The Econometrics of Maritime Safety
Recommendations to enhance safety at sea

Table of Content
List of Abbreviations……………………………………………………………………………………………….i
Acknowledgements………………………………………………………………………………………………. iii
Abstract …………………………………………………………………………………………………………………..1
Chapter 1: Research Question and Methodology………………………………………………….3
1.1. Research Questions…………………………………………………………………………………………..3
1.2. Methodology …………………………………………………………………………………………………….4
1.3. Overview of Datasets and Variables Used ………………………………………………………….6
PART I …………………………………………………………………………………………………………………….9
Chapter 2: Analysis of the Present Safety Regimes…………………………………………….11
2.1. The Complexity of the System …………………………………………………………………………11
2.1.1. The Players of the Regime …………………………………………………………………………11
2.1.2. The Relevant Legal Instruments across the Regimes…………………………………..15
2.1.3. Targeting Systems and Deficiency Coding ………………………………………………….15
2.1.4. Inspection Systems and Remedy for Non Compliance ………………………………….16
2.1.5. The Definition of a Substandard Vessel………………………………………………………17
2.1.6. The Importance of Ship Types and Trade Flows………………………………………….17
2.2. The Total Exposure to Inspections …………………………………………………………………..23
2.2.1. Overview of Inspections in the Name of Safety……………………………………………23
2.2.2. Mandatory Inspections/Surveys/Audits ………………………………………………………23
2.2.3. Non Mandatory Inspections……………………………………………………………………….25
2.2.4. Comparison of Inspection Areas…………………………………………………………………26
2.3. Summary of Costs of Inspections and Insurance Claims ……………………………………34
2.4. Remedies to Enhance Compliance and New Developments………………………………..38
2.4.1. Industry Based Incentives …………………………………………………………………………38
2.4.2. Current Regulatory Based Incentives…………………………………………………………39
2.4.3 New Developments in the EU …………………………………………………………………….39
2.4.4. The New ILO Consolidated Maritime Labor Convention (2006)……………………41
2.4.5. The Voluntary IMO Member Audit Scheme………………………………………………..43
2.4.6. Newest Developments in the Area of PSC at IMO……………………………………….44
2.5. Summary of the Safety Regimes and Inspection Regimes ………………………………….45
Chapter 3: Datasets and Variable Preparation …………………………………………………..47
3.1. Port State Control Dataset and Casualty Datasets……………………………………………47
3.2. Variable Transformations and Definitions………………………………………………………..49
3.2.1. Basic Port State and Casualty Variables…………………………………………………….49
3.2.2. List of Definitions Used in Casualty Analysis……………………………………………..51
3.2.3. The Selection of PSC Relevant Casualties…………………………………………………..52
3.3. The Selection of Ship Types …………………………………………………………………………….52
PART II………………………………………………………………………………………………………………….55
Chapter 4: The Global View on Port State Control……………………………………………..57
4.1. Key Descriptive Statistics for PSC …………………………………………………………………..57
4.1. Key Figures on Registered Vessels ……………………………………………………………….57
4.2. Key Figures on Combined PSC Inspections……………………………………………………58
4.2. The Probability of Detention ……………………………………………………………………………66
4.2.1. Description of Model and Methodology……………………………………………………….66
3.3.2. Step 1 Results: Per MoU and Ship Type……………………………………………………..70
3.3.3. Step 2 Results: Coefficient Testing (Performed in 2 Rounds)………………………..71
3.3.4. Step 3 Results: Final Models per Ship Type………………………………………………..71
3.3.5. Step 4: Visualization of Results………………………………………………………………….77
3.3.6 Individual Results per Ship Type………………………………………………………………..78
3.3.7. Results for the Caribbean MoU………………………………………………………………….82
3.3.8. Differences in Deficiencies across the MoU’s……………………………………………….82
3.3.9. Differences in Port States ………………………………………………………………………….87
3.3.10. Average Probabilities based on Inspector’s Background …………………………….88
3.3.11. Overall View Based on Average Probabilities ……………………………………………90
3.4. Summary of Major Findings: Port State Control……………………………………………….91
PART III ………………………………………………………………………………………………………………..93
Chapter 5: Key Statistics on Casualties and Overview of Datasets……………………95
5.1. Selection of Port State Control Relevant Casualties ………………………………………….95
5.2. Key Descriptive Statistics for Casualties ………………………………………………………….96
5.3. Overview of Inspections and Casualties……………………………………………………………99
5.4. Overview of Dataset Combinations for Regressions …………………………………………103
Chapter 6: Probability of Casualty – Overall View……………………………………………105
6.1. Preparation of Datasets and Sample Sizes ……………………………………………………..105
6.1.1. The Selection of Relevant Casualties………………………………………………………..105
6.1.2. The Selection of Relevant Datasets…………………………………………………………..106
6.2. Selection of Variables and Model Explanation…………………………………………………107
6.3. Model Assessment and Final Results ………………………………………………………………110
6.4. Visualization of Results: Casualty Normal Models ………………………………………….111
6.4.1. Overall View of Inspected versus Non-Inspected Vessels……………………………112
6.4.2. The Partial Effects of Inspections on Casualties………………………………………..117
6.4.3. Assessment of the Target Factor: Can targeting be improved? …………………….121
6.5. Summary of Findings: Casualty Overall View…………………………………………………131
Chapter 7: Probability of Casualty-Refined View……………………………………………..135
7.1. Description of Methodology for Data Preparation ……………………………………………135
7.1.1. The Selection of Relevant Datasets to be Used ………………………………………….135
7.1.2. Methodology Used to Match Ships ……………………………………………………………136
7.2. Explanation of Variables and Base Model Used ………………………………………………141
7.2.1. Variables Used for the Regression ……………………………………………………………141
7.3. Model Assessment and Final Results ………………………………………………………………144
7.3.1. Type I & II Model: Casualty Refined View ………………………………………………..144
7.3.2. Type III Model: Casualty First Events and Deficiencies …………………………….146
7.4. Interpretation and Visualization of Twin Models…………………………………………….147
7.4.1. Refined View on PSC Inspections: Summary of Partial Effects …………………..147
7.4.2. The Effect of Time in-between Inspections………………………………………………..151
7.4.3. The Effect of Inspections and Detentions ………………………………………………….151
7.4.4. PSC Deficiencies and the Probability of Casualty………………………………………158
7.5. Summary of Major Findings: Refined View …………………………………………………….166
PART IV……………………………………………………………………………………………………………….169
Chapter 8: Conclusions and Recommendations ………………………………………………..171
8.1. General Overview of the Safety Regime………………………………………………………….171
8.2. The Overall Magnitude for Improvement Possibilities …………………………………….173
8.3. Effect of Inspections and Improvement for Targeting ………………………………………176
8.4. Identified Areas for Improvement of Inspections……………………………………………..178
8.5. Suggestions for Further Research ………………………………………………………………….180
Summary in Dutch (Nederlandse Samenvatting) ……………………………………………..183
Summary in Spanish (Resumen en Español) …………………………………………………….187
References …………………………………………………………………………………………………………..191
Biography ……………………………………………………………………………………………………………203
Appendices ………………………………………………………………………………………………………….205
Appendix 1: List of Member States of each MoU……………………………………………………….205
Appendix 2: List of Detainable Deficiencies ……………………………………………………………..206
Appendix 3: IMO Definitions of Selected Major Ship Types……………………………………….208
Appendix 4: Variable List and Respective Coding for Regressions ……………………………..209
Appendix 5: Grouping of Countries of Ownership……………………………………………………..218
Appendix 6: Results of Correspondence Analysis for Ship Type Selection …………………..219
Appendix 7: Step 3: Final Models: General Cargo……………………………………………………..221
Appendix 8: Step 3: Final Models: Dry Bulk……………………………………………………………..225
Appendix 9: Step 3: Final Models: Tanker………………………………………………………………..229
Appendix 10: Step 3: Final Models: Container ………………………………………………………….233
Appendix 11: Step 1: Results of Regressions: Passenger Vessels ………………………………..237
Appendix 12: Step 1: Results of Regressions: Other Ship Types …………………………………240
Appendix 13: Step 1: Results of Regressions: Caribbean MoU……………………………………243
Appendix 14: Step 3: Probability of Detention: General Cargo …………………………………..244
Appendix 15: Step 3: Probability of Detention: Dry Bulk …………………………………………..246
Appendix 16: Step 3: Probability of Detention: Tanker ……………………………………………..248
Appendix 17: Step 3: Probability of Detention: Container………………………………………….250
Appendix 18: LM Test for Very Serious Casualties……………………………………………………252
Appendix 19: Estimation Results of Greene Model, Very Serious Casualties ………………254
Appendix 20: Average Probabilities of Corrected versus Uncorrected Model……………….257
Appendix 21: Results of Regression: Very Serious Casualties…………………………………….260
Appendix 22: Results of Regression: Serious Casualties…………………………………………….263
Appendix 23: Results of Regressions: Less Serious Casualties …………………………………..267
Appendix 24: Results of Regression: Fishing Fleet (above 400gt) ……………………………….272
Appendix 25: Probability of Casualty per DoC Country of Residence………………………….275
Appendix 26: LM-Test Type I (Very Serious) and Type II Models ………………………………276
Appendix 27: Matching Models: Type I Models…………………………………………………………281
Appendix 28: Matching Models: Type II Model …………………………………………………………289
Appendix 29: LM Test: Type III Models……………………………………………………………………292
Appendix 30: Matching Models: Type III Models ………………………………………………………299
List of Figures
Figure 1: Description of Research Areas and Questions ………………………………………………..3
Figure 2: Overview of Research Methods……………………………………………………………………..4
Figure 3: Overview of Datasets Used…………………………………………………………………………..7
Figure 4: Overview of Variables Used………………………………………………………………………….8
Figure 5: Players of the Safety Regime in General ……………………………………………………..11
Figure 6: The Selection of Ship Types………………………………………………………………………..18
Figure 7: Ship Types Inspected (Before and After Regrouping) ……………………………………20
Figure 8: Overview of Trade Flows (Port Calls per Ship Type, 2004) ……………………………21
Figure 9: Major Trade Flows (Average Million Ton-Miles for 2000 to 2004)………………….22
Figure 10: Summary of Total Inspection and Audit Exposure ……………………………………..23
Figure 11: Inspection and Detention Frequency of Vessels (1999 to 2004) ……………………33
Figure 12: Inspection Costs versus Insurance Claims in % to Total ……………………………..36
Figure 13: Average Claims of Inspected versus Non-Inspected Vessels ………………………..37
Figure 14: Average Claims of Inspected versus Non-Inspected Vessels per Ship Type…..37
Figure 15: Mean Amount of Deficiencies on Greenaward Certified Ships……………………..38
Figure 16: Auditing Process………………………………………………………………………………………44
Figure 17: Port State Control Dataset Preparation …………………………………………………….47
Figure 18: Casualty Dataset Preparation …………………………………………………………………..48
Figure 19: Ship Types and Deficiencies ……………………………………………………………………..53
Figure 20: Composition of World Fleet (by number of vessels)……………………………………..57
Figure 21: Ships Eligible for Inspection……………………………………………………………………..58
Figure 22: Split up of Ship Sizes (gt) – World Fleet……………………………………………………..58
Figure 23: Key Figures – Total PSC Dataset ………………………………………………………………59
Figure 24: Amount of Deficiencies per Inspection per MoU …………………………………………60
Figure 25: Ship Types Inspected per MoU………………………………………………………………….60
Figure 26: Ship Types and Detention per MoU …………………………………………………………..61
Figure 27: Detention and Mean Deficiencies of Classification Societies………………………..62
Figure 28: Percentage of Detention per Flag State and MoU……………………………………….63
Figure 29: Mean Deficiencies per Flag State and MoU………………………………………………..64
Figure 30: Ownership of Inspected Vessels ………………………………………………………………..64
Figure 31: Visualization of Methodology for data preparation……………………………………..67
Figure 32: Probability of Detention – General Cargo …………………………………………………..78
Figure 33: Probability of Detention – Dry Bulk…………………………………………………………..79
Figure 34: Probability of Detention – Tankers ……………………………………………………………79
Figure 35: Probability of Detention – Container …………………………………………………………80
Figure 36: Probability of Detention – Passenger Vessels……………………………………………..81
Figure 37: Probability of Detention – Other Ship Types………………………………………………81
Figure 38: Probability of Detention – Caribbean MoU ………………………………………………..82
Figure 39: Contribution Weight towards Detention: All Ship Types …………………………….83
Figure 40: Contribution Weight towards Detention: General Cargo …………………………….84
Figure 41: Contribution Weight towards Detention: Dry Bulk …………………………………….85
Figure 42: Contribution Weight towards Detention: Tanker………………………………………..85
Figure 43: Contribution Weight towards Detention: Container ……………………………………86
Figure 44: Contribution Weight towards Detention: Passenger and Other Ship Types ….86
Figure 45: Probability of Detention and Selected Port States ………………………………………88
Figure 46: Average Probability of Detention per Inspector’s Background……………………..89
Figure 47: Average Probability of Detention per Inspector’s Background……………………..89
Figure 48: Probability of Detention per Ship Type (> 15 deficiencies, 5,212 ships) ………..90
Figure 49: Probability of Detention per Ship Type (No deficiencies, 98,953 ships)…………91
Figure 50: Seriousness of Casualties (1999 to 2004) ……………………………………………………96
Figure 51: Casualty First Events per Ship Type (1999 to 2004) …………………………………..97
Figure 52: Ship Types and Casualties (1993 to 2004) ………………………………………………….97
Figure 53: Seriousness of Casualty per Region (1999 to 2004)……………………………………..98
Figure 54: Pollution Type per Region (1999 to 2004) …………………………………………………..99
Figure 55: The Overall View on Inspections and Casualties (1999 to 2004) ………………..100
Figure 56: Ships Inspected in Relation to Ships with Casualties (1999-2004) ……………..101
Figure 57: Ships Detention and Seriousness of Casualty (1999-2004)…………………………101
Figure 58: Mean Amount of Deficiencies per Flag State: 6 months prior to casualty……102
Figure 59: Mean Amount of Deficiencies per Class: 6 months prior to casualty …………..102
Figure 60: Mean Amount of Deficiencies per Seriousness of Casualty ………………………..103
Figure 61: Description of Dataset Combinations for Casualty Regressions …………………103
Figure 62: Description of Methodology Used …………………………………………………………….105
Figure 63: Visualization of Variable Structure: Normal Models …………………………………109
Figure 64: Average Probability of Casualty………………………………………………………………112
Figure 65: Probability of Casualty (Inspected versus Non-Inspected Ships)………………..113
Figure 66: Commercial Fleet versus Fishing Fleet-Flag State Grouping …………………….114
Figure 67: Average Probability of Casualty per Ship Type…………………………………………115
Figure 68: Probability of Very Serious Casualty per Ship Type………………………………….115
Figure 69: Probability of Serious Casualty per Ship Type………………………………………….116
Figure 70: Probability of Less Serious Casualty per Ship Type ………………………………….117
Figure 71: Average Effect of Inspection across Regimes on Very Serious Casualties……119
Figure 72: Effect of Age on the Probability of Casualty ……………………………………………..120
Figure 73: Improvement Areas for PSC eligible ships (1999-2004)……………………………..122
Figure 74: Probability of Very Serious Casualty per Flag State Group……………………….123
Figure 75: Probability of Serious Casualty per Flag State Group……………………………….123
Figure 76: List of Top 30 Flag States: Very Serious Casualty…………………………………….124
Figure 77: List of Fishing Fleet (>400gt, more than 50 ships) ……………………………………125
Figure 78: Probability of Casualty per Classification Society……………………………………..126
Figure 79: Probability of Casualty and Class Groups ………………………………………………..127
Figure 80: Probability of Very Serious Casualty per Ownership Group………………………128
Figure 81: Probability of Serious Casualty per Ownership Group………………………………128
Figure 82: Probability of Casualty of Detained versus not Detained Vessels……………….129
Figure 83: Probability of Casualty: Greenaward Certified (oil tankers) ………………………129
Figure 84: Probability of Casualty: Rightship Inspected (bulk carriers and oil tankers) 130
Figure 85: Probability of Casualty and Number of Deficiencies………………………………….130
Figure 86: Description of Methodology Used …………………………………………………………….135
Figure 87: Visualization of Matching Methodology (per Ship) ……………………………………139
Figure 88: Visualization of Variable Structure: Twin Models …………………………………….143
Figure 89: Effect on Time in-between Inspections: Seriousness………………………………….151
Figure 90: Effect on Time in-between Inspections: First Events…………………………………152
Figure 91: Inspection Effect across Regimes: Very Serious ………………………………………..153
Figure 92: Inspection Effect across Regimes: Serious ………………………………………………..154
Figure 93: Inspection Effect across Regimes: Less Serious ………………………………………..154
Figure 94: Average Inspection Effect per Casualty First Event………………………………….156
Figure 95: Probability of Casualty per Frequency of Inspection (6 months prior) ………..157
Figure 96: Probability of Casualty per Frequency of Detention (6 months prior) …………157
Figure 97: Very Serious and Serious Casualties (Negative Effects)…………………………….158
Figure 98: Very Serious and Serious Casualties (Positive Effects) ……………………………..159
Figure 99: Less Serious Casualties and Deficiencies …………………………………………………160
Figure 100: Fire and Explosion and Deficiencies ………………………………………………………161
Figure 101: Engine Related First Events and Deficiencies ………………………………………..162
Figure 102: Deck Related First Events and Deficiencies ……………………………………………163
Figure 103: Wrecked/Stranded/Grounded and Deficiencies ……………………………………….164
Figure 104: Collision and Contact and Deficiencies …………………………………………………..165
Figure 105: The Overall View on Inspections and Casualties (1999 to 2004) ………………173
Figure 106: Improvement Areas for PSC eligible ships (1999-2004)……………………………175
List of Tables
Table 1: Key Figures on Safety Regimes…………………………………………………………………….12
Table 2: List of Legal Instruments, Targeting and Inspection Systems per MoU (part 1) 13
Table 3: Description of Main Deficiency Codes……………………………………………………………16
Table 4: Summary of Harmonized System of Survey and Certification…………………………24
Table 5: Inspection Matrix – Main Areas of Inspection in Comparison………………………….27
Table 6: Total Estimated Port State Control Inspection Costs (USD) …………………………..34
Table 7: Summary of Inspection Frequency, Allocated Time and Costs (USD/year) ………34
Table 8: Average P&I Club Claim Figures per Vessel and Year (2000 to 2004)……………..35
Table 9: Average Inspection Costs versus Insurance Claims in USD (2000 to 2004) ……..36
Table 10: Summary of Variable Groups……………………………………………………………………..49
Table 11: Inspection Data Summary per MoU ……………………………………………………………59
Table 12: Key Figures on Classification Societies – Total Dataset ……………………………….62
Table 13: Key Figures on Flag States – Total Dataset…………………………………………………63
Table 14: Detention and DoC Country of Residence ……………………………………………………65
Table 15: Summary of Datasets per MoU and Ship Type…………………………………………….67
Table 16: Binary Logistic Models: List of Total Variables Used per MoU……………………..68
Table 17: Binary Logistic Models: List of Variables Used per ST – step 2 Models ………….69
Table 18: Step 1: Classification Tables……………………………………………………………………….70
Table 19: Cut Off Rates (based on observed detention rate) per ST and MoU……………….71
Table 20: Variables changed based on QML versus non QML estimation……………………..71
Table 21: Step 2: Results – Testing of Equality of Coefficients across the Regimes………..72
Table 22: Summary of Key Statistics and Classification Table…………………………………….76
Table 23: Grouping of Deficiency Codes for Visualization ……………………………………………77
Table 24: Split up of Ships with casualties versus non-casualties (1999 to 2004)…………106
Table 25: Split up of Ship Types of Sample versus World Fleet (1999 to 2004) ……………106
Table 26: Number of Observations in the End Model ………………………………………………..107
Table 27: List of Variables Used in Casualty Normal Models …………………………………….108
Table 28: LM Test for Tonnage and Age …………………………………………………………………..110
Table 29: Key Statistics of Final Models: Probability of Casualty ………………………………111
Table 30: Summary of Main Variables: Casualty Normal Models ………………………………118
Table 31: Testing of Restrictions – Inspection Variables…………………………………………….119
Table 32: Datasets for Port State Control and Casualty Merges Performed ………………..136
Table 33: List of Variables used to Match Ships ……………………………………………………….136
Table 34: Ship Type Groups for Matching ………………………………………………………………..137
Table 35: Year Built Ranges for Matching………………………………………………………………..137
Table 36: Tonnage Ranges for Matching…………………………………………………………………..137
Table 37: Overview of Groups used for Matching Ships per Dataset…………………………..138
Table 38: Summary of Matches by Degrees for Round 1(by Ship)……………………………….140
Table 39: Summary of Matched Datasets (by Ship) …………………………………………………..140
Table 40: List of Twin Models and their Variables of Interest ……………………………………141
Table 41: Variables Used in the Twin Regressions (Type I, II and III models)…………….142
Table 42: Summary of Matched Dataset by Seriousness of Casualty (by Ship) ……………144
Table 43: Test Statistics for LM-Test: Type I and II models ………………………………………145
Table 44: Summary of Statistics – Type I and II Model……………………………………………..145
Table 45: Summary of Matched Dataset by Casualty First Events (by Ship) ………………146
Table 46: Summary of Statistics – Type III Models (6 month time frame)…………………..146
Table 47: Test Statistics for LM-Test: Type III models………………………………………………147
Table 48: Summary of Main Variables and their Significance: All Twin Models………….148
Table 49: Testing of Restrictions (Wald Test) – Inspection Variables: Type I Models……153
Table 50: Testing of Restrictions (Wald Test) – Inspection Variables: Type III Models …155
List of Equations
Equation 1: Probability of Detention (either per regime or ship type)…………………………..66
Equation 2: Definition of term xiβ of Step 1 Model………………………………………………………68
Equation 3: Definition of term xβ of Step 2 Model ………………………………………………………69
Equation 4: Probability of Casualty Standard Model…………………………………………………107
Equation 5: Definition of term xiβ of Casualty Standard Model………………………………….108
Equation 6: Probability of Casualty allowing for heteroscedasticity……………………………110
Equation 7: Detailed Effect of Inspection on Casualties…………………………………………….143
Equation 8: Definition of term xiβ of Casualty Detailed Model …………………………………..144
i
List of Abbreviations
AIS Automatic Identification System
AMSA Australian Maritime Safety Authority
BC Code of safe practice for solid bulk cargoes
BCH Code for the construction and equipment of ships carrying
dangerous chemicals in bulk
BLU Code of practice for the safe loading and unloading of bulk carriers
CAP Condition Assessment Program
CAS Conditions Assessment Scheme
CBT Clean Ballast Tanks
CCSS Code of Safety for Caribbean Cargo Ships
CDI Chemical Distribution Institute
CL Classification Society
CMOU Caribbean Memorandum of Understanding
COLREG Convention on the International Regulations for Preventing
Collisions at Sea
COW Crude Oil Washing
CSR Continuous Synopsis Record
CSS Code of safe practice for cargo stowage and securing
DoC Document of Compliance
DWT Deadweight
EC European Community
EEBD Emergency Escape Breathing Device
EMSA European Maritime Safety Agency
EPIRB Emergency position indicating radio beacons
ESP Enhanced Survey Program
EU European Union
FL Flag State
FSA Formal Safety Assessment
FSS Fire Safety Systems Code
GCH Old Gas Carrier Code for ships constructed before 1st October 1994
as per resolution MSC 5 (48)
GISIS Global Integrated Shipping Information System
GDP Gross Domestic Product
GMDSS Global Maritime Distress and Safety System
GPS Global Positioning System
GT Gross Tonnage
HS High Speed
HSC International Code of safety for high-speed craft
IACS International Association of Classification Societies
IBC International Code for the Construction and Equipment of ships
carrying dangerous chemicals in bulk (constructed after 1st July
1986)
IGC International Code for the construction and Equipment of Ships
Carrying Liquefied Gas in Bulk
ILO International Labor Organization
IMDG International Maritime Dangerous Goods Code: contains
classification, packing, marking, labeling and stowage of dangerous
goods in packaged form
IMO International Maritime Organization
ii
IMOU Indian Ocean Memorandum of Understanding
INTERCARGO International Association of Dry Cargo Shipowners
INTERTANKO International Association of Tanker Owner
ISM International Safety Management Code
ISPS International Ship and Port Facility Security Code
ITOPF International Tankers Owners Pollution Federation
LI Legal Instrument
LM Lagrage Multiplier
LMIU Lloyd’s Maritime Intelligence Unit
LN Natural logarithm
LSA International Life Saving Appliance Code
MARPOL International Convention for the Prevention of Pollution from Ships
ML Maximum likelihood
MoU Memorandum of Understanding
OBO Combination Carrier
OCIMF Oil Companies International Marine Forum
OECD Organization for Economic Co-operation and Development
OLS Ordinary Least Squares
OPA Oil Pollution Act
OWN Ship’s Owner
PMOU Paris Memorandum of Understanding
P&A Procedure and Arrangement Manual
P&I Protection and Indemnity
PS Port State
PSC Port State Control
QML Quasi-maximum likelihood
RO Recognized Organization
Ro-Ro Roll On – Roll Off
RS Russian Maritime Register of Shipping (Russia)
SART Search and rescue transponders
SBT Segregated Ballast Tanks
SIRE Ship Inspection Report Program
SIReNAC Central information system for port state inspection records
SMPEP Shipboard Marine Pollution Emergency Plan
SOLAS International Convention for the Safety of Life at Sea
SOPEP Ship Oil Pollution Emergency Plan
ST Ship Type
STCW International Convention on Standards of Training, Certification
and Watch keeping for Seafarers
TBT Tributyltin
TONNAGE International Convention on Tonnage Measurement of Ships
UN United Nations
UNCLOS United Nations Convention on the Law of the Sea
UNCTAD United Nations Conference on Trade and Development
UMS Unmanned machinery spaces
US United States of America
USCG United States Coast Guard
VMOU Viña del Mar Agreement on Port State Control
VDR Voyage Data Recorder
VLCC Very large crude carrier
iii
Acknowledgements
Much improvement in the safety of life at sea has been accomplished since the era of the
last sailing vessels beating their way around Cape Horn where lives were lost each
voyage. Today, the legal framework to enhance safety at sea and to protect against
pollution is complex and the seagoing profession has changed over the last ten years
where more control and complex daily operations have eroded some of the seamanship
found on ships that had to face the challenges of the seven seas without shore-side
Helpance.
Despite the fact that there is a lack of trust amongst most players in the industry, there
is considerable concern about safe and environmentally friendly operations from most
ship owners and operators. Nevertheless, the safety system, as complex as it may be, due
to its international structure provides loopholes for substandard ships to trade and to
distort competition and there is still room to improve the daily lives and working
conditions of some seaman.
Having sailed myself for ten years across the seven seas, the shape of my life has been
strongly influenced by my dedication towards the sea despite the fact that I come from a
non maritime nation. I hope that the work presented in this thesis provides a useful
contribution towards finding a solution to a topic that I feel is complicated and very
political in nature as too much emphasis is placed on national interest. None of this work
would have been possible without the help and cooperation of many people who I would
like to acknowledge shortly here and I hope that I have not forgotten anybody.
First of all, I would like to express my appreciation to my promoter and mentor Prof.dr.
Philip Hans Franses who has guided me through this process and whose support and
patience was crucial for the successful completion of the thesis. It was certainly a
privilege for me to have Prof. Franses as my promoter for this thesis. Next, I would like to
thank my friend and ex-colleagues Ratan Sing Rathore who has helped me with some of
the computer and software issues concerning my large dataset. I would also like to thank
the members of the inner committee for their feedback which helped me to improve my
final document.
Very important for this project was the cooperation of some of the port state control
regimes or maritime administrations in order to be able to use their data and I would like
to express my gratitude to them and hope that some of the interim and final results are
useful. Unfortunately, one of the major regimes opted not to participate which is the
Tokyo Memorandum of Understanding (MoU).
In this respect, I would like to thank the other regimes who did and who also showed
openness towards independent research in order to improve transparency. From the
secretariat of the Paris MoU in The Hague, I would like to thank all members but in
particular Alexander Sindram and Richard Schiferli. From the Secretariat of the Viña del
Mar Agreement in Buenos Aires, I would like to thank Roberto Annichini for his support
and patience to answer all my questions. I especially enjoyed meeting Louis Alberto
Zecchin who was the former head of port state control in Argentina and was very helpful
in helping me understand the Viña del Mar Agreement on PSC. From the USCG in
Washington DC, I would like to especially thank E.J. Terminella for his kindness to
arrange for me to visit the New York Office on Staten Island and to go along on a couple
of inspections in the US. Also his support in extracting US data for this study out of the
iv
whole US database was very helpful and took considerable amount of time on his side in
order to bring the data into a compatible format.
From the Australian Maritime Safety Authority in Canberra and as my first liaison to the
secretariat of the Indian Ocean Memorandum of Understanding, I would like to thank
Chris Barnes, Michael Kinley and Brad Groves. From the secretariat of the Indian Ocean
MoU, I would further like to acknowledge Bimalesh Ganguli. I especially enjoyed meeting
the Australian delegates at IMO who always found some time to meet up with me despite
their very busy schedule. Finally, I would like to thank the secretariat of the Caribbean
Memorandum of Understanding and in particular Capt. Dwain Hutchinson from the
Bahamas Maritime Authority and Dwight C.G. Gardiner from Antigua & Barbuda as
well as Ltd. Cdr. Bennett from the Secretariat of the Caribbean MoU in Jamaica. It was
not easy for this regime to supply me with data since its database is relatively new but
nevertheless, an effort was made to participate in the project. Since I lived myself in the
Caribbean for two years of my life and have sailed there for five years, I know the region
quiet well and I appreciated this effort.
The data providers in this industry have also been very considerate towards my low
research budget and I would like to thank Alex Grey and Trevor Downing from Lloyd’s
Register Fairplay for providing me with casualty data and for conducting a couple of
custom made queries to obtain more detailed information on the history of ships. In
addition, it was always very nice to meet up with both and to discuss some of my findings
over a glass of wine and dinner in London. The second industry provider of data was
LMIU2 and there I would like to acknowledge Lynn Browning.
During the course of the project, I was allowed to participate as observer at several IMO
proceedings which I always enjoyed very much. It was a privilege for me to spend some
time at IMO as an intern and to meet the delegates and discuss some of my findings. In
order to facilitate all this, I would like to thank Brice Martin-Castex, Jo Espinoza and
Rouba Ruthnum. My discussions with Brice Martin-Castex on the subject in general were
always very interesting and helpful and I appreciate the time he has given to me on each
of these trips. I would also like to mention and thank my friend Regina Figl, Mengh
Sheng and Heinz Peter for their hospitality during my trips to London and that I was
allowed to stay at their homes.
The next group of people I would like to acknowledge here are all the inspectors, ship
owners, ship operators, classification societies and representatives of the vetting
inspection regimes. With this respect, I would like to thank my friend Yuri Sakurada
(DNV) for her time on discussions and for giving me her advice as a naval architect.
During the course of my time in Rotterdam, Yuri became a real friend with whom I will
certainly stay in touch regardless of our locations in this world. I would also like to thank
Rob Pijper (Lloyd’s Register) for taking me along on several surveys including an ISM
audit. I enjoyed going along on his inspections very much – in particular his very positive
attitude to deal with problems as they arise and his insight into some of the technical
issues. Further I would like to mention Pieter Andringa (Germanischer Lloyd).
From the vetting inspection regimes, the most helpful persons for this project were Henk
Engelsman (CDI/OCIMF) and Capt. Warwick Norman (Rightship). I was allowed to
observe vetting inspections on chemical tankers, oil tankers and dry bulk carriers. In
particular Henk was always very helpful and provided me with information and his
extensive knowledge about shipping derived from his experience as being a former
2 Lloyd’s Maritime Intelligence Unit
v
captain and inspector for many years. Finally, I would like to acknowledge the
Greenaward Foundation, in particular Capt. Jan Fransen for supplying me with data
which I incorporated into my overall dataset. I would like to extend special thanks to
Aarnout Salwegter from the Dutch Shipping Inspectorate for taking me along on several
port state control inspections in Rotterdam and for sharing his opinion about inspections
in general. The same applies to the Belgium PSC office and in particular to Walter de
Graeve and J.P. van Byten for taking me along on several inspections in Antwerpen of
which I could also witness the process of a detention. I would also like to acknowledge
Walter Vervlosem who has taken me on a P&I Club inspection in Ghent and who has
brought the maritime insurance world a bit closer to me.
In particular I would like to thank all the seaman and officers onboard the vessels I
visited including their designated persons ashore and superintendents. The cooperation of
the ship owners or operators was in general very good. It is always interesting to visit a
vessel since one is treated well by the people onboard. For me, each of these visits
reminded me at my own time at sea but also showed me the reality of shipboard life and
operations in the commercial shipping industry which is quite different from the cruise
industry – the industry I came from. It also gave me insight into the complex daily
operations that are on some of the vessel. In particularly, I enjoyed the discussions and
interesting stories from various captains and chief engineers connected with these visits
and their genuine hospitality. I would like to thank Capt. George Ntallaris for his
explanation on iron ore cargo operations, Capt. John Dudley for his insight into tanker
operations, Dick Pas for his time and discussions about IMO and Dennis de Witte for his
support in organizing additional ship visits.
Finally, I would like to thank the last group of people from the Econometric Institute at
Erasmus University for their time and patience to help me improve my econometric
skills. I would like to mention Michel van de Velden and thank him for letting me use a
program of his for the correspondence analysis, Govert Bijwaard, Michiel de Pooter and
Richard Paap – for their time and discussion about the logit model, maximum likelihood
estimation and heteroscedasticity. I would also like to thank Michiel, Francesco and Chen
for their friendship which helped me fight the loneliness in writing this thesis. Further, I
would like to thank Christiaan Heij and Jan Brinkhuis for allowing me to follow one of
their courses and for their moral support. I particular enjoyed my lunch sessions with Jan
and the encouraging words of both – Christiaan and Jan during the last phase of my PhD.
I will certainly miss that and hope to be able to work on future publications in this area
with some of the members of the Econometric Institute. Also special thanks to Jan and
Chantal Cheung-Tam-He for helping me in producing a Dutch translation of the abstract.
Many thanks are also given to my friend and long-time ship mate, Wayne Bornio who has
taken the time to perform some of the proof-reading of this thesis. But last not least, I
would like to thank my friend Gert-Jan Huisink whose discussions and friendship has
helped me throughout my time in Rotterdam. Gert-Jan always had a supportive word for
me and encouraged me in the last and most difficult phase of this thesis. The same
applies to Jose Lucas who always had some good words of support for me. I would like to
thank my parents and family for their understanding and their help in adapting to shore
based life during the last three years.
Overall, I rate the experience in conducting my research and in writing up this thesis as a
positive accomplishment. The only regret I do have is that I did not have enough time to
look into all possible aspects of my vast dataset but in the final stage of the PhD, I
accepted that some areas can be left open for further research.
vi
1
Abstract
This thesis should be seen and understood as a first attempt to study port state control on
a global scale by measuring the effect of inspections on the probability of casualties and
by identifying areas for improvement. What is new in this thesis is the combination of
port state control data from various regimes and casualty data from three different
sources of the same time frame. The corresponding research questions are 1) What is the
present state of the safety regime? 2) Can targeting of ships for inspections be improved?
3) What is the effect of inspections on casualties? and 4) How can inspections be
improved? It is based on approx. 183,000 port state control (PSC) inspections and 11,000
casualties from 1999 to 2004. The thesis will hopefully open up a new chapter in research
in the area of maritime safety of which the future potential, when political barriers are
overcome and more transparency is accepted, should not be ignored by the industry and
regulators. Several econometric techniques are used to produce probabilities of detention
and casualties either per seriousness or casualty first events. The author does not take
any political dimensions of flag state implementation or port state control into
consideration but solely concentrates on the technical aspects of the topic in question.
The maritime industry is characterized by a heavy legal framework based on
international law with limited legal enforcement possibilities in case of non compliance.
This creates loopholes and distortion to competition due to the existence of a market of
substandard ships. From a public perspective, the desired situation is to promote safe,
secure and environmentally friendly maritime transportation and to decrease the number
of substandard vessels. Flag states are to be seen as the first line of defense in
eliminating sub-standard vessels followed by the second line of defense, the port states.
The lack of trust in the industry has created a playground for inspections on certain ship
types. A considerable amount of industry driven inspections are performed where total
inspections are estimated to be at 11 inspections per year for tankers, 6 for dry bulk
carriers and 5 for all other ship types.
Two areas of potential improvement have been identified: 1) the targeting of ships for
inspections and 2) the inspections and follow up on deficiencies themselves. At first sight,
it seems that too many ships with zero deficiencies are targeted but when aggregated by
ship and across regimes, the percentage of ships with zero deficiencies diminishes
significantly over the given time period. Nevertheless, a certain group of vessels (about
7% of ships eligible for port state control) has been identified to be over-inspected and
efforts should be made to shift inspection efforts towards the groups of ships that can
benefit from an inspection which is estimated to be at about 14% of all port state control
eligible ships based on the time period used for this analysis. The effect of port state
control inspections towards the probability of casualty can be measured for very serious
casualties but not for serious and less serious casualties. Depending on the overall risk
profile of a vessel, an inspection can potentially decrease the probability of having a very
serious casualty by approx. 5% per inspection where the effect can be as strong as 10% for
very high risk vessels.
While key figures on deficiencies and detentions vary accordingly across the regimes, the
difference towards the probability of detention is merely reflected by the differences in
port states and the treatment of deficiencies and not necessarily by age, size, flag, class or
owner. The target factor can be improved by developing a target factor on combined
inspection data taking the total ship’s inspection history into account. Furthermore
regional differences can be allowed since high risk areas were identified for West Africa,
2
the Indian Ocean region, the North Atlantic East region and the South China Sea. The
difference in the effect of inspections on the probability of casualty for either seriousness
or casualty first events further confirms a shift of sub-standard vessels from areas such
as the Paris MoU and the USCG to other areas of the globe such as the South American
Region, the Indian Ocean Region or Australia. Inspections of these regimes have shown to
decrease the probability of having a casualty.
The classic variables such as ship type, age, size, flag, the classification society,
deficiencies found in prior inspections and detention are all valid variables for targeting
sub-standard ships. Flag is only one variable out of many variables that can be used to
target sub-standard ships. Age only remains significant for very serious casualties and as
the age of the vessel increases, the probability of having a very serious casualty increases
by about 12% over a 35 year period which translates into about 0.35% per year. The
probability of casualty further confirms that general cargo vessels are ships with the
highest probability of a casualty which is confirmed by the probability of detention. Black
listed flag states or non inspected ships show a higher probability of a very serious
casualty compared to grey and white listed flag states while the same does not hold for
serious and less serious casualties. Average insurance claim costs on the other hand
reveal that highest claim costs are associated with tankers and passenger vessels.
Further improvements on targeting sub-standard ships can be made by adding the
variable indicating the ownership or DoC3 Company of a vessel and certain data on ship
history such as change of class, class withdrawal and change of ownership over time or
where the ship was primarily built, all of which have either shown a positive or negative
effect on the probability of casualty. Another possibility would be to include if a vessel
had been inspected by one of the vetting inspection4 regimes (for dry bulk) or certified by
the Greenaward Foundation.
A refined view on the effectiveness of inspections reveals that detention does not seem to
be significant towards the probability of having a casualty and which is a surprising
result. It does not necessarily mean that detention is not relevant but that the effect
might be captured by the inspection. Furthermore, the time span in-between inspections
is not significant for very serious casualties but is for less serious and serious casualties.
On average and regardless of the seriousness of casualty, the probability increases by
2.3% within the time frame of one year. Basic ship risk profile across all regimes lies
between 0.5% to 1.5% for most ship types and regimes while the average probability of
casualty aggregated for all ship types has been identified to be 0.06% for very serious,
1.6% for serious and 1% for less serious casualties.
The probability of detention models revealed the highest contribution of deficiencies
towards detention in the areas of certificates, ship and cargo operations, the ISM5 code
and safety & fire appliances while lowest contribution is found for machinery and
equipment. The probability of casualty either per seriousness of casualty or casualty first
event also revealed three areas of interest – the ISM code, ship and cargo operations and
machinery and equipment. Those are the main areas which have been identified where
room for improvement exists for port state control inspections in order to decrease the
probability of a casualty.
3 Document of Compliance Company, the designated company responsible for the safety
management onboard a vessel according to the International Safety Management Code (ISM)
4 vetting inspections are performed by the industry (the oil majors or the chemical industry)
5 International Safety Management Code
3
Chapter 1: Research Question and Methodology
1.1. Research Questions
The maritime industry is characterized by a heavy legal framework based on
international law with limited legal enforcement possibilities due to the absence of a
court of justice who can initiate legal proceedings in case of non compliance. This creates
loopholes and distortion to competition due to the existence of a market of substandard
ships. From a public perspective, the desired situation is to promote safe, secure and
environmentally friendly maritime transportation and to decrease the amount of
substandard vessels. The objective of this thesis is to make some recommendations to
contribute towards this objective. Figure 1 lists the main research areas and questions
which are divided into four parts.
Figure 1: Description of Research Areas and Questions
Part 4: Conclusion
Chapter 8: Main Findings
and Recommendations
Chapter 5:
o What is the probability of detention of each regime by
taking into account the different data sources?
Caribbean
MoU*)
Paris
MoU*)
Probability
of
Detention
Indian Ocean
MoU*)
(w/o Australia)
US Coast
Guard
AMSA
Australian
Maritime SA
Viña del
Mar
Agreement
Part 2: The Global View on Port State Control
Chapter 6 and 7:
o Are the right ships targeted for inspection?
o What is the effect of inspections on casualties?
o How can inspections be improved?
Probability
of
Casualty
Part 3: The Link between Port State Control and Casualties
*) MoU = Memorandum of Understanding
Part 1: Analysis of the Safety Regimes
Chapter 2:
o Introduction to the safety regimes
o Overview of respective legal bases
o Analysis and comparison of all safety inspections
Qualitative
Analysis
4
Chapter 2 gives an overall introduction to the safety regimes, the inspection regimes and
their associated costs and frequencies. It further provides an overview of the legal bases
when applicable. Chapter 3 explains the datasets and the processes for variable
transformations which are used in chapter 4, 5, 6 and 7 of this thesis. It further provides
definitions of variables used in this thesis.
Chapter 4 gives an in-depth analysis of some of the Port State Control Regimes today and
is based on 183,819 port state control inspections from various regimes around the world.
It produces the probabilities of a ship being detained across the regimes split into ship
types and will take into account the fact that the datasets come from different data
sources.
Chapter 5 provides an overview of casualty key statistics as well as an explanation on the
types of regressions used for the casualty analysis. Chapter 6 makes the link between
port state control and casualties by looking at the inspection and casualty history of the
ships from the datasets. It measures the effect of inspections on casualties and tries to
evaluate the target factor. Chapter 7 provides a more detailed view and tries to identify
how inspections can be improved. The main research questions can be summarized as
follows:
1. What is the present state of the safety regime?
2. Can targeting of ships for inspection be improved?
3. What is the effect of inspections on casualties?
4. How can inspections be improved?
Chapter 8 provides an overall conclusion to the findings drawing from all the other areas
and gives a recommendation on how to improve the system.
1.2. Methodology
To analyze the safety and inspection regimes from all possible angles and to obtain the
best overview as well as to establish the connection to the daily shipboard life, the
following research methods for this thesis were used and are shown in Figure 2 below.
Figure 2: Overview of Research Methods
Literature
Review
Inspections
and Surveys
Quantitative Analysis
Qualitative
Analysis
Interviews
Descriptive Statistics
Regression Analysis
Interpretation
&
Conclusion
Assessment of
the system
Review of
Legal Base
5
First, the author conducted a qualitative analysis by reviewing the respective legal bases
in order to gain a better understanding of the complexity of the system. The legal base is
split into international law such as the international conventions and the memoranda of
understanding and applicable national law for each of the port state control regimes.
National laws in particular apply to the US, Australia and the EU where EU law (either
directives or regulations) is supreme to national law.
For the review and understanding of the various inspection regimes, the author joined
inspections and surveys as observer and conducted interviews with the inspectors, ship
owner’s associations, port state control officials, ministry officials’, classification societies
and insurance companies. The different types of inspections and surveys were chosen in
order to cover the mandatory (statutory) inspections performed by flag states,
classification societies and port states and the non mandatory inspections performed by
the industry (vetting inspections). A variety of inspections (27 inspections and 1 visit)
were therefore observed on various ship types (general cargo, dry bulk, oil tanker,
chemical tanker, container vessels) as follows6:
1. Flag state inspections (2)
2. Port state control: initial/more detailed (5 including one detention) and
expanded inspection (1), security inspection (2)
3. Classification Societies: annual (4) & renewal (1) inspection, ISM audit (1),
detention follow up (1)
4. Vetting inspections: CDI7 for chemical tankers (2), OCIMF8 (4) for oil tankers
5. Vetting inspection for Rightship for dry bulk carriers (1)
6. Insurance Companies: Protection & Indemnity (P&I) Club (1)
7. Other Inspections and Visits: MARPOL (2), 1 Visit to a VLCC (1)
The port state control inspections were performed in the Netherlands, Belgium and in the
US (New York) where security inspections are performed separately from safety
inspections. Security has not been taken into consideration for this thesis since the ISPS9
Code only came into force in July 2004. The knowledge and oversight gained by the
inspections and interviews was then used in the decision process of the variables used in
the various analyses as well as to help interpreting the data. In addition, an insight into
the various shipboard operations which vary according to the ship types was gained.
A literature review was performed primarily on relevant topics in econometrics and on
the literature connected to the legal framework as well as literature connected to safety
and casualty analysis. Soma (2004)10 lists the many different levels of maritime safety
which have been addressed by researchers so far and splits the areas into five levels
(accidents, ship standards, organization, management and environment) where each level
contributes towards the overall safety level of a vessel. In addition, Talley (1999 ff)11 in
his work has looked at determinants of crew injuries (1999), oil spillage costs (2001) and
vessel damage costs (2002) using similar econometric techniques presented here.
This thesis does not look into root causes for casualties due to the inadequate quality of
casualty data nor does it cover all the levels described by Soma (2004) but emphasis is
given on the Assessment of the effectiveness of the safety regimes, in particular port state
6 Number in brakes is the amount of inspections
7 CDI – Chemical Distribution Institute
8 OCIMF – Oil Companies International Marine Forum
9 International Ship and Port Facility Security Code
10 Soma, T. (2004), Blue-Chip of Sub-Standard, Tronheim, page 190
11 Talley, W.K – for further reference on his articles, please refer to the references.
6
control by measuring the effect of inspections on the probability of a casualty. It is looking
at the subject from a more regulatory or public perspective which is to enhance safe and
environmentally friendly maritime transportation. The combination of the data used in
this study is new due to the various barriers in obtaining raw data on a ship level. The
data for this section comprises incidents, accidents and casualties and the definitions
thereof are presented in Chapter 3. The quantitative part of the thesis consists mainly of
regression analysis based on various combinations of datasets.
The author would like to point out that with respect to the rule making process at IMO,
the organization has developed in 2001 and 2002 guidelines for Formal Safety Assessment
(FSA)12 which should enhance rule making based on risk assessment and cost benefit
analysis. It is worthwhile noticing that some results of this thesis could possibly be
incorporated into future FSA studies.
1.3. Overview of Datasets and Variables Used
Three datasets have been used for the analysis and their relation can be seen in Figure 3.
Set A consists of the inspection database of 183,819 inspections from various Memoranda
of Understanding (MoU13) listed in Figure 1 for the time period January 1999 to
December 2004 where the time period is not fully covered by all regimes. This total
dataset is a combination of six individual inspection datasets and when aggregated, it
accounts for approx. 26,020 ships14 where the average amount of inspections per vessel is
by 7 per ship or 1.7 inspections per ship per year.15
Set C represents an approximation of the total ships in existence16. Out of these vessels,
ships below 400 gt17 and ship types which are not eligible for port state control inspection
such as fishing vessels, government ships, yachts and ferries (for the Paris MoU) have
been eliminated from this dataset which leaves approx. 43,817 ships (46,75% of the total)
for inspection. Since the amount of inspections from the Paris MoU is the dominating part
of this dataset and ferries are treated separately in the EU, ferries have been excluded
from PSC eligible ships. The total estimated inspection coverage by the regimes in
question of eligible ships is 59.4% between set A and the eligible ships of Set C for the
time period in question (1999-2004).
Besides the port state control inspection dataset, a small industry inspection dataset has
been collected of vetting inspection information18 on oil tankers and dry bulk carriers
from Rightship. In addition, oil tankers which are certified by Greenaward have also been
12 MSC/Circ.1023, MEPC/Circ.392 of 5th April 2002 and MSC 81/18 of 7th February 2006
13 A memorandum of understanding (MOU) is a legal document describing an agreement between
parties but is less formal than a contract.
14 25,836 exact ships plus 184 estimated ships. Since there are 1,288 ships with missing IMO
numbers out of the total port state control dataset and the average number of inspections per ship
lies by 7, the unidentified ships can be aggregated to another 184 inspected ships.
15 Based on an average of 4 inspection years which is the average of the total months per regime to
bring the different years of data to the same level for all regimes. The total time period Jan. 1999
to Dec. 2004 therefore represents a total of full 4 inspection years instead of 6 years.
16 As per data received from Lloyd’s Register Fairplay.
17 As per Marpol 73/78, Annex I, Regulation 4 which identifies the vessels subject to mandatory
surveys (page 51)
18 Rightship Rating Data (48,834 records of which 37,080 are used) and Greenaward Data on
certified ships (244 records)
7
identified. The casualty and industry data is linked to the port state control data by the
IMO number and within the same time frame.
Figure 3: Overview of Datasets Used
Set B is the casualty dataset which consists of 11,701 records for time period 1993 to 2004
and is a combination of data received from Lloyd’s Register Fairplay, LMIU19 and the
IMO (International Maritime Organization). The time period 2000 to 2004 is the most
complete casualty dataset since it draws from all three datasets. Aggregated, this dataset
accounts for approx. 9,598 ships or 10% of the total ships in existence from Set C where
the average amount of casualties per ship is by 1.2. Port State relevant casualties without
the fishing fleet aggregate to 6005 ships for the time period 1999 to 2004 or 13.7% of the
total PSC eligible ships. The elimination of port state control relevant casualties is
explained in detail in Chapter 5. A portion of the fishing fleet (above 400gt) is treated
separately in the casualty analysis.
The sets are used in various ways depending on the kind of analysis which is conducted.
In essence, the combination of these datasets gives insight into the amount of ships that
are inspected/not inspected, detained/not detained and have/do not have a casualty with
their respective combinations. Figure 4 gives an overview of the variables used for all
types of analysis for port state control and casualties where the link between the two
datasets is given by the IMO number and the dates of inspection/casualty respectively.
Depending on the type and method of analysis, either dummy variables for each variable
are used or the data is coded into groups (e.g. flag states can be used individually or
grouped into black, grey or white listed flag states)20. The incorporation of the ownership
of a vessel is not a straight forward task in shipping and requires some careful thinking.
19 Lloyds Maritime Intelligence Unit
20 it follows a ranking performed by the Paris MoU each year where white listed flag states
perform best followed by grey and black listed flag states.
Total
Ships in Service
= 93,719 ships
where 47% are
eligible for
PSC inspection
= 43,817 ships
PSC eligible
(43,817 ships)
Set A
Ships
Inspected
183,819 insp.
= 26,020 ships
Set B 59% of Set C
11,701
Casualties
= 9,598 ships
10 % of Set C
Set C
Industry
Data
8
Two types of variable groups have therefore been used. The first one concerns information
on the Document of Compliance Company (DoC) of a vessel based on information received
from Lloyd’s Register Fairplay and the second one and due to the lack of the completeness
of information on the DoC Company is the addition on the ownership of a company which
represents the “beneficial owner”21. Definitions on both types are provided in Chapter 3
where variables, their groupings and transformations are explained in detail. The
Document of Compliance Company is the company that is responsible for the safety
management of the vessels.
Figure 4: Overview of Variables Used
Note: DoC = Document of Compliance Company, an ISM requirement
This short introduction to the research questions, the methods and datasets used to
conduct the analysis should provide enough evidence that the subject is covered from
various angles and that great care was placed on the selection of the datasets and the
data preparation.
Given the datasets used for the quantitative part, it can be assumed that with almost
60% of coverage of port state control data, a sensible interpretation can be made even
with the lack of data from one of the major safety regimes – the Tokyo MoU where
cooperation for this analysis unfortunately could not be obtained. It is important to
indicate that a more refined result in part III of this thesis could have been obtained if
additional data from the Asian region would have been made available for this thesis. The
following chapter continues with an overall introduction to the safety regime as well as its
current legal bases.
21 Based on Lloyd’s Register Fairplay data of the “World Shipping Encyclopedia CD” and Lloyd’s
“Maritime Database CD” plus custom made queries on ownership and construction detail history.
At time of construction
At time of inspection/casualty
PORT STATE CONTROL CASUALTIES
Construction Information
Vessel Particulars (Age, Size, Ship Type)
Classification Society
Vessel Registration (Flag State)
Beneficial Owner
DoC Company
Date of Inspection
Location of Inspection
(either country or port)
Deficiencies
(main deficiency coding)
Detention
Date of Casualty
Location of Casualty
Casualty First Events
Seriousness
Pollution
Loss of Life, Loss of Vessel
Link:
IMO Number
Industry Data
Rightship Ranking
Greenaward Cert.
9
PART I
This part contains two chapters and provides a general overview of all players in the
safety regime and how they interact as well as the major trade flows in order to provide
the necessary understanding of the industry for the quantitative analysis.
It further analyzes all inspections that are performed onboard ships in the name of safety
and tries to establish their frequency and costs. In addition, insight into average
insurance claim costs (above the deductibles) is given. Applicable legal bases and their
respective new developments in the area of safety and port state control are listed and
analyzed.
Part I ends with an explanation of all the variables and the dataset preparations that
were performed to be used for the quantitative part including the selection of ship types.
10
11
Chapter 2: Analysis of the Present Safety Regimes
This chapter provides an introduction to the safety regimes and the inspection regimes by
explaining the players of the safety regime and gives a short overview of the legal bases
and target factors of each of the port state control regimes. In addition, the various
inspections which are performed on ships and their associated costs are explained in
detail to provide a better understanding of the complexity of the system.
2.1. The Complexity of the System
2.1.1. The Players of the Regime
Figure 5 provides an overview of the players of the safety regime which at first side seems
complex. The legal framework is created by three major international organizations
namely, the UN, ILO, and the IMO22 and country specific legislation23. The classification
societies provide the technical expertise during ship building and technical maintenance
of the vessel. In addition, classification societies can be authorized to perform statutory
responsibilities on behalf of the flag states that have the ultimate responsibility to enforce
their legal base which can be a combination of the international conventions of which the
flag state is signatory or its own legal base while the ship owner has the ultimate
responsibility to comply with the combined legal bases.
Figure 5: Players of the Safety Regime in General
The line between the actual ship owner, operator or technical manager of the vessel is not
completely clear in shipping and therefore complicates enforcement of the legal
instruments. In an effort to gain some insight into the relationships, data from Lloyd’s
Register Fairplay was merged with the total dataset as explained previously. The reason
of the existence of the port state control regime derives from the fact that a certain
percentage of ship owners and flag states use the legal “loophole” created by the
international legal framework and try to save costs by operating below the minimum
safety standards. This can cause accidents and damage to the environment, the cargo and
22 UN: United Nations, IMO: Intern. Maritime Organization, ILO: Intern. Labor Organization
23 This could be for instance the “acquis communautaire” for the EU or OPA 90 for the US or any other
country specific legislation
Delegation
Ultimate
Responsibility
Ultimate responsibility
Flag
States
Charterer
Cargo
Owner
Ship
Operator
Ship
Manager
Port
State Control
Ship
Yards
Ship
Owner
Classification
Societies
Insurance Banks
Companies
Legal
Framework
Vetting
Inspections
12
human lives. According to the OECD the percentage of sub-standard ships in the world
commercial fleet is estimated to be between 10-15%24. The industry solution to this
problem is represented by the vetting inspections which are performed on oil tankers,
chemical tankers and bulk carriers. The vetting inspections create a strong commercial
incentive for the ship owner to comply to the vetting inspection requirements since the
outcome of these inspections will determine if the ship gets cargo or not. The various
types of inspections that are performed on ships including port state control inspections
will be explained in detail later on in this chapter. A possible first step to improve
enforcement is the implementation of the IMO Voluntary Flag State Audit Scheme which
has been adopted in December 2005 and will also be explained later on in this chapter.
Port State control can be seen as a last resource of safety to eliminate substandard ships
from the seas. Worldwide, there are currently ten safety regimes in place to cover most of
the coastal states. Those regimes are as follows:
1. Europe and North Atlantic (Paris MoU)
2. Asia and the Pacific (Tokyo MoU)
3. Latin America (Acuerdo de Viña del Mar)
4. Caribbean (Caribbean MoU)
5. West and Central Africa (Abuja MoU)
6. Black Sea (Black Sea MoU)
7. Mediterranean (Mediterranean MoU)
8. Indian Ocean (Indian Ocean MoU)
9. Arab States of the Gulf (Riyadh MoU)
10. US (US Coast Guard).
The first port state control regime that came into place was the Paris MoU in 198225
followed by the others listed previously and noting the standards listed in IMO Resolution
A.682 (17)26 calling for regional co-operation in ships inspections while Resolution A 787
(19) with its amendment of A 822(21) provides guidelines on the procedures to conduct
port state control inspections. Table 1 gives an overview of the safety regimes used in this
analysis with an indication of the years they exists and the number of members as of
2005 where a detailed list of the member states can be found in Appendix 1 for further
detail.
Table 1: Key Figures on Safety Regimes
Safety Regime MoU Signed
Years in
Existence
(in 2005)
Number of
Members
Paris MoU 1982 23 22
Caribbean MoU 1996 9 22
Viña del Mar 1992 13 13
Indian Ocean MoU 1998 7 18
Tokyo MoU (AMSA) 1993 12 18
USCG Primarily 1994 n/a n/a
Although AMSA is part of the Tokyo MoU and the Indian Ocean MoU, the details of the
Tokyo MoU are only listed as a reference and for the various analyses; AMSA is treated
as individual country like the USCG.
24 Peijs, K. (2003). Ménage a trois. Speech at Mare Forum (November 2003: Amsterdam)
25 Paris Memorandum of Understanding on Port State Control, 19th May 2005
26 IMO Resolution A.682 (17) calls for the regional co-operation in ship inspections
13
Table 2: List of Legal Instruments, Targeting and Inspection Systems per MoU (part 1)
Paris MoU Caribbean MoU Viña del Mar
Memorandum Signed/Take
Effect 1982 1996 1992
Years in Existence as per 2005 23 9 13
Relevant Instruments Convention Protocols Convention Protocols Convention Protocols
1) Load lines 66 & Protocol 88 yes yes yes not
specified yes yes
2) SOLAS 74 & Protocols 78 & 88 yes yes yes yes yes yes
3) MARPOL 73/78 with Protocol
78 & 97 yes yes yes yes yes yes
4) STCW 78 & Amendments of 95 yes n/a yes n/a yes n/a
5) COLREG 72 yes n/a yes n/a yes n/a
6) TONNAGE 69 yes n/a yes n/a yes n/a
7) ILO Conv. Nr. 147, 1976 &
Prot.96 yes yes yes not
specified not specified not
specified
8) The Intern. Conv. on Civil
Liability for Oil Pollution Damage,
1992 Protocol.
n/a yes not specified n/a not specified n/a
9) Treatment for ships flying a flag
not party to the legal instruments no favorable treatment no favorable treatment no favorable treatment
10) Other Legal Instruments EC Directive 21/95 and
106/01
Code of Safety for
Caribbean Cargo Ships
(CCSS Code) for ships
below 500gt
not relevant
Targeting System
1) Amount of ships to be
inspected:
25% of ships entering area
per year
15% of est. number of
foreign merchant ships per
year
at least 15% of foreign ships
per year
2) Variables used for targeting:
Flag, Class, Size, Age, Ship
Type, inspections history of
the area only
Flag (above average
detention), Size (below
500gt), Ship Types (carrying
harmful substances),
Previous Deficiency History
Flag, Class, Size, Age, Ship
Type, inspections history of
the area only
3) Targeted ship types either by
type/age or size:
general cargo, bulk carrier,
gas carrier, chemical tanker,
oil tanker, passenger
general cargo ships,
passenger ships, Ro-Ro, oil
tankers, gas carriers,
chemical tankers
passenger ships, ro-ro
vessels, bulk carriers, oil
tankers, gas carriers,
chemical tankers
4) Time criteria
if entering area for the first
time in the last 12 months or
if not inspected in the last 6
m
ships not inspected within
the last six months by
another member state
avoid inspection if vessel
was inspected within the
preceding 6 months
5) Quantitative Target Factor: > 50 points inspection
mandatory no quantitative system yes but not obligatory
6) Deficiency Coding standardized deficiency
coding coding as per Paris MoU coding as per Paris MoU
7) Incentive System for good
ships: no no no
Inspection Systems and References to IMO Resolution A.466(XII), A.787(19) and A.822(21)
References to IMO Resolutions A.787(19) for Marpol A I&II A.682(17), A.787(19) A.682(17), A.787(19),
A,822(21)
Priority inspections yes yes yes
Initial inspections yes yes yes
More Detailed inspection if clear
grounds yes, clear grounds defined yes, clear grounds defined yes, clear grounds defined
Expanded inspection yes no no
Re-examination/Follow up
inspections not stated not stated not stated
Remedies for no compliance detention detention detention
Source: Compiled by author from MoU’s, IMO Resolutions and interviews
14
Table 2 continued Indian Ocean MoU AMSA*) US Coast Guard
Memorandum Signed/Take Effect 1998 n/a n/a
Years in Existence as per 2005 7 n/a n/a
Relevant Instruments Convention Protocols Convention Protocols Convention Protocols
1) Load lines 66 & Protocol 88 yes not
specified yes yes yes yes
2) SOLAS 74 & Protocols 78 & 88 yes not
specified yes yes yes yes
3) MARPOL 73/78 with Protocol 78 &
97 yes yes yes not Annex
VI yes not Annex
IV and VI
4) STCW 78 & Amendments of 95 yes n/a yes n/a yes n/a
5) COLREG 72 yes n/a yes n/a yes n/a
6) TONNAGE 69 yes n/a yes n/a yes n/a
7) ILO Conv. Nr. 147, 1976 & Prot.96 yes not
specified not ratified not
ratified yes not
ratified
8) The Intern. Conv. on Civil Liability
for Oil Pollution Damage, 1992
Protocol.
not specified not
specified n/a yes not ratified n/a
9) Treatment for ships flying a flag
not party to the legal instruments no favorable treatment no favorable treatment no favorable treatment
10) Other Legal Instruments not relevant
Tokyo MoU and Indian
Ocean MoU, Austr.
Navigation Act 1912 section
190AA and 210
Title 46 US Code (Ch33),
OPA 90, MTSA 02
Targeting System
1) Amount of ships to be inspected: at least 10% of estimated
number of foreign ships
minimum of 50% of eligible
ships defined under point 4) not relevant
2) Variables used for targeting:
ships suspended from class
for safety reasons within
preceding 6 months
target factor based on ship
risk profiles (probability of
detention)
based on management, flag,
class, ship type and vessel
history of the area
3) Targeted ship types either by
type/age or size:
ships carrying dangerous
cargos and failed to report
information to authorities
bulk carriers due to the high
% of bulk carriers arriving in
Australia
oil and chemical tankers,
gas carrier, bulk, passenger
ship, general cargo
4) Time criteria if entering area for the first
time or absent of 12 months
cargo ships every 6 mts,
tankers over 15 yrs and
passenger ships every 3 mts
yes, built into the matrix
5) Quantitative Target Factor: no quantitative system
yes – based on a risk factor
on arrival (probability of
being unseaworthy)
yes, different boarding
priorities according to points
6) Deficiency Coding coding as per Paris MoU coding as per Paris MoU different coding but recoding
made for Equasis
7) Incentive System for good ships: no
no but charges for more
detailed inspections – AD
185/hr
yes, Qualship 21
Inspection Systems and References to IMO Resolution A.466(XII), A.787(19) and A.822(21)
References to IMO Resolutions A.682(19), A.787(19) A.787(19), A.822 (21) A.787(19)
Priority inspections yes yes yes
Initial inspections yes yes yes, annual examination
More Detailed inspection if clear
grounds yes, clear grounds defined yes, clear grounds defined yes, clear grounds defined
Expanded inspection no no
no, more expanded
examinations are more
detailed inspections
Re-examination/Follow up
inspections not stated yes, follow up on
deficiencies yes, follow up inspections
Remedies for no compliance detention
detention, for noncompliance
to Sect. 190AA
of the Navigation Act, fines
can be imposed
detention, for noncompliance
to US law – civil
penalty action
*) AMSA is part of the Indian Ocean MoU and the Tokyo MoU but is treated individually in this
analysis
15
The regimes were compared based on their legal relevant instruments for inspection on
foreign vessels, their targeting system and the inspection systems including the
deficiency coding and detentions and a detailed list of these findings is given in Table 2
for easier comparison. The historical development of the regimes varies which can easily
be seen by the number of years of existence. The USCG started with its inspection
program in a limited version in 196827 for passenger vessels and expanded the program to
all vessels calling US ports with the 1978 Port and Tanker Safety Act. After the accident
of the Exxon Valdez in 1989 and the creation of the Oil Pollution Act (OPA 90), inspection
on foreign vessels were re-enforced and clear guidelines of the requirements to comply
with US law were given. As the world fleet changed in the composition of flags and the
US became primarily a port state, it started to emphasize on developing a targeting
matrix for foreign flagged vessels and implemented the PSC program in 1994.
2.1.2. The Relevant Legal Instruments across the Regimes
The relevant legal instruments are listed in Table 2. In essence, there are some
differences in the legal instruments across the MoU’s but all regimes do not give
favorable treatment to ships flying a flag which is not a party to the international
conventions. The Paris MoU area, the US Coast Guard and AMSA have other legal
instruments besides the international conventions which are added or replace some of the
international conventions. The Caribbean Memorandum refers to the Code of Safety for
Caribbean Cargo ships (CCSS Code) for ships below 500 gt and the USCG also
acknowledges the application of this code for the eligibility of inspections for vessels
trading to U.S. ports (7th district)28 but below 500 gt. For failure to comply to US law, the
US Coast Guard can respond with civil action or criminal proceedings which are a very
strong incentive to comply for ship owners. AMSA based on the Australian Navigation
Act can impose fines. For the EU area, due to the European Court of Justice and its
power to start legal proceedings against its flag and port states, the incentive to comply is
also much stronger than in the other regions and new legislative developments in the EU
area should further enhance enforcement in the future.
2.1.3. Targeting Systems and Deficiency Coding
Some port state control regimes use custom made target factors to decide if a vessel
should be inspected or not since available resources should be allocated to inspect high
risk vessels. Those target factors might reflect the regional needs since port state control
is linked to the trade flows which determine the ship types. All regimes target certain
ships types and use a time criteria and the history of previous inspections as an input
variable. The regimes do not take inspections of each other into consideration. The Paris
MoU, the US Coast Guard and AMSA29 have a quantitative system which ranks ships in
the order of risk and determines if certain ships have to be inspected if a certain risk level
is reached. The other regimes might have an internal target factor in place but
enforcement is not guaranteed by the memorandum. The deficiency codes are more or less
synchronized with the deficiency codes used by the Paris MoU with the exception of the
USCG who uses its own deficiency coding. For the purpose of this analysis, the deficiency
27 USCG Port State Control Speech, https://monkessays.com/write-my-essay/uscg.mil/hq/g-m/pscweb/psc_speech.pdf
28 the USCG inspection coverage is divided into districts
29 In this respect, it is worth noticing, that AMSA has already used similar econometric techniques
in 2002 in order to determine targeting of vessels for inspections and has also implemented the
findings into practise in 2003. The administration has evaluated the implementation and believes
that it has been successful in detecting sub-standard vessels. It is currently reviewing the system
to further enhance it.
16
codes were re-classified by the USCG to fit into the Paris MoU deficiency codes. A full list
of the codes is provided in Table 3. Not all codes are used in the analysis since for
instance code 2700 is related to security which is not part of this thesis. The two
miscellaneous codes 9800 and 9900 where also left out in most of the regressions.
2.1.4. Inspection Systems and Remedy for Non Compliance
Some regimes partly refer to IMO Resolution A.787(19) for operational procedures either
to define clear grounds, inspections procedures or grounds for detention which was
amended by Resolution A.822(21) incorporating the ISM Code. All regimes divide
between priority inspections, initial inspections and more detailed inspections in case of
“clear grounds”.
Table 3: Description of Main Deficiency Codes
Code Deficiency Code Description Code Deficiency Code Description
100 Ship’s certificates and documents 1600 Radio communications
200 Crew certificates 1700 MARPOL Annex I (Oil Pollution)
300 Accommodation 1800 Gas and chemical carriers
400 Food and catering 1900 MARPOL Annex II (Noxious Liquids)
500 Working spaces & accident prev. 2000 SOLAS Operational deficiencies
600 Life saving appliances 2100 MARPOL related oper. deficiencies
700 Fire Safety measures 2200 MARPOL Annex III (Pack.Harmf.Sub.)
800 Accident prevention (ILO147) 2300 MARPOL Annex V (Garbage)
900 Structural Safety 2500 ISM related deficiencies
1000 Alarm signals 2600 Bulk carriers
1100 Cargoes 2700 Security (ISPS Code)
1200 Load lines 2900 MARPOL Annex IV (Sewage)
1300 Mooring arrangements (ILO 147) 9800 Other def. clearly hazardous safety
1400 Propulsion & auxiliary engine 9900 Other def. not clearly hazardous
1500 Safety of navigation
The definition of clear grounds can vary across the regimes but not significantly and
covers in essence the power of the officer to conduct a more detailed inspection if he/she
feels that it is necessary an in particular, if the crew onboard is not familiar with the safe
operation of the vessel. Guidance on clear grounds and a list of detainable deficiencies as
defined by the IMO guidelines (Resolution A.787(19)), Chapter 2.3 are as follows30:
1. the absence of principal equipment or arrangements,
2. ship’s certificates are clearly invalid,
3. certificates are incomplete, not maintained or falsely maintained,
4. evidence from general impression and observation reveals serious hull or structural
deterioration that may place at risk the structural, watertight or weather tight
integrity,
5. evidence from general impression and observation reveals serious deficiencies in the
area of safety, pollution prevention or navigational equipment,
6. master or crew is not familiar with essential shipboard operations relating to the
safety of ships or the prevention of pollution,
7. key members cannot communicate with each other,
8. emission of false distress alerts followed by proper cancellation procedures,
9. receipt of a report of complaint containing information that the ship is substandard
30 based on IMO guidelines for PSC, Resolution A.787(19), chapter 2.3
17
For the USCG, the more detailed inspection is called expanded examination. In the Paris
MoU area, the expanded inspections are a unique concept and apply to oil tankers31 (>
3000 gt and > 15 years), bulk carriers (> 12 years), passenger ships (> 15 years) and gas
and chemical tankers (> 10 years). This type of inspection is based on the EU Council
Directive 106/2001/EC which is currently under review (port state control directive) and
is therefore only applicable for the Paris MoU members. Follow up is specified within the
US Coast Guard Regime and AMSA.
2.1.5. The Definition of a Substandard Vessel
The circumstances leading to a detention is defined in all MoU’s and in all cases,
detention is justified, if the ship is to be seen as substandard. A list of detainable
deficiencies can be found in Appendix 2 for further reference. The guidelines provided by
the IMO PSC procedures for the identification of a substandard vessel and which is also
in adapted form used by the USCG are as follows:32:
“A ship is regarded as substandard if the hull, machinery, or equipment, or operational
safety, is substantially below the standards required by the relevant conventions or if its
crew is not in conformance with the safe manning document, owing to, inter alia:
1. The absence of required principal equipment or arrangement.
2. Non-compliance of equipment or arrangement with relevant specifications.
3. Substantial deterioration of the ship or its equipment.
4. Insufficient operational proficiency, or unfamiliarity of the crew with essential
operational procedures.
5. Insufficiency of manning or insufficiency of certification of seafarers.
6. Noncompliance with applicable operational and/or manning standards;
If these evident factors as a whole or individually make the ship unseaworthy and put at
risk the ship or the life of persons on board or present an unreasonable threat of harm for
the marine environment if it were allowed to proceed to sea, it should be regarded as a
substandard ship”.
2.1.6. The Importance of Ship Types and Trade Flows
The division of ship types is crucial for this analysis for various reasons. First, the legal
base which is the basis for the port state control inspections clearly distinguishes the
ships types and second, port state control is influenced by the trade flows as they
determine the ship types that will visit a particular country of a regime. The decision
factors taken into consideration to divide the datasets per ship types are given in Figure 6
and can be summarized as follows:
• Point 1: Legal Base including the major conventions and related codes distinguishing
different applications based on ship types and the deriving differences in conducting a
port state control inspection.
• Point 2: World Trade Flows to capture exposure of the regimes.
• Point 3: Analysis of Casualties per ship type and their severity.
• Point 4: Correspondence analysis to provide an overall confirmation on the selection of
the grouping of ship types with respect to the port state control deficiencies.
31 Paris Memorandum of Understanding, Annex I, page 38
32 IMO Resolution A.787(19), Chapter 4 and USCG Marine Safety Manual, Vo. II, Section D, page
D1-5
18
While point 1 and 2 is explained in detail here, the rest is presented in later chapters of
this thesis when the variables used to perform these analyses are explained in detail. The
casualty analysis is shown in Chapters 5 while the Correspondences Analysis is treated
under Chapter 3 where ship types and deficiencies are correlated.
Figure 6: The Selection of Ship Types
*) Note: For the legal base, only major codes are taken into account, abbreviations are explained in
the following text to come
The three most important conventions and amendments create a different operating
environment for ship types. Sometimes, age and size of a vessel determines application of
regulations within the ship types. Depending on the convention or code, exact definitions
of ship types are provided and can be found in Appendix 3 for further detail. In general,
the following items underline the decision to split up the ship types.
1. Legal Base (Major Conventions and respective Codes)
SOLAS*)
Passengers
> 12 pax
Cargo Ships
> 500 gt
MARPOL*)
Annex I: Oil Tanker
Annex II: Chemical Tanker
Annex III: Containers
IGC/GCH Annex IV & V, VI: All Ships
BC/IGC/BLU/CSS
STCW &
Load Line*)
Different
requirements
per ship type
Ship Types:
1. General Cargo & Multipurpose
2. Dry Bulk
3. Container
4. Tankers
5. Passenger Vessel
6. Other
Combine with
2. Trade Flows: per MoU (% inspected, % detained, Commercial characteristics)
3. Casualties Statistics: per ship type & severity, per casualty first events
4. Correspondence Analysis: ship types in relation to deficiencies
FSS, LSA, ISM, ISPS
IBC/BCH
IMDG
HSC
Connects with Annex II
and Annex III of Marpol
Oil Tanker > 150gt
All Other > 400 gt
19
SOLAS (Safety of Life at Sea)33
SOLAS is for the requirements of construction of a vessel and all items related to the
safety of life at sea. It divides between passenger and cargo ships. It applies to all ships
above 500 gt on international voyages. Passenger ships are ships exceeding 12 passengers
and all other ships are cargo vessels. SOLAS refers to several codes which can be
particular to certain ship types as follows:
• HSC – International Code of safety for high-speed craft (for ships built after
January 1996),
• IGC – International Code for the construction and Equipment of Ships Carrying
Liquefied Gas in Bulk (for ships constructed after 1st October 1994,
• GCH – Old Gas Carrier Code for ships constructed before 1st October 1994 as per
resolution MSC 5 (48),
• BC – Code of safe practice for solid bulk cargoes,
• BLU – Code of practice for the safe loading and unloading of bulk carriers,
• CSS – Code of safe practice for cargo stowage and securing,
• IBC – International Code for the Construction and Equipment of ships carrying
dangerous chemicals in bulk (for ships constructed after 1st July 1986),
• BCH – Code for the construction and equipment of ships carrying dangerous
chemicals in bulk (for ships constructed before 1st July 1986),
• IMDG – International Maritime Dangerous Goods Code: contains classification,
packing, marking, labeling and stowage of dangerous goods in packaged form –
mainly Containers,
• FSS – Fire Safety Systems Code: Can have specific requirements of fire fighting
equipment per ship type,
• LSA – International Life Saving Appliance Code: can have specific life saving
appliances equipment depending on the ship type,
• ISM – International Safety Management Code and ISPS – International Code for
security of ships and of port facilities.
MARPOL (Prevention of Pollution from Ships)34
o General Application: all ships but surveys (Regulation 4) only apply to oil tankers
above 150gt and all other ships above 400 gt.
o Annex 1: Oil Pollution – Ship Type: All ships but in particular oil tankers.
o Annex 2: Noxious Liquids in bulk – Ship Type: Chemical Tankers and link to IBC
Code.
o Annex 3: Harmful Substances in Packaged Form – Ship Types: Containers and link to
IMDG Code.
o Annex IV (Sewage) and Annex V (Garbage): applies to all ships with some size
restriction on certification requirements.
o Annex VI (Air Pollution): applies to all ships with certain size restriction on
certification requirements.
STCW (Standard Training and Watchkeeping)35 and LL (Load Lines)
STCW applies to all seagoing ships not engaged on national voyages and lists the crew
certification requirements. Requirements can vary by ship type (tankers, chemical
tankers, gas carriers) – e.g. special requirements for tankers for cargo operations and fire
fighting, pollution prevention. The load line convention applies to all ships engaged on
international voyages. It provides different requirements based on ship types and vessel
33 International Convention for the Safety of Life at Sea, 1974 and Protocols 1978 and 1988, IMO
34 International Convention for the Prevention of Pollution form Ships with Annexes I to VI, IMO
35 International Convention on Standards of Training, Certification and Watchkeeping for
Seafarers, 1978 and International Conventions on Load Lines, 1966
20
age and size. The resulting ship types can be seen in Figure 7 before and after regrouping.
Containers are part of general cargo but more sophisticated and although the port states
control inspection of a general cargo ship does not significantly differ from a container
ship, due to the different commercial setup of the two segments, the container ship is kept
separate from the rest of the general cargo ships. In addition, Annex II of MARPOL refers
to Container Ships due to the IMDG Code (Stowage of dangerous goods in packaged
form).
Figure 7: Ship Types Inspected (Before and After Regrouping)
Oil Tanker,
7.9%
Mobile Offshore, 0.5%
Factory Ship, 0.3%
Offshore, 0.7%
RoRo Pax, 0.9%
OBO, 0.9%
Gas Carrier, 1.8%
Special Purp. 0.2%
Heavy Load, 0.1%
HS Pax, 0.1%
Passenger Ship,
2.4%
Reefer Cargo,
2.4%
Other, 3.8%
RoRo Cargo,
5.0%
Tanker, 4.0%
Chemical Tanker,
4.0%
Container,
9.9%
Bulk Carrier,
26.2%
General Cargo,
28.9%
After Regrouping
Passenger
3%
Container
10% Dry Bulk
26%
Other ST
6%
Tankers
19%
General &
Multipurp.
36%
Source: Compiled by author based on PSC dataset
Figure 8 provides an overview of the trade flows based per number of port calls and split
up per ship type for the year 2004 while Figure 9 provides the same information based on
average billion ton-miles36 shipped across the globe for the major commodities. Both
figures provide an overview of the major trade flows and how they affect the port state
control areas.
36 Tonnage of cargo shipped times average distance transported
Figure 8: Overview of Trade Flows (Port Calls per Ship Type, 2004)
Based on Data from LMIU (Lloyd’s Maritime Atlas of World Ports and Shipping Places, 2004)
Figure 9: Major Trade Flows (Average Million Ton-Miles for 2000 to 2004)
Based on Data compiles from Fearnley’s and ISL (Institute of Shipping Economics and Logistics)
23
2.2. The Total Exposure to Inspections
2.2.1. Overview of Inspections in the Name of Safety
The following section will provide a short overview of the different kind of inspections and
surveys that are carried out on ships besides port state control inspections. An overview
of the total exposure to inspections is given in Figure 10. The inspections originate from
various sources and are as follows:
• Port state control inspections and flag state control inspections.
• ISM and ISPS audits due to statutory requirements and which are still sometimes
performed by the flag states but most of the time also delegated to recognized
classification societies.
• Classification surveys on behalf of flag states and to remain in class37.
• Insurance companies such as P&I Clubs for insurance coverage purposes.
• Industry inspections such as vetting inspections performed on oil tankers, chemical
tankers, gas carriers and bulk carriers on behalf of oil majors or other cargo owners or
on behalf of the ship owner. (CDI, OCIMF/SIRE, Rightship, Oil Majors).
• Commercial incentives: These inspections are on request of the ship owner in order to
obtain a quality certificate which will then help in obtaining commercial incentives.
Figure 10: Summary of Total Inspection and Audit Exposure38
Source: compiled by author from various legal sources and inspections
2.2.2. Mandatory Inspections/Surveys/Audits
Port state control and flag state inspections cover the statutory requirements.
Classification societies perform most of the surveys based on the statutory requirements
and by authorization of a flag state. The IMO has tried to synchronize the various types of
37 a ship does not necessary have to be in “class” in order to trade but it is highly recommended.
38 Note: CAS = Condition Assessment Scheme, ESP = Enhanced Survey Program, CAP = Condition
Assessment Program
24
inspections and in essence, four types of mandatory inspections can be identified and are
shown in the graph which covers the inspection areas listed next to the inspection types.
Depending on the type of survey (e.g. initial, annual, renewal, etc.) the content and
intensity of the inspection areas is changed accordingly. An initial survey is a complete
inspection before the vessel comes into service. In addition to the mandatory inspection
types and areas, two mandatory survey programs are identified and are also normally
provided by the classification societies. The first one is CAS (Condition Assessment
Scheme) based on MARPOL and the second is the ESP (Enhanced Survey Program)
based on SOLAS.
The Condition Assessment Scheme originated from an amendment to Annex I of
MARPOL Annex I (Regulation 13G) and can be applied to single hull tankers above 15
years of age. It is intended to complement the requirements of the Enhanced Survey
Program of SOLAS which applies to bulk carriers and oil tankers. Both require a
different scope of survey depending on the age of the vessel including thickness
measurements and rate the coating conditions of the tanks as GOOD, FAIR and POOR
which is sometimes important information for vetting inspections.
To facilitate the various mandatory inspections/survey types shown in Figure 10 and
which need to be carried out, the IMO established the “Harmonized System of Survey’s
and Certification” which can be seen in summarized version in Table 439 where the
following abbreviations are used40:
• A – Annual: general inspection of the items relating to the certificate to ensure that
they have been maintained and remain satisfactory for the service for which the ship
is intended.
• P – Periodical or I – Intermediate: inspection of the items related to the certificate in
order to ensure that they are in satisfactory conditions and fit for the service for which
the ship is intended. It is a more detailed inspection compared to the annual
inspection and is called periodical with reference to the radio equipment and
intermediate for all other types of surveys.
• R – Renewal: same as periodical but more detailed and leads to the issue of a new
certificate and normally involves dry docking.
Table 4: Summary of Harmonized System of Survey and Certification
Years 1 2 3 4 5
Months 9 12 15 21 24 27 33 36 39 45 48 51 57 60
Certificates/Inspection Areas
Passenger Ship Safety Cert. R R R R R
CS Safety Equipment Cert. A A or P P or A A R
CS Safety Radio Certificate. P P P P R
SC Safety Construction Cert. A A or I I or A A R
CF Gas (IGC/GC) A A or I I or A A R
CF Chemical (IBC/BCH) A A or I I or A A R
Load Line Certificate A A A A R
IOPP (MARPOL Annex I) A A or I I or A A R
IPP (MARPOL Annex II) A A or I I or A A R
Based on IMO Resolution A 746 (18)
Abbreviations: CS = Cargo Ship, CF = Certificate of Fitness, IOPP = Intern. Oil Prevention Pollution
Certificate, IPP = Intern. Pollution Prevention Certificate for Carriage of Noxious Liquid Substances in Bulk
39 Extract from IMO Resolution A 746 (18), page 246 and amendment
40 Based on IMO Resolution A.746 (18), page 151 and amendment
25
The table shows the time periods and within which time periods the different types of
surveys can be conducted. It allows a harmonized approach between the various SOLAS
and MARPOL requirements. Passenger vessels have to follow stricter survey schemes
(renewal surveys) than other ship types and a renewal survey has to be carried out each
year versus every five years. Intermediate surveys come into the picture between the 2nd
and 3rd year in order to decrease the inspection time required for a full renewal survey.
Besides the items listed above, two types of audits are identified in Figure 10 – the ISM
(International Safety Management) audit and the ISPS (International Ship and Port
Security) audit which are both SOLAS requirements. This certification is split into a
shipboard part and a company part where the shipboard part has to be completed every
five years with one intermediate audit half way). Some flag administrations have not yet
authorized classification societies to perform these audits but many flag states have done
so and this area is therefore also widely covered by classification societies.
2.2.3. Non Mandatory Inspections
Cargo owners have considerable power through their vetting inspections for certain ship
types (oil tankers, chemical tankers, gas carriers and dry bulk carriers). Sometimes these
inspections originate from the cargo owner or sometimes the ship owner will ask for the
inspection in order to show a certain quality level for a potential cargo owner. Going
through an inspection does not necessarily mean the ship is accepted for cargo. It becomes
clear from the graph that the targeted ship types are chemical tankers, oil tankers, gas
carriers and bulk carriers for the industry inspections while inspections based on
statutory requirements are valid for all ship types. The various inspection systems do
reference each other but there is no cross-recognition. The following paragraphs will
describe the systems further.
CDI (Chemical Industry Institute): CDI inspections originate from the ship owner and
are therefore owned and paid by the ship owner. The owner requests a CDI inspection
and the inspector is appointed to the vessel. Inspections are based on a standardized
questionnaire covering all areas of shipboard operations and are split up into “statutory
requirements” (based on the international conventions), “required” (as per industry Code
of Practice) and “desired” (required by CDI participants or users of the reports)
requirements. An inspection normally takes around 8-10 hours where particular
emphasis is placed on cargo operations and the competence of crew. CDI inspections are
primarily performed on chemical tankers. After the inspection, the report is uploaded to
the CDI system and the ship owner can provide comments to the inspection results. After
that, the ship owner can decide if the report goes alive or not and becomes visible for the
CDI users.
SIRE (Ship Inspection Report Program) and inspection from Oil Majors: Sire
inspections are performed by OCIMF (Oil Companies International Marine Forum) and
originate from cargo owners. The inspectors are appointed by OCIMF and the information
is however owned by the cargo owner but partly made available to other OCIMF members
who can obtain parts of the inspection results for a fee. The inspections also cover more or
less the same areas as CDI with a heavy influence on cargo operations and can take 8 to
10 hours. Ship Owners have some time to comment to the issued report before it becomes
available online. These types of vetting inspections are primarily for oil tankers. While
the standardized questionnaire serves as a basis, some oil majors have additional
requirements and will add these requirements during an inspection which can be
confusing for the ship owners and their crew since no split between statutory
26
requirements and other requirements is made. In addition, oil majors normally perform
their own inspections where the basic requirements are according to the SIRE inspections
but additional requirements per oil major are added to the inspection and are not
published in the SIRE report.
Rightship: Rightship is a ranking system which combines information obtained through
vetting inspections, port state control, casualties, ship particular information and ship
owner information. It ranks vessels according to a rating score (1 to 5 stars where 5 stars
represents a very good vessel with low risk). It is based on a joint venture between BHP
Billiton Freight Trading and Logistics and Rio Tinto Shipping. The inspections cover
tankers and bulk carriers but are primarily for dry bulk carriers. A Rightship Inspection
can take from 8 to 48 hours and covers all aspects of shipboard operations in addition to
ship structure and cargo handling equipment including hatch covers which is important
for dry bulk carriers. Inspectors perform ballast water tank inspections and evaluate the
conditions of the cargo holds.
Greenaward: The last kind of inspection that is performed on vessels (oil tankers)
originates from the Greenaward Foundation. These inspections are paid by the ship
owner. An initial inspection will take approx. 9 hours and cover all aspects of shipboard
operations. In addition to the shipboard audit, an office audit (2 days) is performed to
evaluate the shore based management systems and support to the vessels. After
successful completion, the ship receives a certificate (Greenaward) and the ship owner
can obtain discounts on harbor dues from ports participating in the program. Once the
vessel is “Greenaward Certified”, it needs to undergo annual or intermediate surveys to
remain certified. The Greenaward Foundation is a non-profit foundation. Over the years,
the Greenaward Certificate is not yet officially recognized by port state control regimes.
The approach is more complete and includes shore-side and ship-side elements of the
operations.
In addition to the statutory requirement for CAS and the ESP, some oil majors ask a ship
owner to participate in CAP (Condition Assessment Program) for either hull or
machinery. Those programs are offered by classification societies and are purely
voluntary and provide the ship owner with a rating (CAP 1, 2 or 3 where CAP 1 represent
the best rating) which is important for some oil majors. There is an overlapping of CAP
with CAS where the main difference is that CAS is a statutory requirement and its end
users are the flag states while CAP is a voluntary program required by oil majors who
decides on the minimum of the CAP rating.
2.2.4. Comparison of Inspection Areas
The next section will provide a comparison between the various inspections (excluding
ISPS) that are performed on the vessels and explained previously. It will only concentrate
on inspections performed on ships and only highlight the main areas and items that are
inspected in comparison with each other. The inspection matrix can be seen in Table 5 for
easier reference and was compiled based on the experience the author collected by
observing some inspections and the check-lists of some of the inspectors. The legend and
color coding for the table is provided here below:
x = part of inspection round
r = referenced during inspection
i = actual physical inspection/testing/interviews
s = depends on situation, for class on the type of survey (annual, intermediate, renewal)
Table 5: Inspection Matrix – Main Areas of Inspection in Comparison
Note: Compiled by author Party performing the inspection/survey/audit
Inspection Matrix – Main Areas of Inspection Source of Inspection Port & Flag State or Class Industry
International
Conventions (statutory)
Flag State
Add. Industry
Requirements
Port State (more
detailed insp.)
Flag State
Class Surveys
ISM (emphasis on the
system)
Insurance (P&I Clubs)
CDI/OCIMF
Rightship
Greenaward (Shipside
Part)
Ship Crew Involved
Average Time
onboard (hrs.) 6-8 8 24-48 8 8 8-10 8-48 9
Registration & Administration (Certificates)
Statutory Certificates various x r r r r r r r
Crew Certificates (plus Endorsements) SOLAS/STCW x r r r r r r r
Crew Nationality x r
Medicals x r r r r r r r
Other Certificates for Equipment Testing various x r r r r r r r
Previous Port State Control/Flag State Reports x r r r r r r
Vetting Inspection Reports x r r
Master, Chief
Officer
Living and Working Conditions
Accommodation ILO x x x x x x x
Food (Inspection of Freezers and Galley) ILO x x x x x x x
Living Conditions/Public Spaces ILO x x x x x x x
Rest Periods and Watch Keeping Hours STCW x r r r r r x r
Safety Signs, Protection Equipment SOLAS x x x x x x x x
Gas Detection and Calibration SOLAS/ISM x x i x x x x x
Decontamination showers and eyewash on deck SOLAS/ISM x x i x x x x x
Mooring Arrangements Safe & Maintained SOLAS/ISM x x x x x x x x x
Hospital and Medical Attention x x x x x x x x x
Chief Officer,
Third Officer,
Cook
Note: Compiled by author Party performing the inspection/survey/audit
Inspection Matrix – Main Areas of Inspection Source of Inspection Port & Flag State or Class Industry
International
Conventions (statutory)
Flag State
Add. Industry
Requirements
Port State (more
detailed insp.)
Flag State
Class Surveys
ISM (emphasis on the
system)
Insurance (P&I Clubs)
CDI/OCIMF
Rightship
Greenaward (Shipside
Part)
Ship Crew Involved
Management ISM
Safety Management System/Master’s Authority SOLAS/ISM x r r i r r r i
Safety & Environmental Policy SOLAS/ISM x r r i r r r i
DoC Company and Designated Person Ashore SOLAS/ISM x r r i r x r i
Company Internal Audits SOLAS/ISM x r r i r x r i
Records of Incidents/Near Misses/Accidents SOLAS/ISM x r r i x x r i
Maintenance Routines, Non-conformities SOLAS/ISM x r r i r x r i
Operational Safety – Safety Procedures (Hot Work,
Entry into enclosed spaces) SOLAS/ISM x r r i r r r i
Safety, Fire and Abandon Ship Drills SOLAS/ISM x i(s) i(s) r r x r i
Onboard Communication satisfactory x x x x x x x
Crew Familiarization ISM x x i r x i x
Company Drug and Alcohol Policy and Testing x r x r x
Crew Working Experience x x i x
Manning and Training Policy x r x i x
Security Related Items SOLAS/ISPS x x x x x
Master, Chief
Officer, Third
Officer
Safety and Fire Appliances
SOLAS Training Manuals SOLAS x x x x x x x x x
Muster Lists and Emergency Instructions SOLAS x x x i x x x x x
Lifesaving Appliances (Lifejackets, Immersion Suits,
etc) SOLAS x i i i x i i x x
Lifeboat, Life rafts, Equipment and Launching SOLAS x i i i x i i x x
Chief Officer,
Third Officer
Note: Compiled by author Party performing the inspection/survey/audit
Inspection Matrix – Main Areas of Inspection Source of Inspection Port & Flag State or Class Industry
International
Conventions (statutory)
Flag State
Add. Industry
Requirements
Port State (more
detailed insp.)
Flag State
Class Surveys
ISM (emphasis on the
system)
Insurance (P&I Clubs)
CDI/OCIMF
Rightship
Greenaward (Shipside
Part)
Ship Crew Involved
Rescue Boat and equipment SOLAS x x x i x x x x x
Pilot Ladder, Embarkation Ladders for Lifeboats SOLAS x i i i x i i x x
Oxygen & Acetylene Storage, CO2 room SOLAS x i i i x i i x x
Fire Control Plan SOLAS x r r i x r r r r
Fire Fighting Equipment and Detection SOLAS x i i i x i i x x
Fireman’s outfit, breathing apparatus, air bottles,
EEBD SOLAS x x x i x x x x x
Fire/Foam Hydrants SOLAS x x x i x x x x x
Industry Guidelines/Publications x x x i x
Navigation and Communication
Company Navigation Procedures STCW x x x x x x x x x
Bridge Standing Orders SOLAS x x x x x x x x x
Passage Planning STCW x x x x x x x x x
Chart Corrections SOLAS x x x x x x x x x
Nautical Publications up to date various x x x x x x x x x x
Navigational Equipment Working (GPS, Speed Log,
Radar, Echo Sounder, Compass, Navtex etc.) SOLAS x x x i x x x x x
Dead man Alarm (when applicable) x x x x x x x x x
Guidelines for the prevention of fatigue x r
Crew knows how to operate equipment STCW x x x x x x x x x
VDR/AIS SOLAS x x x i x x x x
Compass Error Log STCW x x x x x x x x
Chief Officer,
Second
Officer
Note: Compiled by author Party performing the inspection/survey/audit
Inspection Matrix – Main Areas of Inspection Source of Inspection Port & Flag State or Class Industry
International
Conventions (statutory)
Flag State
Add. Industry
Requirements
Port State (more
detailed insp.)
Flag State
Class Surveys
ISM (emphasis on the
system)
Insurance (P&I Clubs)
CDI/OCIMF
Rightship
Greenaward (Shipside
Part)
Ship Crew Involved
Compass Deviation Card SOLAS x x x x x x x x
Navigation Lights COLREG x x x i x i x x
GMDSS Operations and Testing SOLAS/STCW x x x i x x x x
EPIRB and SART SOLAS x x x i x x x x
Ship and Cargo Operations including Pollution Prevention
Loading and Stability Manuals IBC/BCH x r r r x r r x x
Cargo loading limitations IBC/BCH x r r r x r r x x
Damage/survival stability guidelines IBC/BCH x r r r x r r x x
Procedures and Arrangement Manual MARPOL x r r r x r r x x
High level alarms operative IBC x x x i x x x x x
Bilge Alarms SOLAS x i x i x i i x i
Portable or fixed gas detection systems SOLAS x x x i x x x x x
Inert gas system (for oil tankers) or other systems to
blanket cargo (e.g. nitrogen) x x x x x x x x
15 ppm Alarm MARPOL x i i i x i i x i
Oil-Mist Detector SOLAS x i i i x i i i i
SOPEP, SMPEP MARPOL x r r r x r r x x
Cargo Record Book, Oil Record Book, Garbage RB MARPOL x r r r x r r x x
Tank cleaning and washing including COW MARPOL x r r x x x x x
Industry Guidelines/Publications x x x x x
Cargo Operations in General including Pump Room various x x x x i x x
Chief Officer,
Chief
Engineer
Cargo Transfer Operations various x x x x i x x
Note: Compiled by author Party performing the inspection/survey/audit
Inspection Matrix – Main Areas of Inspection Source of Inspection Port & Flag State or Class Industry
International
Conventions (statutory)
Flag State
Add. Industry
Requirements
Port State (more
detailed insp.)
Flag State
Class Surveys
ISM (emphasis on the
system)
Insurance (P&I Clubs)
CDI/OCIMF
Rightship
Greenaward (Shipside
Part)
Ship Crew Involved
Fuel Testing, sulphur content measurement x r
Anti-fouling system for hull coating (TBT free) MARPOL x r r r
Additional Oil Pollution Prevention Measures x r r
Machinery Related Areas including Engine Room
Engine Room Standing Orders SOLAS/ISM x x x x x x x x x
Planned Maintenance System SOLAS x r r i x r i x x
Emergency Steering Gear SOLAS x i i i i i i i
Emergency Fire Pump SOLAS x i i i i i i i
Emergency Generator SOLAS x i i i i i i i
Emergency Batteries SOLAS x x x x x x x x
Testing of Black Out and Reverse Polarity i(s) i i(s) i(s) x
Overall Cleanliness and Appearance of ER x x x x x x x x
Chief
Engineer,
First or
Second
Engineer
Stability & Structure
Enhanced Survey Program, Thickness
Measurements SOLAS x r r i(s) r r r r r
CAS (Condition Assessment Scheme) MARPOL x r r i(s) r r r r r
Inspections of Ballast Tanks, Cargo Tanks, Void
Spaces, Cofferdams for Condition of
Coating/Corrosion SOLAS/MARPOL x r i(s) r r x i r
Rating System for Condition of Coating/Corrosion as per ESP/CAS x r r i(s) r r r i r
Conditions of Hull and Superstructure (e.g. Hatch
covers) Good/Fair/Poor x x x i(s) x i(s) x i i
Class Status Report/Outstanding Class Conditions
and Memoranda x r r r r r r r
Master, Chief
Officer, Chief
Engineer
32
The table is split into the main areas of inspection such as an administrative part, living
and working conditions onboard the ship, the safety management system, areas related to
safety and fire appliances, navigation and communication, ship and cargo operations
including pollution prevention, machinery related areas and stability and structural
related areas. The source of inspection is listed when applicable which can be a
combination of the international conventions plus flag state requirements and additional
industry requirements besides the statutory requirements. Next, the parties performing
the inspections are identified and their coverage is indicated. The last column provides
guidance on the crew that is involved in the inspections. For some vetting inspections and
class surveys, the ship superintendent will normally also be onboard the vessel to Help
the crew.
The inspection normally starts with a short briefing of the master and review of the ship’s
certificates and crew certificates. This is followed by a deck round starting from the top
(bridge) down to the main deck areas with stops at the life boats, safety lockers, fire
fighting equipment. The bridge will also cover more detailed questions about passage
planning, chart corrections and the checking of the navigational equipment, lights and
radio equipment. Deck rounds can entail stops at the paint locker, the CO2 room (if
applicable), storage location for Acetylene and Oxygen Cylinders, the pump room (if
applicable), the emergency generator, checking of fire hoses and lifebuoys, mooring
arrangements and winches as well as visits to the forepeak. The last section of the
inspection normally covers the cargo control room and the engine room with the testing of
the emergency fire pump and emergency steering gear and a general round around the
engine room including the areas used for welding. If ballast water tank inspections or
inspections of the cargo holds are performed, the inspector will announce this in the
beginning of the inspection so that it can be prepared accordingly. It is not easy to access
ballast water tanks or cargo holds during normal cargo operations. The inspection is
normally finished up with a round of the galley storage areas for food (dry store, freezers,
etc.) and the crew mess and day room.
One can see from the table, that certificates are referenced by everybody and that the
main areas of inspections are more or less covered by all types of inspections. Living and
Working Conditions of the crew are mainly covered by the inspection rounds and the
actual living space of the crew (their cabins and other facilities) is hardly inspected.
The industry inspections such as CDI/OCIMF, Rightship and Greenaward pay more
attention to ship and cargo operations and spend considerably more time with crew
members to interview them on operational issues. These items are primarily referenced
during port and flag state inspections. Drills might be performed by some safety regimes
such as the USCG or flag states but are not performed frequently by other inspectors and
the inspection of the lifeboat primarily emphasizes on the overall condition of the lifeboat,
its launching devices and embarkation procedures as well as the lifeboat equipment. The
inspection of safety and fire appliances is also covered by all types of inspections. For
some items, the inspection might go into more details and entail the actual testing of the
equipment which is merely performed during class surveys while other will only refer to
expiry dates of the last survey/inspection that was performed shore side (e.g. for life
rafts).
Items related to navigation and communication is also covered by all inspection types
including chart corrections, passage planning, nautical publications and the overall
impression of the officer on watch with reference to the handling of the equipment (radar,
echo sounder, radio equipment, etc.)
33
Difficult to inspect is the safety management system since it draws from all areas. All
inspections do cover some ISM related questions and the actual validity of the presented
paperwork only becomes evident after a general deck round and interview with crew
members. It might be that the paperwork related to ISM is in compliance but not
implemented onboard. Some of the findings in part III of this thesis support the idea of
lack of proper implementation of ISM despite all the inspections that are performed in the
name of safety. Inspection systems such as the vetting inspections do emphasize more on
this aspect where Greenaward also performs company audits shore-side. Authorized
classification societies or flag states perform separate audits to ensure that the safety
management system is implemented in practice but inspections due to the time
constraint in conducting surveys is normally only looking at the surface.
As mentioned earlier, ballast water tank and cargo holds inspections are difficult to
perform and are primarily done by classification societies. Rightship pays more attention
to actual physical inspections while port states will only proceed either required by their
policies (e.g. expanded inspections in the EU) or when perceived necessary. The various
programs (ESP, CAS or CAP) for the conditions of coatings in the ballast tanks and cargo
tanks (when applicable) are normally only referenced and physical inspections thereof are
kept to a minimum. The table gives a good indication of some of the overlapping of the
inspections that are performed on ships from port states, flag states, vetting inspections
and other industry inspections. The inspections performed by classification societies on
behalf of the flag state to a certain extent have a different scope since they are the basis
to extend or renew the validity of a certificate and are therefore statutory. The flag state
inspections performed in addition to the surveys from classification societies primarily
serve as a means to check the performance of classification societies as a recognized
organization to conduct these surveys on behalf of the flag state.
The system generates a substantial amount of inspections performed on vessels with
areas that are inspected and re-inspected frequently. In the case of port state control and
based on the total dataset, one can see the total inspection and detention frequency in
Figure 11 which is based on an average of four years41 since not all regimes provided data
for the whole time frame. Based on the 183,819 port state control inspections and 26,020
aggregated ships, this aggregates to 7 inspections within four years or approx. 1.7
inspections per year per ship.
Figure 11: Inspection and Detention Frequency of Vessels (1999 to 2004)
Inspection Frequency (1999 to 2004)
2 to 5 times
28%
6 to 10
times
33%
more than
20 times
1%
once
15%
16 to 20
times
5%
11 to 15
times
18%
Detention Frequency (1999 to 2004)
4 and
more, 6%
once
66%
3 times
7%
twice
21%
41 The total amount of years for each regime was converted into month of inspections and then
converted into total amount of years (291 total months/12 = 24.25 years/6 regimes = 4 years)
34
2.3. Summary of Costs of Inspections and Insurance Claims
Table 6 and Table 7 give an overview of the estimated costs of port state control
inspections and other inspections that are performed onboard ships. The port state
control inspection costs are divided into inspections with zero deficiencies and inspections
with deficiencies who might take more amount of time onboard the ships. In addition, a
20% administrative charge42 is added to the costs. The surveyor costs change from
country to country and this change is not taken into consideration since data from 53
countries are in the total port state control inspection dataset. In reality, the presented
figures might therefore be different but for the purpose of this study, the figures should
merely give an overall indication on the costs associated with port state control.
Table 6: Total Estimated Port State Control Inspection Costs (USD)
# of Inspections # Hours/Insp. Rate Total 4 years Per Year Per Insp.
zero def. 98,895 395,580 4 126 50,038,229 12,509,557 506
with def. 84,924 509,544 6 126 64,453,914 16,113,479 759
Total 183,819 905,124 114,492,143 28,623,036 623
+20% Admin 137,390,572 34,347,643 747
Note: 1 hour surveyor = 72 British Pounds43, 1 GBP = 0.5692 USD, Administrative Costs = +20%,
compiled by author
The estimated inspection costs of a port state control inspection is USD 747 per
inspection or a total of USD 34,3 million for all types of inspection. Inspections associated
with zero deficiencies and without administrative costs are estimated to be at USD 12,5
million per year or USD 50 million for the total four year period. Looking at the total
estimated costs per year per vessel and including shore based costs for ship owners and
operators, the result can be seen in Table 7.
Table 7: Summary of Inspection Frequency, Allocated Time and Costs (USD/year)
in USD
Estim.
Frequency Time (hrs)
Estim.
Costs
Estim.
Costs
Estim.
Total
Cost
Inspection Type yearly*)
Allocated
Onboard
Shore
Side/Insp.
Ship
Side/Insp.
Per
Year
Port State Control 2 5 747 288 2,070
Flag State Control 1 8 747 441 1,188
Class Annual Survey 1 10 10,362 517 10,879
ISM Audit 0.5 9 2,682 487 1,584
Insurance (P&I Club) 0.5 8 3,048 441 1,744
Industry Inspections: Tankers 6 10 17,663 566 29,702
Industry Inspections: Bulk 1 10 6,250 566 6,816
Industry Inspections: Other 0 0 0 0 0
Total Tankers 11 50 35,248 2,739 47,166
Total Dry Bulk 6 50 23,835 2,739 24,280
Total Other Ship Types 5 40 17,585 2,173 17,464
Note: compiled by author, *) the ISM Audits and P&I Club Inspections are not performed yearly;
For Industry Inspections, administrative portion of 20% are added which might be higher in
reality due to substantial amount of preparation work
42 as per information obtained from the Maritime and Coast Guard Agency, UK
43 as per information obtaine from the Maritime and Coast Guard Agency, UK
35
The data is a summary from several sources from the industry such as classification
societies and ship owners of which the companies would like to remain anonymous. The
table is split up into three groups. The estimated total frequency of inspection for tankers
(oil and chemical tankers) is estimated to be at 11 inspections per year which can of
course vary per ship type and age of the vessel. As the age increases (above 10 or 15
years), the frequency of industry inspections can increase. For dry bulk carriers, the
inspection frequency is estimated to be 6 inspections and all other ship types, it is
estimated to be at 5 inspections.
Shore based costs include the costs for the inspections itself including travel expenses as
well as an administrative portion of preparing the inspections and to comment on the
inspection reports which can take considerable amount of time on the ship operator’s or
owner’s side. Total costs per year per vessel associated with inspections vary from approx.
USD 47,000 for tankers to USD 17,500 for other ship types which are not part of the
industry vetting inspection system. These costs represent total costs where the ship
owner’s portion would be the portion without port state control and the flag state
inspections.
It is difficult to bring these costs in relation to the costs that are associated with
casualties. One attempt was made to gather insurance claim data but only two sources
from the industry could be obtained of P&I Clubs44 who were willing to provide claim
figures for the years 2000 to 2004 per ship type and claim category. An average claim
figure per ship was calculated and is presented in Table 8.
Table 8: Average P&I Club Claim Figures per Vessel and Year (2000 to 2004)
Average Claim in
USD (2000 to 2004)
Cargo/GA
Collision
Contact
Personnel
Pollution
Other
H&M
Average ST
GG & Container 9,794 36,071 18,084 14,396 46,796 16,303 151,181 41,804
Dry Bulk 14,767 58,311 9,955 11,495 51,078 73,207 182,399 57,316
Tanker 42,936 88,277 21,079 18,216 272,016 44,596 609,252 156,624
Passenger 1,885 56,142 9,209 15,310 18,616 9,015 883,549 141,961
Other 9,231 18,801 478 6,446 6,886 38,357 557,692 91,127
Average/vessel 15,722 51,521 11,761 13,172 79,078 36,296 476,815 97,766
Note: compiled by author, GA = general average45, H&M = Hull and Machinery
In reality, the figures will be higher than presented in the table due to the fact that the
claim figures are based on actual claims above the deductible. The deductible can vary
per ship type, size, or ownership of the vessel. In addition, it varies considerably between
hull and machinery (H&M) and other P&I club claims46. The figures presented in the
table can therefore only be seen as a very rough idea of the magnitude of casualty claims
per ship type. It is difficult to compare the costs of inspections with the insurance claim
costs but an overall comparison per ship type is given in Table 9. The result indicate that
44 The P&I Clubs wish to remain anonymous.
45 legal principal of maritme law according to which all parties in a sea venture proportionally
share any losses resulting from a voluntary sacrifice of part of the ship or fleet to save the whole
in an emergency (definition from: http://en.wikipedia.org/wiki/General_average)
46 As per industry sources, the deductible for Hull and Machinery can be between USD 50,000 to
250,000 and for P& I Clubs between USD 5,000 – 30,000 for personnel and USD 10,000 to 100,000
for all other claims.
36
the total inspection costs per ship of USD 24,768 seems to be reasonable in relation to the
average insurance claim costs of USD 97,766 which in reality might be an even higher
figure.
Figure 12 shows the split up of the inspection costs and insurance claims per ship type in
order to see the relation between the two categories. One can easily see that the
percentages are not in line for passenger vessels where the insurance claims are
substantially higher than the inspection costs. For tankers on the other hand, the higher
inspection costs seem to be in line with the insurance claims due to the high costs that are
for instance involved if pollution is involved in a casualty.
Table 9: Average Inspection Costs versus Insurance Claims in USD (2000 to 2004)
In USD per vessel
Inspection
Costs
Insurance
Claims
GG & Container 17,464 41,804
Dry Bulk 24,280 57,316
Tanker 47,166 156,624
Passenger 17,464 141,961
Other 17,464 91,127
Average per Vessel/year 24,768 97,766
Compiled by author
Figure 12: Inspection Costs versus Insurance Claims in % to Total
14.1%
19.6%
38.1%
14.1% 14.1%
8.6%
11.7%
32.0%
29.0%
18.6%
0%
5%
10%
15%
20%
25%
30%
35%
40%
GG & Container Dry Bulk Tanker Passenger Other
inspection costs insurance claims
Compiled by author
It is difficult to conclude if the inspection costs are in relation to the insurance claims and
if the relative high frequency of inspections on oil and chemical tankers is justified since
the costs of preventing accidents due to inspection are not known. In addition, the
insurance claims costs are in reality higher than shown here and only based on two P&I
Clubs. For the regression analysis on casualties and the effect of port state control in the
probability of having a casualty, the insurance claim costs were not taken into
consideration but are based on the seriousness of a casualty instead.
To get an impression about the difference in insurance claims of vessel that were
inspected with vessels that were not inspected, the following graphs should give an
37
impression to see the difference based on claim costs. The graphs were produced the
following way. The total casualty dataset was combined with the insurance claim costs
listed in Table 8 and then aggregated per IMO number in order to obtain an average
claim amount per ship since one ship can have more than one type of claim. The result
was then merged with the inspection dataset in order to identify if a ship has been
inspected or not inspected by port state control. The figures do not match the figures
presented in Table 9 since they are averages across all ship types and based on the total
casualty dataset and not the claim information received from the P&I Clubs directly.
Figure 13 gives an overview of the total average claims of inspected vessels versus not
inspected vessels. One can easily see that not inspected vessel have higher average claim
costs than inspected vessels. The same applies for Figure 14 for the average claim costs
per ship type based on the casualty dataset but using the average claims that were
calculated and shown in Table 8.
Figure 13: Average Claims of Inspected versus Non-Inspected Vessels
103,672
34,685
0
20,000
40,000
60,000
80,000
100,000
120,000
Not PSC Inspected PSC Inspected
Average Insurance Claims
Based on inspections from 1999 to 2004
Figure 14: Average Claims of Inspected versus Non-Inspected Vessels per Ship Type
53,731
61,065
31,551
151,916
81,079
112,306
25,592
29,301
18,935
73,620
32,940
38,866
0
20,000
40,000
60,000
80,000
100,000
120,000
140,000
160,000
180,000
general cargo dry bulk container tanker passenger other ST
Average Insurance Claims
PSC Not Inspected PSC Inspected
Based on inspections from 1999 to 2004
38
One can see that the differences between inspected and not inspected vessels is greatest
for tankers and other ship types which are easily explained with the frequency of
inspections performed on oil tankers. The next chapter will give a short overview of
remedies that could be identified to enhance compliance and new legislative
developments in the area of maritime safety.
2.4. Remedies to Enhance Compliance and New Developments
2.4.1. Industry Based Incentives
Two systems have been identified which take commercial incentives to improve safety
into account. The first one is the system established by the Greenaward Foundation
explained previously where the ship owner receives discounts on harbor dues from
various ports around the world. Data from the Greenaward Foundation was merged with
the total dataset for port state control inspections to see if there is any substantial
difference between ships that have a “Greenaward” certificate versus ships that are not
certified and the result can be seen in Figure 15.
Figure 15: Mean Amount of Deficiencies on Greenaward Certified Ships
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
Code_0100
Code_0200
Code_0300
Code_0400
Code_0500
Code_0600
Code_0700
Code_0800
Code_0900
Code_1000
Code_1100
Code_1200
Code_1300
Code_1400
Code_1500
Code_1600
Code_1700
Code_1800
Code_1900
Code_2000
Code_2100
Code_2200
Code_2300
Code_2500
Mean Number of Deficiencies
Certified Not Certified
Based on inspections from 1999 to 2004
On average, ships that are certified by “Greenaward” show a lower amount of deficiencies
in almost all categories versus ships that are not certified. In the area dealing with safety
and fire appliances (code 600 and 700), this difference is significant. The same applies for
code 900 (structural safety), code 1200 (load lines), code 1500 (safety of navigation) and
code 1700 (MARPOL Annex I: Oil Pollution). Overall, the mean amount of deficiencies per
inspection is at 2.6 deficiencies for not certified vessels versus 0.7 for certified vessel and
one can conclude that Greenaward certified vessels show a better performance in port
state control than non-certified vessels and that the incentive to perform better does pay
off for the ship owner. On the other hand, the certification does not necessarily mean that
the ship gets inspected less within the port state control regimes. Probabilities of casualty
will be shown in part III of this thesis (Figure 83).
39
2.4.2. Current Regulatory Based Incentives
The second system which has been identified to provide incentives for ship owners to
perform better is the Qualship21 program of the US Coast Guard and the revised
directive on port state control of the EU. For the USCG and depending on the vessel’s
inspection history, ownership, flag and classification society, Qualship21 status can be
obtained. Once received and depending on the ship type, the vessel will then be exposed
to less port state control inspections. For cargo ships, this period is 2 years. Passenger
vessels can also participate but will not benefit from fewer inspections due to the
perceived risk. The filtering process used by the US Coast Guard also contains past
casualty history of the vessel while it is not clear how this data is incorporated into the
overall process.
2.4.3 New Developments in the EU
New developments to review the present targeting system of the EU area through the
Paris MoU derive from a new proposal of Council Directive 95/21/EC47 as part of the
Third Maritime Package. Besides the revision on port state control, the EU Third
Maritime Package also contains other measures such as the development of a common
accident investigation48 and reporting scheme and the clarification of ports of refuge. A
very interesting measurement with respect to flag states is the transfer of the IMO
Voluntary Member States49 Audit Scheme into EU law which will then be mandatory and
enforceable under EU law. The IMO Voluntary Member States Audit has been adopted in
December 2005 during the 24th Session of the IMO Assembly and will be explained in one
of the subsequent chapters. This is a welcoming measure to improve enforcement and
acceptance of the scheme in the maritime industry.
The main objective of revision of the port state control directive is to ensure better use of
present resources to fully cover high risk vessels and to decrease the inspection burden on
low risk vessels. This thesis can support this new change in course based on the findings
of this thesis in part III. The main points of the proposal are summarized below50:
o Harmonize application of the directive with respect to the new member states and
harmonized training of port state control officers.
o Revision of the process of banning ships (refusal of access) including all ship types
and the proposal for permanent banning from community ports.
o The increase of transparency through the publication of black lists for substandard
owners and a list of ranking for classification societies.
o Revision of the present targeting system to a risk-based system including a
revision of the application of expanded inspection and the improvement of the
follow up of outstanding deficiencies. The new inspection regime will also try to
establish a system which should allow the division of inspections across the whole
EU area.
o More attention is placed to the human factor such an increased awareness of
falsified crew certificates. In addition, any complaints of crew members regarding
living and working conditions onboard a vessel should be investigated and also
followed up.
47 http://europa.eu.int/comm/transport/maritime/safety/2005_package_3_en.htm
48 Proposal for a Directive establishing the fundamental principles governing the investigation of
accidents in the maritime transport sector and amending Directive 1999/35/EC and 2002/59/EC,
COM (2005). 590 final version of 23rd November 2005
49 IMO Resolution A.973(24) and IMO Resolution A.974(24)
50 Based on the Proposal for a Directive, COM (2005), 588 final version of 23rd November 2005
40
The revision contains a new approach to banning ships from the EU. First, it incorporates
a time frame for period of banning of 3 months for the first banning and 12 months for the
second banning and second, it allows for permanent banning for an additional detention
after the second banning. It is difficult to estimate the effect of this new proposal once it
has been implemented. It will certainly have an effect on one of the categories of ship
types which could not be banned previously but have the highest number of detentions –
the general cargo ships. Some of the shift of substandard vessels across the regions can
already be measured and is shown in part III of this thesis.
On the other hand, the effect of permanent elimination of substandard ships versus a
shift towards other trading areas is difficult to foresee. The act of detention is in general
good for the vessel since it ensures that rectification of the deficiencies found is dealt with
and that the vessel complies with the minimum standards after the release of detention.
After the ship has been released from detention, the banning based on the criteria of a
ship not meeting the minimum safety standards is not correct but is merely a tool to
punish ship owners for not complying. It is questionable if this remedy will really raise
the safety standards and what the economic impacts might be on the long run. A different
approach might be to impose fines to ships that had multiple detentions instead of
banning them altogether. In this way, there is more control over substandard ships that
can benefit from inspections and detentions, in particular for the living and working
conditions of some crew members who serve on substandard vessels.
In applying main criteria for detention including the areas that are covered under the
ILO conventions, the port state control officer is only supposed to look at the forthcoming
voyage. The overall safety of a vessel does not only depend on the forthcoming voyage but
derives from a continuous operation of the vessel, especially of the successful
implementation of the safety management system onboard. One should therefore look at
the safety management system in place and its probability of continuation beyond the
forthcoming voyage since the probability of the ship being inspected at the forthcoming
voyages is lower in the same regime. Cooperation between the regimes would therefore be
beneficial and beyond the boundaries of the ship’s sailing area at the time being but
within the next 2 to 3 months.
Fatigue is defined by the IMO as follows51: “A reduction in physical and/or mental
capability as the result of physical, mental or emotional exertion which may impair nearly
all physical abilities including: strength, speed, reaction time, coordination, decision
making or balance.” In the maritime environment, it is difficult to keep a concise
schedule with respect to resting periods. Besides the quantity of sleep, the quality of sleep
is also important and if one is deprived from sleep for a long period, the sleep deficit
accumulates. It is difficult for a port state control officer or any other inspector to
correctly observe and identify deprivation of sleep and or the signs of fatigue at the time
of the inspection when crew members normally try to accommodate the inspector as good
as possible. It mostly happens that some of the resting periods are also violated due to an
inspection.
Another area of new EU legislation deals with the liability of classification societies and
suggests the introduction of financial sanctions for failures to comply. While these
suggestions are still part of the Third Maritime Package, two additional measures deal
with the harmonization of criminal law enforcement against ship-source pollution
51 Goodwin,S (2006), Does Work Keep you Awake at night?, Learning from Marine Incidents III,
Conference, The Royal Intitute of Naval Architects (January 2006: London, UK)
41
(Council Framework Decision) and penalties for infringements (Directive)52. Both
measures have been adopted and should be implemented in January 2007 and March
2007 respectively. The council framework decision does not fall under the EU court of
Justice jurisdiction53 while the directive does. Both measures complement each other and
provide a framework to address infringements including criminal offences and extend
penalties not only to ship-owners or master of a vessel but also to cargo owners and
classification societies.
A third measure dealing with the civil liability and financial securities of ship-owners in
the form of a directive has been proposed by the Commission in November 2005 and is
under review by the Council and European Parliament. The directive’s objective is to
increase the civil liability of ship owners and to make insurance coverage obligatory
which in the view of the EU is not covered by the international regime. If further
introduces protection for seafarers in case of abandonment. The respective international
conventions are as follows:
1. International Convention on Civil Liability for Oil Pollution Damage, 1992( CCL)
& 1996 Protocol
2. International Convention of 1996 on Liability and Compensation for Damage in
Connection with the Carriage of Hazardous and Noxious Substances by Sea (HNS)
– not in force yet
3. International Convention on Civil Liability for Bunker Oil Pollution Damage, 2001
(Bunker Oil Convention)
2.4.4. The New ILO Consolidated Maritime Labor Convention (2006)
In February 2006, the ILO adopted the new Consolidated Maritime Labor Conventions
which updates and consolidates 6854 existing ILO maritime conventions and
recommendations. The adoption and revision of this new convention is expected to
improve the living and working conditions of the seafarers as well as clearly state their
rights. It consists of five titles as follows55:
Title 1: Minimum requirements for seafarers to work on a ship
Title 2: Conditions of employment
Title 3: Accommodation, recreational facilities, food and catering
Title 4: Health protection, medical care, welfare and social security protection
Title 5: Compliance and enforcement
One of the main purposes of the new conventions is also to enhance implementation of
common standards across the industry which is treated in Title 5. In order to do so, it
provides for a complaint procedure for seafarers and a certification system of labor
standards by the flag states.
The certification consists of two parts – the maritime labor certificate and a declaration of
maritime labor compliance. The maritime labor certificate is issued by the flag state upon
52 Council Framework Decision 2005/667/JHA of 12th July 2005 and Directive 2005/35/EC of 7th
September 2005
53 since it was adopted under Title VI of the Treaty of the European Union and therefore falls
under the third pillar of the EU. Framework decisions are binding for the members states but the
choice of method on how to implement the decision is left to the member state.
54 ILO: https://monkessays.com/write-my-essay/ilo.org/public/english/bureau/inf/event/maritime/index.htm
55 ILO: https://monkessays.com/write-my-essay/ilo.org/public/english/dialogue/sector/papers/maritime/consolcd/conv.htm
42
an inspection where the inspection can be delegated to authorize classification societies
and is valid for five years where periodic inspections are required. The declaration is a
document of the flag state which summarizes the national laws and regulations as well as
a plan of the ship owner or manager on how to implement these and how continuous
improvement can be achieved. The areas that are covered under the declaration are as
follows56:
1. Minimum Age
2. Medical Certification
3. Qualification of Seafarers
4. Seafarer employment agreements
5. Use of a licensed or certified or regulated private recruitment and placement
service
6. Hours of work or resting hours
7. Manning levels for the ship
8. Accommodation
9. On-board recreational facilities
10. Food and catering
11. Health and safety and accident prevention
12. Onboard medical care
13. Onboard complaint procedures
14. Payment of wages
A continuous record of deficiencies that were found during inspections is to be kept
onboard as well as applicable remedies. The system is similar to the safety management
system. These records are then made available to inspectors for flag states and port
states.
Port State control will inspect the certificates in an initial inspection and if clear grounds
indicate the need for a more detailed inspection, the port state control officer will do so
and can detain a vessel is clear evidence of non compliance to the provisions in the
declaration of the maritime compliance is found. In addition, complaints from seafarers or
professional bodies that have an interest in the safety of the seafarers are to be treated
with confidentiality and can constitute grounds for a more detailed inspection. In this
respect, it is worth noticing that IMO has developed PSC guidelines on seafarer’s working
hours during FSI57 14 (June 2006).
The new convention is a very welcoming development in trying to increase the acceptance
of the ILO standards and enforce them. The inspections will probably be performed
primarily by the recognized organizations – the classification societies and will be added
to the other inspections or surveys that are performed onboard the ships. It will be
interesting to see how the complain procedure for seafarers will work in practice and how
inspectors will inspect the areas. It is a first step towards improvement of living
conditions. Hopefully, more attention will be given to the living and working conditions of
some seafarers during inspections so that improvements can be made.
56 ILO: https://monkessays.com/write-my-essay/ilo.org/public/english/dialogue/sector/papers/maritime/consolcd/conv.htm
57 Flag State Implementation Sub-Committee Meeting (June 2006). The correspondent Working
Paper is FSI 14/WP.3, 8th June 2006
43
2.4.5. The Voluntary IMO Member Audit Scheme
The IMO has adopted two important instruments in December 2005 which are Resolution
A.973 (24) namely the Code for the Implementation of Mandatory IMO Instruments and
Resolution A. 972(24) which provides the Framework and Procedures for the Voluntary
IMO Member State Audit Scheme. It comprises flag states, port states and coastal states
and aims in improving the interpretation and implementation of the IMO instruments by
the parties involved. The mandatory instruments in the code are as follows:
1. SOLAS and Protocols of 1978 and 1988
2. MARPOL and Protocol of 1997
3. STCW
4. Load Lines and Protocol 1988
5. Tonnage Convention
6. Convention on the International Regulations for Preventing Collisions at Sea
The code clearly defines the responsibilities of flag states, port states and coastal states in
implementing the conventions mentioned earlier. It further builds on the United Nations
Convention on the Law of the Sea (UNCLOS, 1982) in which administrations are
responsible to ensure that a ship and its crew is fit to operate a vessel. The code allows
authority to conduct surveys to be delegated to recognized organizations (the
classification societies) of which the guidelines are found in Resolution A.739 (18)58. Areas
which are covered in the code are the responsibilities of a maritime administration with
respect to the administration of the registry and enforcement either on the flag or port
state level, accident investigations and the qualifications of surveyors.
The framework lists detailed audit procedures to audit the flag and port states and
coastal states. As of now, it is a voluntary scheme but the first of its kind for IMO to allow
the organization to perform audits. The scheme is seen as a tool to help the
administrations to implement the conventions into national law and to improve their
interpretation. The areas which are covered during the audit are as follows:59
1. Organization and authority
2. Implementation of IMO conventions into national law
3. Jurisdiction and Enforcement arrangements
4. Approval of Authorized organization
5. Appointment of surveyors
6. Accident investigations and Reporting to the IMO
The audit process is shown in Figure 16 below with detailed steps. It is split into a
planning process, an executing process and a reporting and follow-up process. The
member state requesting the audit agrees to make some of the findings available for other
member state which should enable to increase transparency and provide Helpance to
flag states that are willing to learn lessons from the audit of other administrations.
The auditors are appointed by IMO. After the planning stage, the audit is performed and
an interim report is issued to the member state and at this stage is only made available to
the audited member state. The member state then has the opportunity to make comment
on any disagreements in the interim report by writing. The final audit report is then
58 IMO Resolution A.739 (18) adopted on 4th November 1993: Guidelines for the Authorization of
Organizations acting on behalf of the Administration
59 IMO Resolution A.974(24): Framework and Procedures for the voluntary IMO Member State
Audit Scheme
44
issued to the member state and in consolidated version in a form of an audit summary
report will be made available to the IMO secretariat. The member state is also requested
to address any of the audit findings.
Figure 16: Auditing Process
Source: IMO Resolution A.974(24)
The idea of conducting voluntary audits is a very welcoming development within the IMO
framework and if transferred into EU law in the case of the EU will certainly help in
improving the implementation and interpretation of IMO conventions. Although a
voluntary tool, it might become more or less mandatory for some large flag states and
allows for giving an incentive through decreased port state control inspections.
2.4.6. Newest Developments in the Area of PSC at IMO
FSI60 14 held in June 2006 enjoyed many submissions for agenda item 6 – Harmonization
on Port State Control. During the session, it was acknowledged that a framework on
global harmonization and co-operation on PSC activities should be established. The
working group recognized that the following should be reached globally61:
o harmonized (compatible) procedures of PSC inspections,
o harmonized (unified) actions against vessels having deficiencies; and
o availability of PSC inspection results to all officers conducting PSC inspections
worldwide.
The group further identified specific tasks that needed to be implemented in order to
harmonize port state control activities which are listed here below in condensed format:62
60 Flag State Implementation Sub-Committee Meeting, IMO, June 2006
61 FSI 14/WP.3, Report of the Working Group on Harmonization of Port State Control Activities,
page 5
62 FSI 14/WP.3, Report of the Working Group on Harmonization of Port State Control Activities,
page 6
Selection of Auditors
Preparation of Audit
Opening Meeting
Interviews, Document Review
Observations
Closing Meeting
Interim Report & Response
Audit Final Report
Action Plan/Remedial Work
Audit Summary Report
Follow Up
Planning of Audit
Performing of Audit
Reporting and Verification
Follow Up
45
o Ratification of relevant IMO instruments
o Unified interpretation of relevant legal instruments and code of conducts
o Compatibility of PSC procedures including reporting, coding, statistics,
notifications and right of appeal
o Transparency of information and co-operation and information exchange between
member states and MoU’s including PSC data
o Analysis of PSC activities, practices and statistics
o Training including revision of IMO training material
These latest developments are a major step into the direction of increasing the
effectiveness of port state control. This concludes the qualitative analysis of the present
safety regime including some of the newer developments in the areas concerning
maritime safety.
2.5. Summary of the Safety Regimes and Inspection Regimes
This chapter provided an overview of the overall complexity of the safety regimes. Many
players are part of the safety systems consisting of a mandatory (statutory) part and a
non mandatory part (industry driven). The mandatory part based on the legal framework
and normally enforced by the flag states is nowadays more and more performed by
recognized or authorized organizations (the classification societies) and as a last resource
by the port states. New developments in the area of labor related inspections are also
expected to fall within the area of classification societies.
Comparison of the port state control legal bases showed that there are differences in their
target factors to target vessels which might also reflect regional requirements. Port state
control is influenced by the world trade flows and their respective ship types that will
visit a particular regime. Differences in operational standards might be due to the
different experiences that the regimes have due to their number of years in existence and
the different backgrounds their inspectors have. However, all regimes do refer to the
applicable IMO guidelines for port state control inspections and identify “clear grounds”
and “non favorable treatment” for ships flying a flag which is not party to the conventions
in question. The USCG, AMSA, and the Paris MoU (through the EU) have stronger
enforcement possibilities and remedies for non-compliance than for instance the Indian
Ocean MoU or the Viña del Mar Agreement. Each of the segments of the maritime
industry (dry bulk, liquid bulk, containers etc.) have distinctive commercial and operating
surroundings which also influence the level of safety of each of these ship types and the
experience of the inspectors of the regimes.
The lack of trust in the industry between flag states, port states, classification societies,
insurance companies, and cargo owners has created a playground for many inspections
which are performed on certain ship types (oil tankers, chemical tankers and dry bulk
carriers) nowadays in the name of safety. The areas that are inspected in all of these
inspections show a considerable amount of overlapping between statutory and industry
driven inspections. In addition, the safety regimes do not accept port state control
inspections that are performed in another regime. This leaves certain ship types to be
exposed to a relatively large amount of inspections where the inspections are performed
sometimes during critical port operations and take time away from the crew. With
shortened time in ports, the inspections can increase the working hours of shipboard
personnel considerable. None of the inspections takes this into account or actually looks
closer into working and living conditions of the crew in particular the working and resting
46
hours. The new ILO convention is a welcoming development and will hopefully increase
compliance with the minimum working and living conditions onboard ships.
The lack of enforcement of the minimum international standards also shows the political
sensitivity of this topic overall and further underlines the lack of trust and cooperation
between the players and the various port state control regimes. The underlying question
is how the functioning of the safety regimes can be improved and how the money which is
allocated to port state control can be better used to eliminate substandard ships?
The estimated inspection costs of a port state control inspection is USD 747 per
inspection or a total of USD 34,3 million for all types of inspection. Inspections associated
with zero deficiencies and without administrative costs are estimated to be at USD 12,5
million per year or USD 50 million for the total four year period. Total inspection costs
per vessel per year are estimated to vary from USD 47,000 for tankers to USD 17,500 for
other ship types while the frequency of inspections can also vary considerably but is
estimated to be at 11 inspections per year for tankers, 6 for dry bulk carriers and 5 for all
other ship types. Comparing average insurance claim costs of vessels that have been
inspected with vessels that have not been inspected, one can clearly see that the average
insurance claim costs are higher for non inspected vessels and the difference between the
two categories is further highest for tankers.
One could argue that the inspections that are performed on ships with zero deficiencies
which is about 54% of the total inspection dataset and its associated costs (USD 12,5
million per year) could be used for training and to further created the necessary
framework to harmonize port state control activities by Helping emerging regimes where
more substandard ships are to be found. During the last FSI (14) in June 2006,
harmonization of port state control was considered and a working group established
which should create the necessary framework in order to achieve harmonization. This
aspect will further be analyzed in part III of this thesis.
The new IMO Voluntary Audit Scheme for flag states, port states and coastal states is the
first measure to address the interpretation and enforcement problem directly. It is a tool
which can be supported in many ways such as incorporating it into target factors for port
state control or any other inspections.
This chapter provided an overview of the individual players of the safety regime and how
the players interact which is important to understand in the chapter to follow in this
thesis. It further shows the overall complexity and legal framework of port state control
and any other inspections that are performed in the name of safety onboard vessels. The
link to the next chapter is the understanding of the industry and how it operates in order
to construct the variables which will be used in the chapters to follow.
47
Chapter 3: Datasets and Variable Preparation
This chapter explains the database preparations and transformation for port state control
data and casualty data. It further gives an explanation of the variables that are used and
how the variables are grouped together as well as provide a list of definitions used in the
casualty analysis. The chapter ends with a detailed analysis on the selection of ship
types.
3.1. Port State Control Dataset and Casualty Datasets
All existing memoranda of understanding who are in control of an inspection database
were asked to provide raw data for the analysis but not all regimes decided to cooperate
with this study despite the fact that the data is most of the time publicly available on
their home pages. Data from five regimes was received for the analysis and are shown in
Figure 17 where the data from one regime, the Indian Ocean MoU was received in two
different datasets and kept separate. The combined dataset therefore combines six data
sources.
Figure 17: Port State Control Dataset Preparation
Note: Indian Ocean MoU and Australia are treated separately
Since the Australian inspections accounted for a large part of the Indian Ocean MoU
dataset (54%) and around 60% of this dataset are inspections performed on bulk carriers,
17,455
inspections
From 01/99
to 12/04
7,349
inspections
From 01/02
to 12/04
47,108
inspections
From 01/01
to 12/04
708
inspections
From 01/03
to 07/05
89,936
inspections
From 05/00
to 12/04
Data
Merging
Recoding and Re-grouping of Data
Paris
MoU
Viña del Mar
Agreement
Data received on all inspections performed via a certain time frame including inspections
with zero deficiencies. (Total of 183,819 observations)
21,263
inspections
From 01/00
to 12/04
Particulars:
IMO Number
Ship Name
Gross Tonnage
Age (or Year Built)
Ship Type
Inspection Info:
Date of Inspection
Detained (y/n)
Port State or Port
Classification Society
Flag
Ownership (Residence)
Deficiencies
LR Fairplay Data for missing ship
particulars, ownership variable, and
construction information
Ship Types
Flag States
Classification Society
Ownership Countries
Port States (Ports for AMSA and USCG)
Construction Information
USCG
AMSA
Australia
Ind. Ocean
w/o Australia
Caribbean
MoU
48
the Indian Ocean MoU dataset was split between two datasets to decrease the bias of the
dataset versus the Australian data. The data received contains all inspection data within
a certain time frame (from January 1999 to December 2004). The US Coast Guard Marine
Safety Management System (MSMS)63 was changed in 2001 and data from the old system
could not be used for the analysis and therefore starts with January 2001. The Caribbean
MoU did not have any data available prior to 2003 and to increase the amount of records,
data from 2005 was added to the analysis.
Unfortunately, data from the Tokyo MoU could not be obtained and therefore a certain
percentage of vessels which are only inspected in this particular region are not covered by
the PSC dataset of this thesis. This might affect the results of part III of this thesis which
measures the effects of inspections on the probability of casualty and will be addressed in
Chapter 6 and 7 respectively.
The port state control data received was merged with data from LR Fairplay to add any
missing variables for ship particulars (age and size) and for the country of ownership of
the vessel. For the country of ownership of the vessel, a special data merge was performed
to add the variable indicating the Document of Compliance Company (DoC) Company. In
addition, information about the construction details was added to the inspection dataset.
For the casualty datasets and corresponding variables that were used for the casualty
analysis and the merge between port state control data and casualties, three data sources
were used and aggregated into one dataset. Figure 18 shows the process on how this
dataset was aggregated.
Figure 18: Casualty Dataset Preparation
Aggregation started with a match of data from LR Fairplay and LMIU to combine both
datasets. The two datasets combined account for 9,213 records for the time period 2000 to
2004. This dataset was then merged against data from the IMO to add casualties prior to
2000 and any other casualties not included in the combined LR/LMIU dataset. The
63 US Coast Guard Marine Safety Management System (MSMS) combines MSIS (Marine Safety
Information System) which was replaced in 2001 by MISLE (Marine Information for Safety & Law
Enforcement System)
LR Fairplay Merged with
2000-2004
LMIU
2000-2004
IMO
1993 – 2004
LR Fairplay Data:
Scrapping Information (if missing)
Ownership (DoC) Company
Missing Ship Particulars
Ship Types
Flag, Classification Societies
Ownership Country
Location of Casualty
Voyage Details
Type of Casualty
Construction Information
Recoding and Re-grouping of Data
49
additional casualty data without the corresponding port state control data (prior to 1st
Jan 2000) should help with the overall casualty analysis of the data. The total dataset
accounts for around 11,701 casualties. The combined casualty dataset was then merged
with data from LR Fairplay to add information about the scrapping of the data (if not
provided by the casualty dataset already), any other missing ship particulars and the
ownership of a vessel (DoC Company). The next section will explain the variable
transformation and grouping in detail.
3.2. Variable Transformations and Definitions
3.2.1. Basic Port State and Casualty Variables
Variable transformation and regrouping was performed for port state control data and
casualty data. Transformation tables were used to re-code all of the following variables:
1) Flag States (Black, Grey, White, Undefined) – as per Paris MoU since the overall
inspection dataset comprises of 49% data from the Paris MoU region
2) Classification Societies – IACS and Not IACS recognized
3) Ownership of a vessel as per Alderton & Winchester64 or technical management as
per LR Fairplay (DoC Company)
4) Ship Types
Variables were recoded using a transformation table for each MoU and the casualty
datasets into standard codes for each variable group (flag, class, owner, ship type). The
standard coding used for the total datasets were then transferred into dummy variables
for the regressions or regrouped into other groups for the descriptive statistics. Table 10
below gives a list of all variable groups that are used in both types of regressions.
Table 10: Summary of Variable Groups
Variable Group
Total
Number
in Group
Data
Type
Dependent 1: Detention 1 Binary
Dependent 2: Casualty (very serious, serious, less serious) 1 Nominal
Ship Particulars (e.g. age, tonnage, double hull, changes in flag,
class, ownership, vetting inspections) 15 Nominal
Ship Types 7 Nominal
PSC Regimes 6 Nominal
Detention Indicators 6 Sums
Deficiencies (main codes) & multiplicative dummies with ST 29 (156) Sums
Classification Societies 42 Nominal
Flag States 130 Nominal
Ownership Countries 6 Nominal
Port States/Ports 130 Binary
Ship Yard Country 37 Binary
Other Variables 3 n/a
Total for casualty models (multiplicative dummies) 413 (540)
Total for detention models (multiplicative dummies)*) 1575
64 Alderton T. and Winchester N (2002). “Flag States and Safety: 1997-1999”. Maritime Policy and
Management, Vol 29, No. 2, pp 151-162
50
A detailed list of variables with the respective coding65 for the detention and casualty
regressions is given in Appendix 4: Variable List and Respective Coding for Regressions
for further detail. The actual number of the variables that is then used in the respective
regressions depends on the actual dataset or portion of dataset and can therefore vary
accordingly. For the detention models, all variables have been multiplied by ship type and
regimes where the vessel was inspected.
Flag States
Flag States were coded individually or grouped into four major groups according to the
Paris MoU Black, Grey and White List66 where white listed flag states are performing
well followed by grey. Black listed flag states are performing worst. Flag states in the
group “undefined” are flag states that do not have enough inspections for the Paris MoU
or do not trade in the Paris MoU area.
Classification Societies (RO)
Classification Societies have been coded individually or grouped into two groups – either
they are a member of the International Association of Classification Society or not which
serves as a kind of quality indicator. There are currently ten members as follows:67
1) American Bureau of Shipping,
2) Bureau Veritas,
3) China Classification Society,
4) Det Norske Veritas,
5) Germanischer Lloyd,
6) Korean Register of Shipping,
7) Lloyd’s Register,
8) Nippon Kaiji Kyokai (ClassNK),
9) Registro Italiano Navale,
10) Russian Maritime Register of Shipping.
Ownership or Technical Management
Ownership is represented by two variables. It is either the “true owner” (not the
registered one) who has the financial benefit or it is the technical manager on the ISM
Document of Compliance68 The datasets were merged with data from Lloyds Register
Fairplay in order to identify the ownership of a certain vessel for both variables. For the
true ownership, the country of location was then grouped according to Alderton and
Winchester (1999)69 to reflect the safety culture of a certain owner and operator shore side
which is also expected to be reflected onboard. The grouping of the countries into six main
groups is found in Appendix 5: Grouping of Countries of Ownership for further reference
but is as follows:
• traditional maritime nations,
• emerging maritime nations,
• new open registries,
65 similar coding was used for the casualty and detention regressions for importing the data into
Eviews, the software that was used to perform the regressions
66 Paris Memorandum of Understanding Annual Reports for 2000 – 2004.
67 As per IACS, https://monkessays.com/write-my-essay/iacs.org.uk
68 The Document of Compliance is a requirement by the ISM (International Safety Management
Code) Code. The technical manager responsible for the safety management of the vessel needs to
be identified on this document. Sometimes for smaller companies, this can be the owner; otherwise
it is contracted out to manager who runs the vessel on behalf of the owner.
69 Alderton T. and Winchester N (2002). “Flag States and Safety: 1997-1999”. Maritime Policy and
Management, Vol 29, No. 2, pp 151-162
51
• old open registries,
• international open registries,
• “unknown” for unknown or missing entries.
The ISM Code provides a definition for the term “Company” which is denoted DoC
Company (Document of Compliance Company) as follows70: “Company means the owner of
the ship or any other organization or person such as the manager, or the bareboat
charterer, who has assumed the responsibility for operation of the ship from the owner of
the ship and who on assuming such responsibility has agreed to take over all the duties
and responsibilities imposed by the International Safety Management Code”.
3.2.2. List of Definitions Used in Casualty Analysis
For the sake of this thesis, it is important to define the various terms that are used in
conjunction with casualty analysis such as incidents, accidents, casualties, near misses
etc. The understanding of these terms and their use is important and is therefore
explained here below.
Incident, Accident and Casualty71
In insurance terms, events that are not deliberately caused by the insured and that are
not inevitable are an incident. The difference between the two is that an accident is
associated with an injury and a casualty normally entails a fatality.
Near Miss/Near Accident
A near miss or near accidents are incident that could have lead to an accident or casualty
but did not. It is therefore still important to analyze such events in order to understand
what went wrong and how it can be prevented in the future.
Total Loss and Constructive Total Loss72
A total loss of the vessel is when the vessel is lost either by it being completely destroyed
or submerged. A constructive total loss is equal to a total loss but in this case, it is
declared a constructive total loss if the repair of the vessel would be more expensive than
the remaining value of the vessel. These are cases of vessels which have experienced a
substantial amount of damage and are not worth repairing.
For the sake of the thesis, the term “casualty” is used to cover incidents, accidents and
casualties. In addition, the seriousness of the casualties is divided as per the IMO
definition based on the MSC Circular 953 of 14th December 2000 which is as follows:
IMO: MSC Circular 953, 14 December 200073
1. Very serious casualties: casualties to ships which involve total loss of the ship, loss
of life or severe pollution, the definition of which, as agreed by the Marine
Environment Protection Committee at its thirty-sevenths session (MEPC 37/22.
paragraph 5.8), is as follows:
“severe pollution” is a case of pollution which, as evaluated by the coastal
State(s) affected or the flag State, as appropriate, produces a major
70 SOLAS, Chapter IX, Regulation 1, page 425
71 Legal terms: http://en.wikipedia.org/wiki/List_of_legal_terms
72 Lloyd’s Maritime Intelligence Unit, definitions received with the casualty data
73 Reports on Marine Casualties and Incidents, Revised harmonized reporting procedures,
MSC/Circ. 953, MEPC/Circ.372, 14 December 2000
52
deleterious effect upon the environment, or which would have produced
such an effect without preventive action.
2. Serious casualties are casualties to ships which do not qualify as “very serious
casualties” and which involve fire, explosion, collision, grounding, contact, heavy
weather damage, ice damage, hull cracking, or suspected hull defect, etc. resulting
in:
• immobilization of main engines, extensive accommodation damage, severe
structural damage, such as penetration of the hull under water, etc.
rendering the ship unfit to proceed, or
• pollution (regardless of quantity); and/or
• breakdown necessitation towage or shore Helpance.
3. Less serious casualties are casualties to ships which do not qualify as “very serious
casualties” or “serious casualties” and for the purpose of recording useful
information also include “marine incidents” which themselves include “hazardous
incidents” and “near misses”.
3.2.3. The Selection of PSC Relevant Casualties
The preparation of the casualty dataset is very important for the validity of the analysis
and considerable care has therefore been placed on it. The cases were also re-classified
with respect to the seriousness of a casualty as defined previously and their casualty first
event as good as it could be identified by the fragmented data on hand. By combining
three different sources, the most comprehensive dataset of casualties and data available
for the purpose of this analysis was created.
From the total casualty dataset of 11,701 cases or 9,598 ships for the time period 1993 to
2004, more than half have been eliminated due to the incorrect time frame (1999-2004) or
for other reasons which will be explained in the beginning of Chapter 5. The remaining
cases are then matched against the time frame of PSC data (1999-2004) and aggregate to
6005 ships which provide the bases for the regressions in Chapter 6 and 7. Some of the
descriptive statistics of the casualties are based on the total casualty dataset depending
on the type and theme of the graph but is always indicated below the graphs. The next
section will give an overview of the selection of ship types since the selection of ship types
for port state control and casualties is equally important.
3.3. The Selection of Ship Types
As explained in Chapter 2, the selection of ship types for the analyses is important and
therefore considerable amount of time was spent to find the best possible grouping. This
provides a more accurate analysis of the probability of detention and casualty. Taking the
decision points listed under section 2.1.6. The Importance of Ship Types and Trade Flows
into account, the following ship types have been re-grouped out of the 19 original ship
types:
1. General Cargo & Multipurpose (General Cargo, Ro-Ro Cargo, Reefer Cargo,
Heavy Load).
2. Dry Bulk.
3. Container.
4. Tanker (Tanker, Oil Tanker, Chemical Tankers, Gas Carriers, OBO).
5. Passenger Vessel (Passenger Ships, Ro-Ro Passenger, HS Passenger).
6. Other (Offshore, Special Purpose, Factory Ship, Mobile Offshore, Other Ship
Types).
53
The trade flows with respect to ship types is shown in Chapter 2 while the results of the
Correspondence analysis is shown in this section to underline the decision of the grouping
of the ship types.
Quantitative Analysis for the Selection of Ship Types
Correspondence analysis can deal with large contingency (frequency) tables and plot
distances in a two-dimensional space where the distance between the variables in
question represents the association between them. Based on the contingency table, the
relative frequencies (or conditional proportions) and the marginal proportions (critical
masses) are calculated for the rows and columns and in this way, the row profiles can be
created. Each row profile can be represented as a point in space. The average profile is
the weighted average and is also the point of origin. The further a point is away from this
point of origin (also called centroid), the more different it is from the average profile74.
The main reason for this analysis is to further support the groupings that were performed
on the ship types. The variables which were used in the plot are based on all the
inspection data (approx. 183,000 inspections) and all the plots have been produced using
a program written by Michel van de Velden75. The variables are as follows and the result
can be seen in Figure 19:
1. Columns: Number of deficiencies (codes C0100 to C2500)
2. Rows: Ship Types
Figure 19: Ship Types and Deficiencies
Note: The circles were made manually by the author and are not generated by the program, only the
profiles for the ship types with respect to the deficiencies are shown here
74 For more detailed information on Corrspondence Analysis, refer to Clausen, S.E. (1998). Applied
Correspondence Analysis – An Introduction (Sage University Papser Series, Nr. 07-121)
75 Michel van de Velden, Econometric Institute, Rotterdam, 2004
54
The numerical results can be seen in Appendix 6 for further reference. Total inertia
explained is 74% in the first two dimensions. The plot shows that certain ship types group
together based on their violation against the deficiency codes. Container ships are kept
separate due to the commercial setup of the market segment. OBO’s are between oil
tankers and dry bulk carriers. Since they are closer to the tanker group, they have been
aggregated with this group. Gas carriers are difficult to allocate and since there are not
enough observations to create a separate group for gas carriers, the best possible group
which is also underlined by the plot is to add them with the tankers.
It can therefore be concluded that the division of ship types has been conducted in the
best possible way. This section concludes Chapter 3 of the thesis which gave an overview
of the datasets and the variables including the ship types that are used for the remaining
chapters of the thesis. It provides the foundation for the next chapter which will use the
variables explained here for the regressions to follow.
55
PART II
This part of the thesis is the first quantitative part and looks at the differences of port
state control across regimes. It tries to find the answer to the following research question:
1. What is the probability of detention across regimes and is there room for
improvement?
It produces the probability of detention across regimes to see if there is room for
harmonization. The probability of detention is also used in part III of the thesis where
port state control data is linked to the casualty data.
56
57
Chapter 4: The Global View on Port State Control
The first part of this chapter will provide some key descriptive statistics based on the port
state control data followed by regression analysis which produces the probability of
detention. The methods used in this chapter will take into account the different data
sources of the port state control datasets.
4.1. Key Descriptive Statistics for PSC
4.1. Key Figures on Registered Vessels
The first set of graphs give an overview of the world fleet based on number of vessels
registered and vessel size. Figure 20 shows the composition of the world fleet by number
of vessels and Figure 22 gives an insight into the size composition of the fleet by gross
tonnage.
Figure 20: Composition of World Fleet (by number of vessels)
General
20%
Fishing
24%
Other
11%
Tugs
12% Passenger
Ferry 2%
5%
Tanker
13%
Dry Bulk
7%
Container
6%
As per January 2005, Data Source: Lloyd’s Register Fairplay
By comparing the two graphs, one can easily see that most fishing vessels (74%) and tugs
(79%) are below 400 gt while ferries also show a high amount of vessels below 400 gt
(54%). Those ships fall out of the port state control eligibility since they do not have to
comply with SOLAS and MARPOL and fall under a different legal framework. The
commercial fleet is composed of container vessels, dry bulk, general cargo, tankers and
passenger vessels and a portion of the other ship types where smaller vessels can only be
found with general cargo ships (22% below 400 gt).
The actual split up of the commercial fleet which is eligible for inspection versus the total
registered vessels can be seen in Figure 21. Most of these vessels are above 400 gt as seen
in Figure 22. Oil tankers do have to comply with Annex I of the MARPOL convention if
the vessels are above 150 gt otherwise the cut of is 400 gt. Most ships in service by
58
number are general cargo ships (33%) followed by tankers (25%), dry bulk (14%) and
containers (12%).
Figure 21: Ships Eligible for Inspection
Container, 12%
Dry Bulk, 14%
General, 33%
Tanker, 25%
Passenger, 3%
Other, 13%
Commercial
above 400 gt
47%
Not
Commerical
53%
As per January 2005, Data Source: Lloyd’s Register Fairplay
Figure 22: Split up of Ship Sizes (gt) – World Fleet
22.2%
11.9%
54.7%
79.6%
40.3%
74.0%
99.7% 98.7%
77.8%
88.1%
45.3%
20.4%
59.7%
26.0%
12.0%
88.0%
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Container
Dry Bulk
General
Tanker
Passenger
Ferry
Tugs
Other
Fishing
below 400 gt above 400 gt
As per January 2005, Data Source: Lloyd’s Register Fairplay
4.2. Key Figures on Combined PSC Inspections
Table 11 provides a summary of each of the datasets from the various regimes and Figure
23 and Figure 24 provides the visualization of the key figures of the total dataset. The
data is based on all inspections which were conducted during the time frames including
information on inspections with zero deficiencies.
59
Table 11: Inspection Data Summary per MoU
Descriptive
Statistics
Total
Dataset
Paris MoU
Caribbean
MoU
Viña del
Mar
Agreement
Indian
Ocean
MoU
US Coast
Guard
AMSA
From
To 05/00
12/04
01/03
07/05
01/00
12/04
01/02
12/04
01/01
12/04
01/99
12/04
Total Inspections 183,819 89,936 708 21,263 7,349 47,108 17,455
Detentions 10,008 7,005 36 644 732 660 931
Detention Rate 5.44% 7.79% 5.08% 3.03% 9.96% 1.40% 5.33%
Total Deficiencies 471,764 312,305 760 46,977 19,085 42,452 50,185
Mean # of Def. 2.6 3.5 1.1 2.2 2.6 0.9 2.9
Mean Age – yrs 17 17 18 15 18 13 11
Mean Size – gt 22,079 15,327 11,112 22,105 18,215 28,948 36,767
Insp. with zero def 98,953 39,333 597 13,359 3,943 34,560 7,161
% of insp. zero def 53.8% 43.7% 84.3% 62.8% 53.7% 73.4% 41.0%
Note: compiled by author based on total PSC dataset
Out of the total 183,819 inspections, 53.8% are without deficiencies and 5.4% ended in a
detention of the vessel while aggregated by ship, the 53.8% decreases to 16.3% and
detention increases from 5.44% to 24.6% of all inspected vessels. From the total amount of
inspections of ships with deficiencies, 68% had 1 to 5 deficiencies while around 6% showed
more than 16 deficiencies. One can see that the key figures presented in Table 11 vary
accordingly such as the detention rate, the mean number of deficiencies per inspection
and the amount of inspections with zero deficiencies.
Figure 23: Key Figures – Total PSC Dataset
> 16 def., 6%
6 to 15 def.
26%
1 to 5 def.
Inspections 68%
with
deficiencies
46%
Inspections
without
deficiencies
54%
Detentions: 5%
Note: compiled by author based on total PSC dataset
This does not necessarily mean that one regime performs worse than the other. Each of
these datasets is the product of different legal bases and target factors described
previously and the trade flows which influences the ship types. The regression analysis
performed in the next chapter will highlight the differences and look into areas of possible
harmonization across the regimes. A detailed split up of the amount of deficiencies per
60
MoU based on the total inspections of each dataset and of the total dataset is shown in
Figure 24.
Figure 24: Amount of Deficiencies per Inspection per MoU
43.7%
84.3%
62.8%
53.7%
73.4%
41.0%
53.8%
35.4%
9.6%
24.8%
30.1%
22.7%
41.6%
31.2%
16.6%
4.2%
9.7% 13.6% 15.5% 12.1%
4.3% 1.8% 2.6% 2.6% 0.6% 1.9% 2.8%
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Paris MoU Carib
MoU
Vina MoU Indian O.
MoU
USCG AMSA Total
Zero Def 1 to 5 Def 6 to 15 Def Above 16 Def
Note: compiled by author based on total PSC dataset
The inspection frequency of the total dataset per ship for the total time frame where the
average amount of inspection frequency lies at around 7 inspections within 4 years or 1.7
inspections per year while the detention frequency of the total dataset lies at around 1.5
detentions per vessel for the same time period as shown previously in Chapter 2 of this
thesis.
Figure 25: Ship Types Inspected per MoU
53%
44%
31%
44%
15%
17%
20%
35%
24%
26%
61%
16%
23%
27%
13%
7%
8%
13%
6%
14%
8%
4%
2%
13% 5%
16%
15%
3%
11%
3%
2%
14%
2%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Paris MoU
Carib MoU
Vina MoU
Indian O.MoU
USCG
AMSA
General Cargo (1) Dry Bulk Carrier (2) Tankers (3)
Container (4) Passenger (5) Other (6)
1 2 3 4 5 6
Note: compiled by author based on total PSC dataset
61
This clearly shows that the same ships are inspected more than once within each regime
and across the regimes but only 66% of the vessels are detained once, 21% are detained
twice and 13% are detained three or more times within the time period (1999 – 2004).
Most ships inspected are general cargo & multipurpose ships and dry bulk carriers
followed by tankers and container ships. The USCG and AMSA show a lower amount of
general cargo ships but a higher amount of dry bulk carriers for AMSA and tankers for
the USCG. Detention varies per ship type and regime as can be seen in Figure 26 below.
Figure 26: Ship Types and Detention per MoU
10%
8%
5%
12%
6%
8%
1%
2%
7%
2%
5%
5%
9%
1%
4%
3%
7%
2%
6%
1%
7%
6%
1%
6%
3%
7%
19%
2%
4%
2%
2%
3%
2%
5%
2%
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Paris MoU
Carib MoU
Vina MoU
Indian O.MoU
USCG
AMSA
General Cargo (1) Dry Bulk Carrier (2) Tanker (3)
Container (4) Passenger (5) Other (6)
1 2 3 4 5 6
Note: Detention percentages are individually based on total inspections for each ship type and
therefore do not add up to a 100%
The next section will look at the key figures for classification societies which have been
classified into IACS and not IACS recognized classification societies and is shown in
Table 12 and visualized in Figure 27. Most ships inspected are classified by IACS
recognized class in each regime (some 79 to 85%) while detention rate is higher for non
IACS recognized class across all regimes. The same applies to the amount of mean
deficiencies per inspection where the amount of mean deficiencies for ships classified with
non IACS class is more than double to IACS class which can easily be seen by the two
lines in Figure 27.
62
Table 12: Key Figures on Classification Societies – Total Dataset
IACS Not IACS
Total Inspections
Detentions
% Detained
% of Total MoU
Mean
Deficiencies
Total Inspections
Detentions
% Detained
% of Total MoU
Mean
Deficiencies
Paris MoU 77272 4688 6.07% 85.9% 3.0 12664 2317 18.30% 14.08% 6.1
Carib. MoU 545 15 2.75% 77.0% 0.6 163 21 12.88% 23.02% 2.8
Viña MoU 19029 484 2.54% 89.5% 2.0 2234 160 7.16% 10.51% 4.4
Ind. O. MoU 6530 491 7.52% 88.9% 2.2 819 241 29.43% 11.14% 5.8
USCG 44210 539 1.22% 93.8% 0.8 2898 121 4.18% 6.15% 2.4
AMSA 16954 883 5.21% 97.1% 2.8 501 48 9.58% 2.87% 5.4
Total 164540 7100 19279 2908
Note: compiled by author based on total PSC dataset
Figure 27: Detention and Mean Deficiencies of Classification Societies
2.8
0.8
2.2
2.0
0.6
3.0
5.4
2.4
5.8
4.4
2.8
6.1
0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
30.00%
35.00%
Paris MoU Carib MoU Vina MoU Indian
O.MoU
USCG AMSA
% of Ships Detained
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
Mean Deficiency per Inspection
IACS Not IACS Mean Def IACS Mean Def Not IACS
Note: compiled by author based on total PSC dataset
Table 13 gives and overview of the flag states which have been grouped into white, grey
and black flag states according to the Paris MoU76 “Black, Grey, White List” and
undefined flag states as explained previously. The table shows the percentage of black,
grey, white or undefined flag states which have been detained and their respective mean
deficiencies per inspection.
76 Paris MoU Black, Grey and White List for the years 2000 to 2004
63
Table 13: Key Figures on Flag States – Total Dataset
FS_Black
% Detained
Mean
Deficiencies
FS_Grey
% Detained
Mean
Deficiencies
FS_White
% Detained
Mean
Deficiencies
FS_Undef
% Detained
Mean
Deficiencies
Paris 36595 68.8% 5.1 9244 8.6% 3.0 43980 22.4% 2.2 117 0.2% 4.4
Carib. 378 80.6% 1.7 20 0.0% 0.3 229 16.7% 0.4 35 2.8% 0.4
Viña 9444 69.1% 3.0 1361 7.6% 2.6 9859 17.7% 1.3 599 5.6% 4.0
Indian 3257 58.7% 3.1 1600 13.7% 2.3 2186 13.3% 1.4 306 14.3% 7.3
USCG 18241 58.2% 1.2 3158 6.1% 1.0 24695 33.5% 0.7 1014 2.3% 1.4
AMSA 7230 45.5% 3.1 1993 14.8% 3.8 7998 36.7% 2.4 234 2.9% 5.8
Total 75145 17376 88947 2305
Note: compiled by author based on total PSC dataset
The table is visualized in Figure 28 for the percentage of detention and in Figure 29 for
the mean deficiencies. Both figures give an interesting overview on the split up of flag
states. Most ships detained are black listed flag states while the USCG and AMSA also
show a higher amount of detention with white listed flag states.
Figure 28: Percentage of Detention per Flag State and MoU
69%
81%
69%
58%
46%
9% 8%
14%
6%
15%
17% 18%
13% 33% 37%
59%
22%
0.2% 2.8% 5.6% 14.3% 2.3% 2.9%
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Paris MoU Carib MoU Vina MoU Indian
O.MoU
USCG AMSA
% of Ships Detained per Flag State
Black Grey White Undefined
Note: compiled by author based on total PSC dataset
The amount of mean deficiencies varies between each MoU and is highest for black listed
flag states and undefined flag states with the exception of AMSA and the USCG. It is
almost double compared to the mean deficiencies of white listed flag states. For the
Indian Ocean MoU, one can see a high percentage of undefined flag states that trade in
the Indian Ocean MoU area but not in the Paris MoU area where the mean amount of
deficiencies (7.3) and detention rate (14.3%) is significantly higher with the rest of the
flag states.
64
Figure 29: Mean Deficiencies per Flag State and MoU
5.1
1.7
3.0
1.2
3.1
3.0
0.3
2.6
2.3
1.0
3.8
2.2
0.4
1.3
1.4
0.7
2.4
4.4
0.4
4.0
7.3
1.4
5.8
3.1
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
Paris MoU Carib MoU Vina MoU Indian O.MoU USCG AMSA
Mean Number of Deficiencies
Black Grey White Undefined
Note: compiled by author based on total PSC dataset
Looking at the dataset with reference to the ship owner, one can see from Figure 30 that
half of the vessels inspected were owned by traditional maritime nations followed by
emerging maritime nations and countries from open registries.
Figure 30: Ownership of Inspected Vessels
67%
59%
64%
48%
66%
56%
20%
16%
17%
28%
22%
21%
4%
8%
4%
3%
3%
5%
5%
3%
7%
7%
7%
18%
12%
8%
13%
4%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Paris MoU
Caribbean MoU
Vina del Mar
Indian O. MoU
USCG
AMSA
Traditional Maritime Nation (1) Emerging Maritime Nations (2)
Old Open Registry (3) New Open Registry (4)
Intern. Open Registry (5) Other/Unknown (6)
1 2 3 5 6
Note: compiled by author based on total PSC dataset
65
This split up does vary across the regimes which also underlines the trade flows of each of
these regimes and the corresponding ship profiles which serve the areas. The ship profiles
will be analyzed in the next section of this chapter. The Indian Ocean MoU shows a
higher percentage of owners from emerging maritime nations compared to the rest of the
regimes which can be explained by the fact that the area has more regional trade using
general cargo ships.
Table 14 looks into the “Document of Compliance Company” as per the ISM77 Code. A
separate merge was conducted78 in order to combine inspections with the corresponding
Document of Compliance Company and their respective country of location. Due to the
amount of missing data, this variable was not used as a separate variable in the
regressions to come and can only be seen as a general indication. A similar graph is
produced based on the probability of casualty in part III of this thesis. The results are not
presented with individual company names but by their respective country of location.
Table 14: Detention and DoC Country of Residence
Country of Residence Not
Detained
Detained
Total
Inspected
%
Detained
Bangladesh 32 15 47 31.9%
Cuba 60 20 80 25.0%
Papua New Guinea 23 7 30 23.3%
Algeria 16 4 20 20.0%
Romania 71 13 84 15.5%
Sri Lanka 20 3 23 13.0%
Channel Islands (British) 27 4 31 12.9%
Tunisia 23 3 26 11.5%
Syria 117 15 132 11.4%
Morocco 36 4 40 10.0%
Pakistan 27 3 30 10.0%
Brazil 129 14 143 9.8%
Egypt 37 4 41 9.8%
Lebanon 19 2 21 9.5%
Turkey 1021 99 1120 8.8%
Ukraine 384 37 421 8.8%
Portugal 55 5 60 8.3%
Argentina 69 6 75 8.0%
Latvia 181 15 196 7.7%
India 390 32 422 7.6%
Azerbaijan 233 18 251 7.2%
United Arab Emirates 294 22 316 7.0%
Bulgaria 289 20 309 6.5%
South Africa 30 2 32 6.3%
Unknown 48 3 51 5.9%
Fiji 69 4 73 5.5%
Russia 1985 115 2100 5.5%
Note: Based on total PSC dataset and more than 20 inspections, detention rate > 5%
The next chapter will produce the probability of detention and look at the differences
across the regimes.
77 International Safety Management Code
78 Merge conducted by Lloyd’s Register Fairplay
66
4.2. The Probability of Detention
4.2.1. Description of Model and Methodology
This model will provide the estimated probability (P) of a ship being detained based on
each ship type defined previously for each safety regime. The dependent variable (y) in
this case is “detained” or “not detained”. In a binary regression, a latent variable y* gets
mapped onto a binominal variable y which can be 1 (detained) or 0 (not detained). When
this latent variable exceeds a threshold, which is typically equal to 0, it gets mapped onto
1, other wise onto 0. The latent variable itself can be expressed as a standard linear
regression model
y*i = xiβ + εi
where i denotes ship i. The xi contains independent variables such as age, size, flag,
classification society or owner, and β represents a column vector of unknown parameters
(the coefficients). The binary regression model can be derived as follows, where the same
can apply to either the probability of detention or the probability of casualty (Chapter 6
and 7) later on in this thesis79, that is,
P (yi = 1|xi) = P (y*i > 0| xi) = P (xiβ + εi > 0|xi) = P (εi > – xiβ|xi) = P (εi ≤ xiβ|xi)
The last term is equal to the cumulative distribution function of εi evaluated in xiβ, or in
short:
P (yi = 1|xi) = F (xiβ)
This function F can take many forms and for this study two were considered, namely the
cumulative distribution function of the normal distribution (probit model) and the
cumulative distribution function of the logistic function (logit model). The general model
can therefore be written in the form of Equation 1 where the term xiβ changes according
to the model in question and is defined separately with each respective model of this
thesis.
Equation 1: Probability of Detention (either per regime or ship type)
x β)
x β)
i
i
P (
(
i 1 e
e
+
=
All probabilities for the models to follow are probabilities for individual ships. To estimate
the coefficients, quasi-maximum likelihood (QML)80 is used as method of estimation in
order to give some allowance for a possible misspecification of the assumed underlying
distribution function.
For the final models, logit and probit models are compared to see if there are any
significant differences and logit models are used for the visualization part. Since the
datasets originate from different sources, a test is performed to see whether the
79 for further reference, refer to Franses, P.H. and Paap, R. (2001). Quantitative Models in
Marketing Research. Cambridge University Press, Cambridge, Chapter 4
80 for further details on QML, refer to Greene H.W. (2000), Econometric Analysis, Fourth Edition,
page 823ff
67
coefficients obtained by the regressions differ significantly from each other across the
regimes. The analysis is therefore spilt up into four main steps which are visualized in
Figure 31 below for better understanding.
The amount of variables and observations used in the models change across the ship
types and safety regimes. In total, there are six datasets generating from five PSC
regimes and six ship types as shown in Table 15 which also shows the amount of total
observations for each ship dataset and the number of observations entered into the
combined ship models excluding the Caribbean MoU (708 observations). The Caribbean
MoU had to be excluded from the combined models due to the lack of sufficient data.
Figure 31: Visualization of Methodology for data preparation
Table 15: Summary of Datasets per MoU and Ship Type
Notation
Number of
Variables
Start/End.
Paris
MoU
r=1
Carib.
MoU
r=2
Viña
MoU
r=3
Ind.O.
MoU
r=4
USCG
r=5
AMSA
r=6
General 424 to 133 GC1 GC3 GC4 GC5 GC6
Dry Bulk 390 to 108 DB1 DB3 DB4 DB5 DB6
Container 245 to 72 CO1 CO3 CO4 CO5 CO6
Tanker 299 to 82 TA1 TA3 TA4 TA5 TA6
Passenger 93 to 38 PA1 PA3 PA4 PA5 PA6
Other ST 130 to 35 OT1
One Model
with all 708
observations
OT3 OT4 OT5 OT6
Total 1,581 to 468
# of Regressions Performed 6 1 4+1 4+1 5+1 4+1
Remark concerning the
regression models All ST
Only
one
model
No separate model
for passenger and
other ships types
No
separate
model
for PA
Same
as Ind.
Ocean
MoU
Note: GC = general cargo, DB = dry bulk, CO= container, TA = tanker, PA = passenger, OT = other
ship types
The number of variables used in the combined models is split up into the number of
variables entered in the model at the beginning and the number that was left in the final
models after reduction. The total number of variables for all combined models is 1,58181
and narrows down to 468 in the final models. The four steps shown in Figure 31 are
81 number of total multiplicative dummy variables
Step 1
Step 2
Step 3 & 4
Individual regressions per Ship Type and MoU (total
of 28 regressions) by using Equation 2
Part 1: Regression per ship type for general cargo,
dry bulk, containers and tankers using Equation 3
Part 2: Coefficient and significance testing
Part 1: Reducing Models developed under Step 2 by
imposing the restrictions that turned out to be valid
Part 2: Coefficient testing (second round) and
imposing restrictions that are found to be valid
Part 3: Visualization of results
68
explained shortly here and the equations used for the regressions are given in each
section respectively.
Step 1: Individual Regressions
A separate analysis is performed for each dataset listed in Table 15 which adds up to a
total of 28 regressions. The models can be written in the form of Equation 1 where the
term xiβ is given in Equation 2. Table 16 gives a detailed overview of the amount of
variables. The notation is as follows: i = individual ship, ℓ = variable groups, nℓ = total
number of variables within each group of ℓ and k = index from 1 to nℓ
Equation 2: Definition of term xiβ of Step 1 Model
Σ CODE Σ PS Σ OWN
ln(AGE ) ln(SIZE ) Σ CL Σ FS
7
1
6 1
1
1 5 1
4
1
3 1
1
0 1 2 1
5 6 7
3 4
,k k,i
n
,k k,i k
n
k,i k
n
k
,k k,i
n
,k k,i k
n
i i i k
β β β
x β β β β β β
−
=
−
= =
−
=
−
=
+ + +
= + + + +
Table 16: Binary Logistic Models: List of Total Variables Used per MoU
Paris MoU
Caribbean
MoU
Viña del Mar
Indian Ocean
MoU
USCG
AMSA
ℓ Total Number of Variables nℓ nℓ nℓ nℓ nℓ nℓ
Code Detained 1 1 1 1 1 1
AGE 1 Vessel Age C 1 1 1 1 1 1
SIZE 2 Vessel Size C 1 1 1 1 1 1
CL 3 Classification Societies D 29 10 26 19 22 15
FS 4 Flag States D 83 16 62 47 72 45
CODE 5 Deficiency main codes C 26 26 26 26 26 26
PS 6 Port States or Ports *) D 20 8 11 5 47 15
OWN 7 Ship Owner Countries D 6 6 6 6 6 6
Total for each MoU 166 68 133 105 175 109
C = continuous, D = dummy of categorical variables
*) for the USCG and AMSA, ports are used instead of port states
For the step 1 model, a separate regression was performed for each ship type and MoU
individually – a total of 28 regressions. For the Caribbean MoU, the dataset cannot be
split up according to the ship types due to the low number of observations but one
regression using the total dataset is performed including a dummy variable for each ship
type. The same method is also used for passenger vessels and other ship types with a
slightly modified version which will be explained under the step 2 models.
Step 2: Hypothesis and Coefficient Testing
For the step 2 model, the dependent variables except the port states were multiplied
(based on the outcome of the step 1 model) by ship type and PSC regime (r) to create
multiplicative dummy variables. The total dataset was then divided into six datasets (one
69
for each ship type) and a separate regression was performed on each ship type based on
Equation 3. The variables are listed in detail in Table 17 for further reference. In this
equation, the notation for individual ship i is dropped for sake of simplification The rest of
the notation is as follows: ℓ represents the variable groups, nℓ is total number of variables
within each group of ℓ (0-7), k is an index from 1 to nℓ , r represents a respective PSC
regime (1 to 5) and nr is the total number of PSC regimes (5).
Equation 3: Definition of term xβ of Step 2 Model
OWN
F CODE P
REG ln(AGE) ln(SIZE) CL
7,
1
1 1
6,
1
4, 1 1 5, 1 1
1
1 1
3,
1
1 0, 1 1, 1 2, 1 1
7
4 5 6
3
k,r k,r
n
k
n
r
k,r k,r
n
k
n
r
k,r k,r
n
k
n
r
k,r k,r
n
k
n
r
k,r k,r
n
k
n
r
r r
n
r
r r
n
r
r r
n
r
β
β S β β S
xβ β β β β
r
r r r
r r r r
−
= =
−
= = = =
−
= =
−
= = = = =
+ Σ Σ
+ Σ Σ + Σ Σ + Σ Σ
= Σ + Σ + Σ + Σ Σ
Table 17: Binary Logistic Models: List of Variables Used per ST – step 2 Models
All variables are multiplicative dummies
with the exception of the passenger ship
and other ship types
General
Cargo
Dry Bulk
Container
Tanker
Passenger
Other ST
ℓ Total Number of Variables nℓ nℓ nℓ nℓ nℓ nℓ
Code Detained 1 1 1 1 1 1
REG 0 PSC Regime D 5 5 5 5 5 5
AGE 1 Vessel Age C 5 5 5 5 1 1
SIZE 2 Vessel Size C 5 5 5 5 1 1
CL 3 Classification Societies D 73 61 36 41 15 16
FS 4 Flag States D 140 121 51 82 24 36
CODE 5 Deficiency main codes C 107 101 81 93 19 18
PS 6 Port States or Ports *) D 65 71 43 57 23 47
OWN 7 Ship Owner Countries D 23 20 18 10 4 5
Total for each ST 424 390 245 299 93 130
C = continuous, D = dummy of categorical variables
*) for the USCG and AMSA, ports are used instead of port states
As mentioned earlier, the model for the passenger ships and other ship types is not based
on multiplicative dummy variables due to lack of data. Those models follow the same type
of model of Equation 2 based on one total dataset for all passenger vessels or other ship
types respectively with the difference that no constant was used in the model but five
variables indicating the respective regimes as shown in Equation 3.
In order to see if the coefficients across the PSC regimes vary, the Wald-Test for Testing
Restrictions82 was performed on the results obtained from the models and based on the
following hypothesis on a subset of the matrix where ℓ represents the variable groups and
nr is the total number of PSC regimes (5).
82 For further detail on the Wald Test for a Subset of Coefficients, please refer to Greene H.W.,
Fourth Edition, Econometric Analysis, Fourth Edition, page 825.
70
Ho: coefficients within each variable group ℓ across the PSC regimes r do not vary
Ha: coefficients within each variable group ℓ across the PSC regimes r do vary
Ho: coefficients within each variable group ℓ across the PSC regimes r are not significant
Ha: coefficients within each variable group ℓ across the PSC regimes r are significant
Step 3 & 4: Reduction of Models and Visualization of Results
The models per ship type are reduced to the final models as explained in Figure 31 using
a significance level of 5% where the results can be seen in Table 21 for further reference.
After the final reduction of the model, the coefficients were tested again in a second round
applying the hypothesis developed under step 1 at a 5% significance level and restrictions
were imposed when found to be valid. The last step is to visualize the results obtained
under step 3 by calculating out the estimated probabilities using Equation 1.
3.3.2. Step 1 Results: Per MoU and Ship Type
Table 18 gives an overview of the classification tables of the individual regressions that
were performed on each dataset by using SPSS (statistical software). The results then
provide the basis for the creation of the dummy variables used in step 2.
Table 18: Step 1: Classification Tables
Ship Type
Hit Rates for
detained (%)
Paris
%
Carib
%
Viña
%
Indian
%
USCG
%
AMSA
%
General selected 81.4 90.9 85.3 83.5 93.3 80.8
unselected*) 79.2 57.1 84.9 75.6 69.8 65.8
Dry Bulk selected 81.3 90.9 85.3 90.5 88.9 81.9
unselected*) 79.1 57.1 89.1 81.3 66.1 76.2
Container selected 85.6 90.9 95.3 94.4 90.9 80.8
unselected*) 68.5 57.1 80.0 57.1 64.7 75.0
Tanker selected 82.3 90.9 91.2 90.7 87.0 81.0
unselected*) 81.8 57.1 79.2 84.1 66.7 65.4
Passenger selected 77.4 90.9 86.9 86.8 89.2 78.0
unselected*) 80.7 57.1 89.6 76.2 84.4 76.0
Other ST selected 85.6 90.9 86.9 86.8 84.4 78.0
unselected*) 80.0 57.1 89.6 76.2 68.3 76.0
*) unselected means out of sampling forecasting which is an option used in SPSS
The cut off rate used for each of the models is based on the detention rate which varies
accordingly per MoU and ship type and is listed in Table 19 for each ship type and MoU
and for each ship type as a total. The latter is used in step 2 to produce the classification
tables. One can see that the hit rate for detained vessels varies and that the Caribbean
Model due to its low number of observations shows less predictive accuracy with 57% hit
rate for out of sample forecasting. Container vessels also show lower hit rates for all
MoU’s compared to the other main ship types (general cargo, dry bulk and tankers) but in
general, the hit rates are found to be acceptable for the amount of data and variables.
Based on these outcomes, multiplicative dummy variables are computed for each variable
and ship type (e.g. ship type general cargo Paris MoU*Classification Society ABS) and the
datasets for the ship types of each MoU (e.g. all general cargo ships) are aggregated to
71
one dataset which ends of with 4 datasets (general cargo, dry bulk, container and tanker)
to be the basis for the next step.
Table 19: Cut Off Rates (based on observed detention rate) per ST and MoU
Cut Off Rate for Classification Table
Ship Types Total Paris Carib Viña Indian USCG AMSA
General Cargo 0.080 0.097 0.051*) 0.046 0.121 0.023 0.065
Dry Bulk 0.046 0.076 0.051*) 0.021 0.072 0.015 0.053
Container 0.020 0.029 0.051*) 0.019 0.056 0.009 0.066
Tanker 0.031 0.046 0.051*) 0.023 0.090 0.008 0.038
Passenger 0.034 0.057 0.051*) 0.03*) 0.099*) 0.014*) 0.053*)
Other Ship Types 0.037 0.064 0.051*) 0.03*) 0.099*) 0.020 0.053*)
*) regression based on total dataset
3.3.3. Step 2 Results: Coefficient Testing (Performed in 2 Rounds)
Based on Equation 3, the models are estimated and the coefficients are tested according
to the set of hypotheses explained earlier at a 5% significance level. The result can be
seen in Table 21 for detailed reference. The testing was performed in two rounds – first if
the coefficients vary significantly across the MoUs and second, if they are zero. One of the
most interesting findings in performing the testing is that the main differences across the
regimes are based on the port states and the individual deficiency codes and not
necessarily the flag states or classification societies. The next sections will impose the
restrictions that were found to be valid and will after reducing the models and performing
a second test round; visualize the main findings for the probability of detention across the
regimes.
3.3.4. Step 3 Results: Final Models per Ship Type
As a first step, the models were estimated without QML83 and with QML using
Huber/White standard errors and covariance at the time the program first found a
solution. The results were compared to identify significant differences in the coefficients
and the results can be seen in Table 20 which lists the variables at the time the matrix
first solved, the amount of variables which changed significance and the amount of
variables which are changed in the final models.
Table 20: Variables changed based on QML versus non QML estimation
Variables at the
time matrix
first solved
Total
Variables
#of Variables
changed
% Variables
changed
Final # of Variables
changed in reduced
Model
General Cargo 424 15 3.6% 9
Dry Bulk 390 35 9.0% 2
Container 245 18 7.4% 2
Tanker 299 25 8.4% 1
Passenger 93 4 4.3% 0
Other Ship Types 130 6 4.7% 0
83 Quasi Maximum Likelihood – Huber/White standard error & covariance
Table 21: Step 2: Results – Testing of Equality of Coefficients across the Regimes
General Cargo Dry Bulk Tanker Container
Table 21 – part 1 round 1 round 2 round 1 round 2 round 1 round 2 round 1 round 2
Code Variable
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
Age 5 0.6373 0.0000 5 0.0982 0.0000 5 0.0710 0.0000 5 0.8234 0.1127
Size 5 0.0088 0.0000 3 0.5480 5 0.1023 0.0091 5 0.3409 0.0575 5 0.4766 0.0014
ABS American Bureau of
Shipping 5 0.0772 0.1308 5 0.5606 0.4055 5 0.4414 0.5769 4 0.3808 0.5311
BV Bureau Veritas 5 0.1724 0.2491 5 0.0844 0.1323 4 0.5116 0.4663 4 0.2712 0.2489
CCS China Classification
Society 3 0.6979 0.4107 4 0.6253 0.5339
CRR Croatian Register of
Shipping 2 0.0821 0.1120
DNV Det Norske Veritas 5 0.0103 0.0398 2 0.0678 5 0.5740 0.5134 5 0.3149 0.0163 3 0.0215 0.0520
GL Germanischer Lloyd 5 0.0950 0.1329 4 0.7495 0.8373 4 0.0819 0.1147 5 0.1782 0.2770
HIN Honduras Inter. Naval
Survey IB 2 0.2920 0.5739
IBS Isthmus Bureau of
Shipping 2 0.3591 0.6488
INS Intern. Naval Surveys
Bureau 2 0.5745 0.3432 2 0.0817 0.1699
IRS International Register of
Shipping 3 0.1126 0.1850
KRS Korean Register of
Shipping (South) 4 0.0429 0.0187 4 0.4843 0.0814 2 0.7279 0.0648
LLR Lloyds Register of
Shipping (UK) 4 0.0225 0.0118 3 0.0166 5 0.1848 0.1138 5 0.3123 0.0513 4 0.0048 0.0086
NCL No Class Recorded 2 0.5890 0.6730 2 0.3809 0.2977 2 0.9436 0.7701
NKK Nippon Kaiji Kyokai
(Japan) 5 0.0602 0.0487 2 0.3993 5 0.4899 0.2805 5 0.2254 0.1098 4 0.1388 0.1688
PRS Polski Rejestr Statkow
(Poland) 4 0.0039 0.0095 4 0.2166 0.1854
RIN Registro Italiano Navale
(Italy) 4 0.1280 0.1920 3 0.8010 0.8510 4 0.4145 0.3450 2 0.8327 0.5954
RMS Russian Maritime
Register of Shipping 5 0.4846 0.4135 4 0.4306 0.4947 4 0.1455 0.2370
AG Antigua and Barbuda 4 0.4843 0.1403 2 0.0700 0.1910 4 0.5624 0.6223
AN Antilles Netherland 3 0.9356 0.1754
BO Bolivia 3 0.0054 0.0018
General Cargo Dry Bulk Tanker Container
Table 21 – part 2 round 1 round 2 round 1 round 2 round 1 round 2 round 1 round 2
Code Variable
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
BS Bahamas 5 0.1997 0.1331 5 0.8318 0.2093 4 0.2212 0.1177 4 0.6959 0.8277
BZ Belize 2 0.8025 0.5183
BR Brazil 3 0.0053 0.0015 2 0.0000
CN China 3 0.2580 0.4361 4 0.9370 0.8278
CY Cyprus 5 0.3959 0.0075 3 0.2039 5 0.4504 0.0045 5 0.1961 0.1088 5 0.5748 0.6927
DE Germany 2 0.9846 0.4926
DK Denmark 4 0.7012 0.7039 3 0.1124 0.0758
EG Egypt 2 0.3386 0.0721
ET Ethiopia 2 0.5550 0.8109
GE Georgia 2 0.2856 0.0056
GI Gibraltar 2 0.3590 0.0319
GR Greece 2 0.4302 0.5718 5 0.8824 0.2218 3 0.0255 0.0005 2 0.4256 2 0.5040 0.0844
HK Hong Kong 4 0.0185 0.0139 5 0.3508 0.1159 3 0.3795 0.5830
HR Croatia 3 0.0896 0.1644
IM Isle of Man 2 0.4037 0.6447 2 0.2903 0.0536
IN India 3 0.9131 0.0156 3 0.3649 0.2113
IR Iran 2 0.6848 0.0898 2 0.7926 0.0426
IT Italy 3 0.5387 0.0231 2 0.6204 0.6199
KH Cambodia 3 0.2656 0.0010
KP North Korea 2 0.2083 0.0071
KR South Korea 3 0.4980 0.0036
KY Cayman Islands 3 0.9684 0.1840 4 0.5588 0.0644 3 0.5814 0.5005
LR Liberia 5 0.3578 0.3847 5 0.4012 0.0426 5 0.0677 0.0787 2 0.6365 4 0.3077 0.2907
MH Marshall Islands 3 0.6486 0.7862 4 0.7396 0.1916 5 0.4354 0.5560
MT Malta 5 0.2081 0.0116 5 0.1607 0.0018 4 0.0059 0.0004 4 0.5280 3 0.8521 0.2151
MY Malaysia 3 0.6034 0.0262 3 0.4863 0.3045 2 0.6049 0.0711
NL Netherlands 4 0.7683 0.2810 2 0.1530 0.0105 2 0.6718 0.5635 3 0.2780 0.2876
NO Norway 3 0.5029 0.6256 3 0.3497 0.3450 4 0.0069 0.0137
PA Panama 5 0.0271 0.0000 4 0.0789 5 0.5310 0.0091 5 0.0002 0.0000 2 0.6303 5 0.2594 0.0906
PH Philippines 4 0.2720 0.0578 4 0.6348 0.0255
PL Poland 2 0.0758 0.0310
RU Russian Federation 2 0.8107 0.0001 2 0.1928 0.0577
General Cargo Dry Bulk Tanker Container
Table 21 – part 3 round 1 round 2 round 1 round 2 round 1 round 2 round 1 round 2
Code Variable
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
# of variables
Coefficient
Testing
Significance
# of variables
Test of Equality of
Coefficients
SE Sweden 3 0.0842 0.1747
SG Singapore 2 0.7601 0.4617 4 0.5192 0.0097 4 0.3096 0.0103 4 0.2952 0.4082
TH Thailand 3 0.7485 0.5057 3 0.1885 0.3382
TR Turkey 2 0.1148 0.0000 5 0.8337 0.0200
TW Taiwan 2 0.7650 0.0936
UA Ukraine 2 0.8116 0.0008 2 0.4498 0.0890
UK United Kingdom 2 0.4595 0.4395 2 0.9237 0.7673
VC/SV St. Vincent & the
Grenadines 5 0.0398 0.0000 2 0.0544 5 0.1528 0.0541 2 0.1568 0.0013 2 0.0005 0.0001
VU Vanuatu 4 0.2089 0.0715
100 Ship’s certificates and
documents 5 0.0000 0.0000 3 0.0010 5 0.0000 0.0000 4 0.0709 5 0.6376 0.0000 5 0.6486 0.0000
200 Crew certificates 5 0.0000 0.0000 5 0.0000 5 0.0000 0.0000 4 0.0000 5 0.0000 0.0000 4 0.0000 5 0.0000 0.0000 4 0.0045
300 Accommodation 5 0.0535 0.0024 5 0.0601 0.6440 2 0.7453 4 0.0261 0.0117 2 0.0212 3 0.0182 0.0101
400 Food and catering 4 0.1765 0.2878 4 0.1936 0.1469 3 0.9917 0.6672 2 0.2989 0.0357
500 Working spaces and
accident prevention 4 0.1480 0.0028 3 0.0056 3 0.3941 0.4846 2 0.4734 0.6762 2 0.8167 0.4619
600 Life saving appliances 5 0.0010 0.0000 3 0.0056 5 0.0000 0.0000 5 0.0000 5 0.0000 0.0000 5 0.4078 0.0000
700 Fire Safety measures 5 0.0040 0.0000 4 0.0012 5 0.0000 0.0000 5 0.0000 5 0.3463 0.0000 5 0.0000 0.0000 3 0.0513
800 Accident prevention
(ILO147) 5 0.4718 0.5230 5 0.0218 0.0420 3 0.2629 0.4037 2 0.3488 0.6226
900 Structural Safety 5 0.0000 0.0000 5 0.0000 5 0.0064 0.0000 4 0.6498 5 0.0423 0.0000 3 0.0640 5 0.0077 0.0000 4 0.0112
1000 Alarm signals 5 0.6793 0.0000 5 0.8309 0.8068 3 0.6520 0.0938
1100 Cargoes 5 0.0006 0.0000 3 0.0003 5 0.2135 0.1793 3 0.7232 0.1457 4 0.5996 0.4934
1200 Load lines 5 0.0000 0.0000 4 0.8410 5 0.1465 0.0000 2 0.2331 5 0.3015 0.0003 5 0.1293 0.0000
1300 Mooring arrangements
(ILO 147) 5 0.0490 0.0029 2 0.5029 5 0.5774 0.6491 5 0.0009 0.0000 2 0.0012 4 0.2829 0.2674
1400 Propulsion & aux. 5 0.1237 0.0000 3 0.1201 4 0.2512 0.0000 5 0.3776 0.0004 5 0.0905 0.0000 3 0.8884
1500 Safety of navigation 5 0.0001 0.0000 3 0.0666 5 0.0116 0.0004 2 0.2082 5 0.2830 0.0311 5 0.3467 0.4816
1600 Radio communications 5 0.0316 0.0000 4 0.0055 5 0.0082 0.0000 4 0.0004 5 0.0022 0.0017 3 0.0000 4 0.7991 0.0000
1700 MARPOL annex I (Oil) 5 0.0094 0.0000 4 0.2540 5 0.0125 0.0000 5 0.0167 5 0.0034 0.0000 3 0.0000 5 0.0134 0.0000 4 0.2317
1800 Gas and chemical
carriers 4 0.0052 0.0005 2 0.2751
2000 Operational deficiencies 5 0.0021 0.0035 3 0.0005 4 0.0000 0.0000 5 0.0174 0.0144 4 0.5767 0.4367
General Cargo Dry Bulk Tanker Container
Table 21 – part 4 round 1 round 2 round 1 round 2 round 1 round 2 round 1 round 2
Code Variable
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
# of variables
Test of Equality of
Coefficients
Significance
# of variables
Test of Equality of
Coefficients
2100 MARPOL related op.
def. 4 0.1970 0.2691 3 0.7468 0.8013 2 0.6432 0.6629 2 0.5142 0.0001
2200 MARPOL annex III
(Package)
2300 MARPOL annex V
(Garbage) 3 0.4652 0.0477 2 0.0544 0.0390 2 0.0118 0.0024 2 0.0015 2 0.1572 0.3281
2500 ISM related deficiencies 5 0.0000 0.0000 4 0.0000 5 0.0000 0.0000 4 0.0000 4 0.0000 0.0000 3 0.1034 4 0.0020 0.0000 3 0.0001
OOR Owner from Old Open
Registry Country 4 0.0158 0.0341 3 0.6703 0.7165 3 0.1912 0.3330 2 0.8593 0.0010
IOR Owner from Intern. Open
Registry Country 5 0.0796 0.0888 2 0.4555 4 0.1134 0.0954 3 0.0025 0.0042
TMN Owner from Traditional
Maritime Nation 5 0.0173 0.0151 2 0.0079 5 0.8201 0.4477 4 0.4494 0.2132 5 0.0192 0.0339
EMN Owner from Emerging
Maritime Nation 5 0.0325 0.0600 5 0.4915 0.4562 5 0.0215 0.0416
UNK Owner Unknown 4 0.0040 0.0009 2 0.0019 3 0.0053 0.0145 3 0.4341 0.4774 3 0.0146 0.0311
Note: the number of variables depicts the number of variables that were in the test in each round. The first round of testing was performed after the
program found a solution the first time and the second round of testing was performed after the model was reduced to only significant variables.
76
One can see that the significance of some of the variables changed especially for the dry
bulk model. To give a certain allowance for a possible misspecification of the assumption
of the underlying function, QML was used for the final models and both probit and logit
was estimated and the results are shown in Error! Not a valid bookmark selfreference..
The table lists the number of observations that were used in each model, outliers that
were identified and eliminated, the Mc Fadden84 R2 and the hit rates with the respective
cut off values used to produce the of the classification tables for each model, the Hosmer-
Lemeshow-Statistic (HL) and its p-value. The HL test is a goodness of fit test which
compares the expected values with the actual values by group. Its null hypothesis (Ho)
assumes little difference of the expected versus actual values and therefore a good fit of
the model to the data. The alternative hypothesis (Ha) represents not a good fit of the
model to the data.
Table 22: Summary of Key Statistics and Classification Table
General Dry Bulk Container
0 = 60893 0 = 45571 0 = 17785
# observations in final 1 = 5580 1 = 2206 1 = 426
model
Total= 66473 Total= 47777 Total= 18211
# outliers 132 184 6
Cut Off 0.0842 0.0462 0.0240
LOG PRO LOG PRO LOG PRO
Mc Fadden R2 0.433 0.438 0.411 0.419 0.448 0.459
% Hit R. y=0 87.59 86.39 87.55 86.84 90.49 90.12
% Hit R. y=1 82.26 83.33 84.18 85.58 85.92 87.32
% Hit R. Tot 87.14 86.12 87.39 86.78 90.38 90.05
HL-Stat. df=8 130.74 51.83 67.16 47.45 17.82 15.28
p-value 0.0000 0.0000 0.0000 0.0000 0.0226 0.0539
Tanker Passenger Other ST
0 = 32985 0 = 5907 0 = 9699
# observations in final 1 = 1060 1 = 211 1 = 374
model
Total= 34045 Total= 6118 Total= 10073
# outliers 82 12 4
Cut Off 0.0312 0.0345 0.0372
LOG PRO LOG PRO LOG PRO
Mc Fadden R2 0.424 0.435 0.332 0.427 0.388 0.399
% Hit R. y=0 88.81 88.39 84.54 86.58 88.20 87.74
% Hit R. y=1 86.60 87.26 86.73 90.45 83.69 86.36
% Hit R. Tot 88.74 88.36 84.62 86.70 88.04 87.69
HL-Stat. df=8 31.15 19.74 7.53 4.94 16.38 10.55
p-value 0.0001 0.0113 0.4803 0.7640 0.0372 0.2284
The Mc Fadden R2 and the hit rate are acceptable for the amount of observations used in
each model. Outliers were identified at each step and the model was reduced at a 5%
significance level where most variables are significant at a 1% level. The results of some
of the HL statistic indicate that there is not a good fit of the model which can be
84 The Mc Fadden R2 is not provided by the model automatically and was therefore computed separately. For
further details on this statistics, refer to Franses, P.H. and Paap, R. (2000). Quantitative Models in Marketing
Research.
77
explained due to the very large amount of observations. Regardless of this statistic, it is
found that the result is good enough to be used as the hit rates are very good. Not much
difference between logit and probit can be identified and the logit models are used for the
visualization of the results. The remaining statistics of the resulting final models can be
found in Appendix 7 to Appendix 10 while for the passenger ships and other ship types,
the statistics are shown in Appendix 11 and Appendix 12 for further reference.
3.3.5. Step 4: Visualization of Results
This section visualizes the findings in graphical form through the creation of ship profiles
and the grouping of the main deficiency codes into eight main deficiency groups shown in
Table 23. The grouping of the codes reflects the similarity of the deficiency codes by their
nature (e.g. operational deficiencies, management related deficiencies, crew related
deficiencies, etc.). The number in brackets next to the deficiency group represents the
number that is used in the graphs later on in this chapter to facilitate differentiation of
the graphs in black and white versus color.
Table 23: Grouping of Deficiency Codes for Visualization
Deficiency Main Group Description of Codes within the Main Group
Certificates (1) Ship’s certificates Code_0100
Crew certificates Code_0200
Safety & Fire Appliances (2) Life saving appliances Code_0600
Fire safety measures Code_0700
Alarm Signals Code_1000
Equipment/Machinery (3) Propulsion & Aux. Machinery Code_1400
Ship & Cargo Operations (4) Carriage of Cargo & Dang. Goods Code_1100
MARPOL I: SOPEP, Oil Record Book Code_1700
Oil, Chemical Tankers and Gas Carriers Code_1800
MARPOL II: P&A Manual, Cargo Rec.B. Code_1900
SOLAS related operational deficiencies Code_2000
MARPOL related operational deficiencies Code_2100
MARPOL III: Packaging, Documentation Code_2200
MARPOL V: Garbage Management Code_2300
Working & Living Accommodation Code_0300
Conditions (5) Food & Catering Code_0400
Working spaces, accident prevention Code_0500
Accident prevention Code_0800
Mooring Arrangements Code_1300
Stability/Structure (6) Stability/Structure/Equipment Code_0900
Load Lines Code_1200
Bulk Carriers, additional safety measures Code_2600
Navigation/Communication (7) Safety of Navigation Code_1500
Radio communications Code_1600
Management (8) ISM related deficiencies Code_2500
ISPS related deficiencies (not used) Code_2700
In visualizing the results, three approaches are used. First, each ship type is analyzed for
each MoU. Second, the difference in the contribution towards the probability of detention
is shown across the MoU’s and finally, an overall view is presented based on average
probabilities.
78
3.3.6 Individual Results per Ship Type
In order to visualize the results of the regressions, ship profiles are created and the
corresponding probability of detention is computed and shown in Figure 32 to Figure 37
for each ship type and MoU. Due to the amount of graphs, only one ship type per MoU is
shown here and the rest can be seen in Appendix 14 to Appendix 17.
Typical ship profiles for each ship type are created and the probability of detention is
calculated out based on the number of deficiencies for each deficiency category. The
steeper the curve of the graph, the higher the contribution of the deficiency group towards
the probability of detention. In essence, it reflects the ship profiles that trade in the area
as well as the emphasis that was placed on certain deficiencies during an inspection.
For the general cargo ship that can be seen in Figure 32 for the Indian Ocean MoU, 3
deficiencies in the area of certificates lead to a high probability of detention (0.9). The
deficiency groups related to safety and fire and to certificates show the highest
contribution towards detention followed by deficiencies related to navigation and
communications, stability and structure and ship and cargo operations.
Figure 32: Probability of Detention – General Cargo
General Cargo Ship – Indian Ocean MoU
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 13 yrs
Tonnage: 5965 gt
Flag: Panama
Class: GL
Port State: Sudan
Owner: Unknown
1 2
4
5
6
7
8
Overall, the graphs show the differences between the regimes and the ship types. For the
dry bulk carrier in Figure 33 for AMSA, the highest contribution can be found with ISM
related deficiencies (Management) followed by certificates and ship and cargo operations.
ISM related deficiencies reflect how the safety management system is implemented
onboard while the deficiency group ship and cargo operations reflect the actual execution
of the management system. The same applied for one of the most important deficiency
groups – safety and fire appliances.
79
Figure 33: Probability of Detention – Dry Bulk
Dry Bulk – AMSA
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 13 yrs
Tonnage: 38995 gt
Flag: Malta
Class: GL
Port State: Melbourne
Owner: EMN
1
2
3
4
5
6
7
8
Figure 34: Probability of Detention – Tankers
Tanker – Paris MoU
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 10 yrs
Tonnage: 28909 gt
Flag: Panama
Class: GL
Port State: Netherlands
Owner: TMN
1
2
3
4 5
6
7
8
Figure 34 shows the tanker for the Paris MoU region and Figure 35 shows the container
vessel for the USCG. For the first graph, the most important deficiency group is safety
and fire appliances followed by ISM related deficiencies (Management) and ship and
80
cargo operations. The picture is similar to the AMSA picture for dry bulk carriers. What
is interesting to notice is that the group living and working conditions also show a higher
contribution than with other ship types which is counter intuitive since tankers seem to
have a better ship profile to start with than for instance general cargo ships or dry bulk
carriers.
Figure 35: Probability of Detention – Container
Container – USCG
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 8 yrs
Tonnage: 27322 gt
Flag: Panama
Class: GL
Port State: Los Angeles
Owner: TMN
1
2
3
4
5
6
8
For the container vessel as shown in Figure 35 , the most important deficiency group is
the certificates followed by the group safety and fire and then stability and structure. The
last group is also interesting to see for this particular ship type and there is no real
explanation on why this particular deficiency group would show a relative high
contribution. Container ships are normally younger and better maintained vessels.
Figure 36 and Figure 37 show the results for the passenger vessel and other ship types.
The models for those two groups were produced under a slightly different method due to
the lack of observations and detention and are therefore not as accurate as the previous
models. What is interesting to see is a relatively high contribution of work related
deficiencies which might mean that these areas are inspected more with passenger
vessels and a relatively low contribution of safety & fire appliances related deficiencies
which might indicate that passenger vessels perform better in this area than other
vessels due to the relative importance and stringent requirements thereof.
The results of the other ship types are similar to general cargo, dry bulk and tankers but
also show a higher contribution towards detention with codes in the area of working and
living conditions. This group of ship types consists primarily of offshore supply vessels
and mobile offshore vessels, special purpose vessels and factory ships which might
explain the higher contribution of working related deficiencies.
81
Figure 36: Probability of Detention – Passenger Vessels
Passenger – All MoU’s
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 10 yrs
Tonnage: 29006 gt
Flag: Luxembourg
Class: BV
Port State: various
Owner: TMN
1
2
4
5
8
6,7
Figure 37: Probability of Detention – Other Ship Types
Other Ship Types – All MoU’s
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 15 yrs
Tonnage: 2982 gt
Flag: Greece
Class: GL
Port State: various
Owner: TMN
1
2
3
4 5
6,7
8
The next section will show the results for the regression that was performed for the
Caribbean MoU which had to be excluded from the rest of the regressions due to the
insufficient amount of data per ship type.
82
3.3.7. Results for the Caribbean MoU
Due to the lack of data, this section is difficult to analyze for the Caribbean MoU. Only
one model for the whole dataset could be produced where few variables (deficiency codes)
and one classification society remains significant. The statistics can be found in Appendix
13 for further reference. No difference can be seen based on flag, size or age or ship type.
Owners from traditional maritime nations and emerging maritime nations seem to
perform better than the other owner groups.
What is interesting to see is the high contribution for the deficiency code 1500 (safety of
navigation) followed by crew certificates (200) and the deficiency groups for stability and
structure and equipment & machinery. Ship certificates (100) also show a relatively high
contribution. The rest of the deficiency codes are not significant.
Figure 38: Probability of Detention – Caribbean MoU
Caribbean MoU
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Ship Certificates (1A)
Stability & Structure (6)
Equipment & Machinery (3)
Navigation & Commun (7)
Crew Certificates (1B)
Age: 15 yrs
Tonnage: 2982 gt
Flag: n/a
Class: GL
Port State: n/a
Owner: EMN
1B 1A
6
3
7
Note: Deficiency Group “certificates” split into crew and ship certificates
Since it is difficult to analyze each of the graphs individually and to compare the
differences, the next section will produce a series of graphs that allows doing so and
should visualize the differences of the contributions of the deficiencies across the regimes.
3.3.8. Differences in Deficiencies across the MoU’s
Figure 39 provides an overall overview of the percentage contribution of the deficiency
groupings towards the probability of detention per regime. The same basic ship profile
was used for all regimes in order to calculate the probability of detention. The resulting
factor is then converted into a percentage to the total weight of all deficiency codes
towards the probability of detention. By giving all ship types the same basic risk profile
and by assuming that the total weight towards the probability of detention based on the
83
deficiency codes is a 100%, the probabilities obtained through the deficiencies codes is
converted to a percentage weight. The resulting percentages not only take into account
the differences within each regime but also show the percentage weights of the deficiency
groups across the regimes.
The graph below can be read as follows. From the total contribution of the deficiency
groups towards detention for the USCG, 25% of weight towards detention derives from
deficiencies within the area of certificates, 17% within the area of the ISM code
(Management), 21% from deficiencies within ship & cargo operations etc. The lower the
percentage, the lower the overall weight of this deficiency group towards detention.
The graph should not be understood as a ranking of quality of the inspections but it
should merely give an insight into the different emphasizes with respect to the
deficiencies and reflects to a certain extent the average performance of all ships. Looking
at the overall graph in total, one can see that there are some differences across the
regimes but these are quite comparable when aggregated by all ship types.
Figure 39: Contribution Weight towards Detention: All Ship Types
All Ship Types
15% 19%
26% 25% 19%
13% 11%
17%
21%
18% 18% 13%
21% 19%
14% 13% 16%
13%
9% 9% 10%
6%
11% 10%
9%
10% 8%
9% 9% 12%
6% 9%
8%
7%
7%
6%
10%
10%
6% 6%
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
90.0%
100.0%
Paris MoU Vina MoU Indian O. MoU USCG AMSA
Contribution Weight towards Detention in %
Certificates (1) Management-ISM (8)
Ship & Cargo Operations (4) Safety & Fire (2)
Equipment & Machinery (3) Stability & Structure (6)
Working Conditions (5) Navigation & Communication (7)
4
1
8
2
5
3
6
7
4
1
8
2
5
3
6
7
4
1
8
2
5
3
6
7
The actual differences can best be seen when looking at each ship type separately and is
shown in Figure 40 to Figure 44. All graphs show a higher percentage for the deficiency
groups’ certificates, ship and cargo operations, the ISM code and safety & fire which is not
surprisingly.
The weight of these groups changes with respect to the regimes which might reflect the
different emphasis and the trade flows. Certificates are always inspected and are one of
the underlying factors for constituting “clear grounds”. Safety and fire appliances are
always part of the round that is performed during an inspection where life boats and their
84
equipment, launching equipment, lifejackets, immersion suits and fire fighting equipment
and systems are checked. This group also contains the testing of the emergency fire pump
which is not always performed but can be a detainable item if not working.
Figure 40: Contribution Weight towards Detention: General Cargo
General Cargo
17% 13%
26%
10%
22%
11% 14%
41% 19% 13% 16% 8%
9%
11%
20%
22% 25%
16%
12%
9%
9%
7%
11% 9%
12%
9%
7%
12% 10% 16%
6% 11%
5%
5% 7%
7%
7% 5% 6%
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
90.0%
100.0%
Paris MoU Vina MoU Indian O. MoU USCG AMSA
Contribution Weight towards Detention in %
Certificates (1) Management-ISM (8)
Ship & Cargo Operations (4) Safety & Fire (2)
Equipment & Machinery (3) Stability & Structure (6)
Working Conditions (5) Navigation & Communication (7)
2
3
6
5
7
8
4
1
2
3
6
5
7
4
1
8
2
3 6
5
7
4
1
8
Ship and cargo operations are a combination of SOLAS and MARPOL operational related
deficiencies where items such as the 15 ppm Alarm (oil water separator), the oil record
book, SOPEP85 and garbage management as well as fire and abandon ship drills can be
found. In addition, for tankers this group of deficiencies can be more important due to the
more complex cargo operations on chemical tankers, gas carriers and oil tankers. This
group of codes is expected to show higher percentages for the USCG since ships have to
perform fire and safety drills during inspections. Failure to comply with the drills to the
satisfaction of the inspector will show up under this code as well as under the ISM
(Management) code.
What is interesting to notice is the relative high contribution of ISM (Management)
related deficiencies for some ship types and regimes. As mentioned previously, this group
of codes represents the safety management system while the group of codes within ship
and cargo operations and safety & fire appliances represents the actual implementation
in daily shipboard operations. One regime might put more emphasis on the actual
implementation while others will check both aspects. If many deficiencies are found
which show a lack of maintenance and/or a lack of the implementation of operations
onboard, it will also be reflected in this group of deficiencies. The difference in this group
across the regimes also reflects the philosophy in inspecting and recording ISM related
deficiencies.
85 Ship Oil Pollution Emergency Plan
85
Figure 41: Contribution Weight towards Detention: Dry Bulk
Dry Bulk
14% 16% 11%
29%
14%
13% 11%
18%
16%
16%
30%
17%
35%
12% 35%
10%
21%
9%
9%
9.1%
7.9%
8.6%
7.0%
18%
8%
7% 6% 7%
10% 11%
23%
9%
8%
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
90.0%
100.0%
Paris MoU Vina MoU Indian O. MoU USCG AMSA
Contribution Weight towards Detention in %
Certificates (1) Management-ISM (8)
Ship & Cargo Operations (4) Safety & Fire (2)
Equipment & Machinery (3) Stability & Structure (6)
Working Conditions (5) Navigation & Communication (7)
4
1
8
2
5
3
6
7
4
1
8
2
5
3
6
7
4
1
8
2
5
3
6
7
Figure 42: Contribution Weight towards Detention: Tanker
Tanker
10%
21% 26% 25%
19%
17%
9%
7%
13%
14% 12% 15%
41%
19% 13% 33%
15%
12%
7%
8% 11%
11%
8%
6%
16%
12%
8% 7%
15% 11%
4%
5%
6%
5%
11%
14% 5%
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
90.0%
100.0%
Paris MoU Vina MoU Indian O. MoU USCG AMSA
Contribution Weight towards Detention in % Certificates (
1) Management-ISM (
8)
Ship & Cargo Operations (4) Safety & Fire (2)
Equipment & Machinery (3) Stability & Structure (6)
Working Conditions (5) Navigation & Communication (7)
4
1
8
6
2 3
5
7
4
1
8
6
2
3
5
7
4
1
8
6
2
3
5
7
86
Figure 43: Contribution Weight towards Detention: Container
Container
8%
20%
47% 42%
16%
14%
6%
9%
34% 50% 20%
9% 8%
7%
8%
8% 10% 16%
10%
11% 6%
18%
10%
6%
10% 13% 8% 6%
7%
7%
18%
7%
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
90.0%
100.0%
Paris MoU Vina MoU Indian O. MoU USCG AMSA
Contribution Weight towards Detention in %
Certificates (1) Management-ISM (8)
Ship & Cargo Operations (4) Safety & Fire (2)
Equipment & Machinery (3) Stability & Structure (6)
Working Conditions (5) Navigation & Communication (7)
4
1
8
2
5
3
6
7
4
1
8
3
6
7
4
1
8
3
6
7
2
2
5
5
Figure 44: Contribution Weight towards Detention: Passenger and Other Ship Types
Passenger and Other Ship Types
21% 24%
12%
14%
14%
16%
12%
9% 13%
10% 10%
10%
9% 7%
7%
13%
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
90.0%
100.0%
Passenger Other
Contribution Weight towards
Detention in %
Certificates (1) Management-ISM (8)
Ship & Cargo Operations (4) Safety & Fire (2)
Equipment & Machinery (3) Stability & Structure (6)
Working Conditions (5) Navigation & Communication (7)
4
1
8
2
5
3
6
7
The relative low weight percentage for the deficiencies within stability & structure is also
not surprising since it includes such items as ballast water tank or cargo holds
inspections which is difficult to be performed during normal cargo operations. Some
regimes might have a different policy with reference to entering enclosed spaces during
87
an inspection. This group of deficiencies only shows a higher contribution for dry bulk and
container vessels.
The deficiency groups dealing with working and living conditions which is a group of
codes related to the ILO varies across the ship types and regimes. The same applies for
the group of codes for navigation and communication. For passenger vessels and tankers,
the first group shows a higher contribution compared to container vessels and dry bulk
vessels while for the second group, dry bulk and general cargo seems to perform worst
with respect to navigational items. Also these two groups of codes vary the most across
the regimes which indicates the different ship profiles as well as the different emphasis
that is given during an inspection.
The lowest contribution for all ship types and regimes can be found for equipment and
machinery which is also not surprisingly. The engine room and its machinery is normally
part of an inspection round but is not core emphasis of a port state control inspection.
3.3.9. Differences in Port States
This section will look at the probability of detention showing the differences based on
selected ports for several regimes for the five major ship types. Not all cargo types are
handled in each port or port state. The same ship profile was used for all ship types with
the exception of tonnage and is as follows where the result can be seen in Figure 45.
The ship profile used in the graph is as follows:
1. Age: 13 years
2. Gross Tonnage: from 5,900 gt (general cargo), 38,995 gt (dry bulk), 27,322 gt
(container), 28,909 gt (tanker and passenger)
3. Class: Det Norske Veritas
4. Flag: Panama
5. Owner: Traditional Maritime Nation
6. Deficiencies: certificates (1), safety & fire appliances (3), ISM code (1), equipment
& machinery (1)
General cargo ships tend to have the highest probability of detention across all regimes
with the exception of AMSA. The other ship types vary. The USCG shows higher
probabilities for all ship types with the exception of the passenger vessel. The probability
of detention does not vary much from port to port for both the USCG and AMSA while it
can vary for the other regimes. This is understandable since it compares countries with a
group of several countries. This shows that there is room for harmonization of inspections
across the countries of the regimes as well as across the regimes.
It further shows that the worst performing ship type is the general cargo ship which is
not surprisingly since it is also a ship type which is not inspected by any of the vetting
inspection systems. The probability of detention of the ship type tanker varies the most
across regimes followed by dry bulk carriers. Tankers are extensively inspected by the
vetting inspection companies but depending on the deficiency found, might easily be
detained due to the potential high risk impact, an oil tanker or chemical tanker could
have if it is found to be sub-standard. The same should technically apply to passenger
vessels but in this category, political considerations might also play a rule and ships are
less likely to be detained.
88
Figure 45: Probability of Detention and Selected Port States
0.00 0.20 0.40 0.60 0.80 1.00
Canada
Italy
NL
Russia
Brazil
Chile
India
Iran
South Africa
Houston, TX
Los Angeles, CA
New York, NY
San Francisco, CL
Brisbane, QLD
Fremantle, WA
Melborne, VIC
Sydney, NSW
General Cargo Dry Bulk Tanker Container Passenger
3.3.10. Average Probabilities based on Inspector’s Background
The next series of graphs gives an insight into the probability of detention given the port
state control’s inspector previous background. This information was only available for one
of the regimes and is therefore only based on this particular regime. The requirements of
becoming a port state control officer varies across the regimes but most regimes with the
exception of the USCG require previous sea going experience or a background as a naval
architect. Figure 46 shows the average probability of detention per ship type and the
inspector’s background while Figure 47 gives the breakdown per deficiency category. It is
based on 16,773 inspections from the time period 1999 to 2004 where 682 records are
unknown and therefore left out of the total data to be drawn from.
The graphs show that the average probability of detention varies amongst the different
backgrounds of the port state control officers with respect to ship types where the largest
difference is around 5% on container vessels between inspectors with an engineering
89
background versus a naval architect background. Looking at the deficiency codes itself,
one can notice that most of the time the probability of detention of inspectors with an
engineering background seems to be slightly higher compared to a nautical background.
For the other two groups, the results are to be interpreted with caution since not much
data is available for these two groups.
Figure 46: Average Probability of Detention per Inspector’s Background
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
general cargo dry bulk tanker container passenger other st
Engineer Nautical Naval Architect Radio
Note: based on averages of the estimated probabilities obtained from the models
Figure 47: Average Probability of Detention per Inspector’s Background
0.00
0.05
0.10
0.15
0.20
0.25
0.30
C0100
C0200
C0300
C0400
C0500
C0600
C0700
C0800
C0900
C1000
C1100
C1200
C1300
C1400
C1500
C1600
C1700
C1800
C1900
C2000
C2100
C2200
C2300
C2500
C2600
Engineer Nautical Naval Architect Radio
Note: based on averages of the estimated probabilities obtained from the models
90
The two main groups are inspectors with either a nautical background or an engineering
background. The difference between these two groups can be up to 4% for code 800
(Accident prevention) but most of the time lies between 1 to 3%. What is interesting to
observe is that inspectors with engineering background do not necessarily show a lower
probability in deck related deficiencies such as code 1500 (safety of navigation) or 1600
(radio communications) while it does show a difference in code 1400 (propulsion and aux.
machinery) in comparison to inspectors with a nautical background.
This analysis can conclude that there are differences which are expected to exist but that
this type of analysis would require further insight and better underlying data collection
for the other two groups (naval architect and radio) in order to make a final conclusion on
the subject in question. It is a first insight into trying to explain the differences in the
probability of detention and the use of the deficiency codes towards it.
3.3.11. Overall View Based on Average Probabilities
The final section will provide an overall view of the probability of detention based on all
ships in the total inspection dataset with more than 15 deficiencies and with no
deficiencies and their estimated average probabilities. The results are based on 5,212
ships and 98,953 ships respectively and are shown in Figure 48 and Figure 49.
Figure 48: Probability of Detention per Ship Type (> 15 deficiencies, 5,212 ships)
Average Probability of Detention with more than 15 deficiencies
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
general cargo dry bulk tanker container passenger other
Paris MoU (1) Vina MoU (2) Indian O. MoU (3) USCG (4) AMSA (5) Average (6)
1
2
3
4
5
6
Note: based on averages of the estimated probabilities obtained from the models
The difference across the regimes is primarily based on the contribution of the deficiency
codes and the port states. While some differences can be found in flag and class, age and
vessel size are not the major factors contributing the difference as could be seen when
testing the coefficients for restrictions in section 3.3.3. Step 2 Results: Coefficient Testing
(Performed in 2 Rounds).
The graphs should not be used as a measurement of the quality of the inspections. It
shows the differences with respect to detention in mainly the deficiency codes as well as
91
the port states. The results for passenger vessels and other ship types are a less accurate
measurement due to the fact that only one model per ship types could be formed and not
for each MoU. It therefore cannot distinguish the differences based on class, flag, age, size
and deficiencies across the MoU’s but only gives an overview of the differences based on
the port states and a variable indicating the regime (e.g. passenger vessel coming into
MoU 1).
Figure 49: Probability of Detention per Ship Type (No deficiencies, 98,953 ships)
Average Probability of Detention with no deficiencies
0.000
0.005
0.010
0.015
0.020
0.025
0.030
general cargo dry bulk tanker container passenger other
Paris MoU (1) Vina MoU (2) Indian O. MoU (3) USCG (4) AMSA (5) Average (6)
1
2
3
4
5
6
Note: based on averages of the estimated probabilities obtained from the models
The basic probability based on zero deficiencies can be understood as the portion of the
probability based on the ship profile and lies between 0.5% and 1.5% for most ship types
and regimes. This is also the portion that will be related to the probability of casualty in
the next part of the thesis to compare with the probability of casualty. Only other ship
types for the Indian Ocean MoU shows a higher percentage. The picture then changes
when looking at ships with more than 15 deficiencies where the average probability
increases accordingly due to the factor associated with the deficiencies.
3.4. Summary of Major Findings: Port State Control
About half of the world fleet (47%) is subject to port state control. Out of these 47%, most
ships inspected are general cargo ships (36%) followed by dry bulk (26%), tankers (19%),
containers (10%) and passenger vessels and other ship types. Out of the total inspections,
53.8% are inspections without deficiencies and 5% end up in a detention while aggregated
by ship, the 53.8% decreases to 16.3% and detention increases from 5.44% to 24.6% of all
inspected vessels for the time frame 1999 to 2004. 66% of the ships detained (1999 to
2004) have been detained once and 6% have been detained four or more times. The
average amount of inspection frequency lies by 7 over the time period 1999 to 2004. This
amount might be higher in reality since data from some regimes could not be obtained
and not the whole time frame can be covered by all regimes who did supply data. Around
92
68% of the ships with deficiencies have 1 to 5 deficiencies and 6% show more than 15
deficiencies.
The basic ship profiles given by age, size, flag, class and ownership do not vary
significantly across the regimes with respect to the probability of detention. Most
differences across the regimes can be found within the use of deficiencies towards
detention and the port states. When aggregated by ship types, the differences average out
but looking at the ship types individually, one can see that certain codes show higher
contributions compared to each other within each of the regimes. The basic ship risk
profile for all regimes is between probabilities of detention of 0.5% to 1.5%.
Highest contribution can be found for the deficiency groups’ certificates, ship and cargo
operations, the ISM code and safety & fire appliances while lowest contribution is found
for machinery and equipment. Ship and cargo operations seem to be more important for
tankers while stability and structure are highest for dry bulk carriers and containers.
What is interesting to notice is the relatively high contribution of ISM (Management)
related deficiencies for some ship types and regimes. This group of codes represents the
safety management system while the group of codes within ship and cargo operations and
safety & fire appliances represents the actual implementation in daily shipboard
operations. One regime might put more emphasis on the actual implementation while
others will check both aspects. The deficiency groups working conditions and navigation
and communication show the highest variation across the regimes.
The difference between the probabilities of detention given a certain background of an
inspector is reflected for certain deficiency codes but not necessarily as one would expect
intuitively. For inspectors with nautical background versus engineer background, the
differences in the probability of detention can be up to 4% for code 800 (Accident
prevention) but most of the time lies between 1 to 3%. What is interesting to observe is
that inspectors with engineering background do not necessarily show a lower probability
in deck related deficiencies such as code 1500 (safety of navigation) or 1600 (radio
communications) while it does show a difference in code 1400 (propulsion and aux.
machinery) in comparison to inspectors with a nautical background.
This chapter provides a method to calculate the estimated probability of detention which
will be linked to the next chapter which will provide the estimated probability of casualty
in order to see if the correct ships are targeted for inspection or if targeting can be
improved. Contrary to industry knowledge, this analysis does not necessarily confirm that
the performance of flag or class varies significantly across the regimes. What it does
confirm is that the differences across the regimes seem to be influenced by the port states
as well as the use of the deficiency codes towards detention.
93
PART III
This part of the thesis consists of three chapters and will link the inspection datasets to
the casualty dataset and try to find the answer to the main research questions which are
as follows:
1. What is the overall effect of inspections on casualties?
2. Are the correct ships targeted for inspections and can targeting be improved?
3. Can inspections be further enhanced?
The part starts with a short explanation on the selection of port state relevant casualties
and the respective datasets used for the different regressions. It will link the probability
of detention obtained in part II to the probability of casualty.
The regressions are performed on two datasets. First, an analysis is performed on the
casualty dataset itself and second on a matched dataset of inspection ships with
casualties versus inspected ships without casualties.
94
95
Chapter 5: Key Statistics on Casualties and Overview
of Datasets
This chapter will first give an overview of the selection of the casualties used for the
regressions as well as explain the datasets that are used. It will further give some key
descriptive statistics on casualties with reference to ship particulars, casualty severity
and locations.
5.1. Selection of Port State Control Relevant Casualties
Considerate care was given on the selection of casualties for the analysis. From the
casualty dataset within the time period 1999 to 2004 of 9,851 cases, the following cases
were eliminated.
1. Cases due to extreme weather conditions such as hurricanes, typhoons, gales and
very heavy storms
2. Ships attacked by pirates or ships lost due to war
3. Ships involved in a collision with no identified fault86
4. Any other miscellaneous items not relevant to PSC such as drugs found, virus
outbreaks of passengers or accidents which happened in dry docks
5. Not PSC relevant ships types such as ferries, the fishing fleet, tugs or government
vessels. The fishing fleet cases were kept separate and a separate analysis was
performed based only on the fishing fleet above 400gt.
The remaining 6291 cases concern 6,005 ships when aggregated by IMO number and
were then reviewed and re-grouped into the three groups of seriousness as per IMO MSC
Circular 953 of December 2000 and already listed in detail in Chapter 3 of this thesis:87
1. Very serious casualties: casualties to ships which involve total loss of the ship,
loss of life, or severe pollution.
2. Serious casualties are casualties to ships which do not qualify as “very serious
casualties” and which involve fire, explosion, collision, grounding, contact, heavy
weather damage, ice damage, hull cracking, or suspected hull defect, etc. resulting
in: immobilization of main engines, extensive accommodation damage, severe
structural damage, such as penetration of the hull under water, etc. rendering the
ship unfit to proceed, or pollution (regardless of quantity); and/or a breakdown
necessitation towage or shore Helpance.
3. Less serious casualties are casualties to ships which do not qualify as “very
serious casualties” or “serious casualties” and for the purpose of recording useful
information also include “marine incidents” which themselves include “hazardous
incidents” and “near misses”.
In addition to the classification of seriousness of casualties, the cases were also examined
and re-classified according to casualty first events which are used for the regression
analysis in the last section of this thesis. The casualty first events are classified as
follows:
86 The identification of “no fault” in this case was not straight forward and some cases still
included in the dataset might be ships with no fault and were not eliminated due to lack of
exactness of data.
87 as per IMO MSC Circular 953, 14th December 2000
96
o Deck and Hull related casualties: Deck and hull related items such as maintenance
items (cracks, holes, fractures, hatch cover problems, cargo equipment failure,
lifeboat gear failure, anchor and mooring ropes problems), stability related items
such as capsizing, listing, cargo shifts and flooding
o Fire/Explosion: Fire and Explosion anywhere on the vessel (main areas are engine
room)
o Engine or machinery related casualties: Engine related items including engine
breakdown, black outs, steering gear failure and propulsion failure
o Wrecked/Stranded/Grounded: Wrecked, Stranded, Grounded where a large
portion of the ships in this category are stranded or grounded. 112 ships in this
category are ships that were lost and therefore, could probably be classified as
wrecked. Nevertheless, for the purpose of the analysis, this category is to be
interpreted primarily for stranded and grounded vessels.
o Collision and Contact: Collision and Contact
The next section will give a short overview of some key descriptive statistics with
reference to the seriousness of casualty, ship types, locations and the casualty first
events.
5.2. Key Descriptive Statistics for Casualties
Based on the definitions of the seriousness of a casualty used by IMO and the selection of
port state control relevant casualties as explained earlier, Figure 50 gives an overview of
the split up of the seriousness of casualties. The percentage of serious casualties and less
serious casualties changes when comparing total casualties to port state control relevant
casualties only. Although all three categories are taken into consideration in the
regressions in the two following chapters to come, very serious and serious casualties are
to be understood to be more relevant for port state control than less serious casualties.
Figure 50: Seriousness of Casualties (1999 to 2004)
Note: compiled by author
Figure 51 then gives an overview of the split up of the casualty first events. The graph is
not detailed but can be understood as a first attempt to break up the casualty types into
relevant categories. The lack of information and fragmentation of the data does not
permit a better split up. What is interesting to see is the high amount of engine and
machinery related events of about 32% (engine breakdown, engine black out, steering
gear failure and propulsion failure) while the probability of detention has shown a
PSC Relevant Casualties
Serious
53%
Less
Serious
34%
Very
Serious
13%
Total Casualties
Very
Serious
14% Less
Serious
38%
Serious
48%
97
relative low probability of detention based on deficiencies in the area of propulsion and
auxiliary machinery (code 1400).
Figure 51: Casualty First Events per Ship Type (1999 to 2004)
Engine
Breakdown
25%
Fire/Explosion
10%
Contact
19%
Collision
8%
Deck & Hull
Related
12%
Wrecked/
Stranded/
Grounded
19%
Engine
Black Out
2%
Propulsion
Failure
2%
Stearing Gear
Failure
3%
PSC Relevant Cases, compiled by author
Looking at the casualties per ship type for the whole time period (1993 to 2004) as can
bee seen in Figure 52, most casualties can be found with general cargo and multipurpose
ships and dry bulk carriers.
Figure 52: Ship Types and Casualties (1993 to 2004)
Container
7.3%
Dry Bulk
15.7%
General cargo &
Multipurpose
30.2%
OtherST
7.7%
Gas Carrier,1.2%
Passenger, 2.5%
Ro-RoPax, 4.5%
Ferry/Hydr.,2.9%
Fishing Vessel
6.9%
OBO, 0.5%
Tanker, 0.5%
Oil Tanker,
11.4%
Reefer
1.7%
Chemical
Tanker, 1.1%
Ro-RoCargo
6.0%
Total Casualty Data and Time Frame, compiled by author
98
What is interesting to see in the graph is that the casualties on fishing vessels account for
7% of the total casualties of the total dataset but are primarily not part of the port state
control system. Very few fishing factory vessels which are technically not fishing vessels
(0.3%) are inspected by the port state control regimes. In the casualty regression analysis,
the fishing fleet is treated separately and only fishing vessels above 400gt are taken into
account.
Figure 53 gives the split up per seriousness and Figure 54 by pollution type and region.
The regions with the highest percentage of very serious casualties are the Gulf & the
Indian Ocean, the South China Sea, the Philippines and the West African coast. For
serious casualties, regions with above 50% are the British Isles and North Sea, the Baltic
Sea, the Canadian and Russian Arctic, Iceland, the Mediterranean West, the South
Atlantic US West Coast and Panama Canal. A certain percentage of cases were unknown
where about 50% are allocated to serious casualties, 33% to very serious and the rest to
less serious casualties.
Figure 53: Seriousness of Casualty per Region (1999 to 2004)
25.9%
9.8%
18.6%
10.3%
14.4%
14.1%
13.1%
15.4%
14.7%
17.3%
28.2%
33.7%
24.0%
17.2%
43.6%
48.9%
56.9%
50.6%
56.4%
49.5%
59.5%
38.5%
41.9%
54.3%
43.4%
39.5%
48.5%
47.8%
55.6%
39.1%
52.9%
51.3%
38.0%
47.8%
30.5%
41.4%
35.7%
43.0%
25.0%
49.5%
30.2%
47.1%
44.0%
32.6%
41.3%
52.7%
45.9%
37.5%
27.2%
32.7%
38.7%
15.0%
38.0%
35.0%
8.4%
5.6%
7.8%
6.4%
7.5%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Gulf & Indian Ocean
Australasia & South Pacific
Baltic Sea, Kiel Kanal
British Isles and North Sea
Canadian and Russian Arctic, Alaska
Great Lakes
Iceland
Japan, Korea and North China
Mediterranean East & Black Sea
Mediterranean West
NA East Coast and North Atlantic
NA West Coast and North Pacific
Newfoundland
SA East Coast and South Atlantic
SA West Coast and Panama Canal
South China Sea and Philippines
Suez Canal, Red Sea
Unknown
West African Coast
West Indies and Gulf of Mexico
Very Serious Serious Less Serious
Based on casualty data from 1999 to 2004
As for the split up of the type of pollution, the details can be seen in the graph below
which gives a detailed view of the type of pollution per area. Most pollution for the time
period 1999 to 2004 originated from heavy and crude oil followed by chemicals, light oils
and any other oils which are mainly oily waters. The split up varies considerably per
region.
99
Figure 54: Pollution Type per Region (1999 to 2004)
92%
30%
87%
92%
11%
81%
100%
29%
79%
80%
3%
69%
27%
61%
100%
76%
76%
8%
65%
10%
88%
33%
8%
71%
19%
8%
0%
5%
2%
2%
1%
7%
2%
20%
2%
30%
97%
12%
1%
8%
24%
11%
31.0%
10.2%
88.1%
12.3%
5.0%
3%
4.7%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Gulf & Indian Ocean
Australasia & South Pacific
Baltic Sea, Kiel Kanal
British Isles and North Sea
Canadian and Russian Arctic, Alaska
Great Lakes
Iceland
Japan, Korea and North China
Mediterranean East & Black Sea
Mediterranean West
NA East Coast and North Atlantic
NA West Coast and North Pacific
Newfoundland
SA East Coast and South Atlantic
SA West Coast and Panama Canal
South China Sea and Philippines
Suez Canal, Red Sea
West African Coast
West Indies and Gulf of Mexico
% Heavy & Crude Oil % Light Oil % Other Oil % Chemicals
Based on casualty data from 1999 to 2004
5.3. Overview of Inspections and Casualties
The last section in this chapter should give an overview of the PSC eligible fleet in
relation to port state control inspections and casualties which is shown in Figure 55. The
fishing vessels (>400gt) of which the casualties are also considered in the regressions are
also incorporated into this figure.
The graph gives an overview of the total fleet where the portion on top shows the portion
not exposed to port state control minus a portion of the fishing fleet (above 400gt) which
is also considered in this study. The right hand lower part of the graph represents
inspected ships and the left hand side of the graph represents ships that have not been
inspected by the respective regimes88 or not inspected at all. The lower middle portion
summarizes the vessels that had casualties.
One can see that about an equal amount of ships that had not been inspected by the
regimes in question (they might have been inspected by another regime only) did not have
a casualty – about 46% (23.5% plus 23.3%) of the world fleet including fishing vessels
88 As explained previously, some Port State Control regimes decided not to participate in this study
such as the Tokyo MoU, the Black Sea MoU or the Mediterranean MoU while others did not have
any data available yet.
100
above 400gt. Not inspected ships with casualties accounted for 2.4% of the world fleet
versus 1.7% of the vessels which had a casualty and were inspected without any related
time frame and 2.5% of the vessels were inspected six months prior to a casualty.
Figure 55: The Overall View on Inspections and Casualties (1999 to 2004)
inspected
with casualty
w/o time
connection
1.7%
inspected
6 m prior to
casualty
2.5%
not
inspected
with casualty
2.4%
not
inspected
without
casualty
23.5%
inspected
without
casualty
23.3%
not PSC
eligible fleet:
47% with
fishing
(53% w/o
fishing)
Total Ships: 93,719
Note: Casualties are over a time frame (1999 to 2004) and only PSC relevant casualties are shown
in this graph plus the fishing fleet (>400gt)
What is interesting to notice but which is not shown in this graph is that, based on the
individual port state control inspections, 54% of all inspections were inspections with zero
deficiencies while when aggregated by ship and taken as a summary of all inspections
performed on vessels, this percentage reduces to 16% of all inspections for the time frame
1999 to 2004. On the other hand, if only looking at the last inspection six months prior to
a casualty as shown in the figure with 2.5% of the total world fleet, 52.3% of these vessels
were ships with zero deficiencies. The lower left side of Figure 55 then shows the portion
of ships that have not been inspected but had a casualty which accounts for 2.4% of the
vessels. In number of ships, this accounts for 2,213 vessels or approx. 369 ships per year.
This could indicate a possible room for improvement of targeting vessels.
Zooming into inspected ships only, the result can be seen in Figure 56. From a total of
25,836 ships with inspections, 3,956 ships had a casualty and 2,321 ships were inspected
within a time frame of six months prior to a casualty. Of these 2,321 ships, 162 were
detained which accounts for 0.6% of total inspected vessels (25,836), 4% of vessels with
casualty (3956 ships) or 7% of ships with casualty and inspection six months prior to
casualty.
It is interesting to see that the percentage of very serious and serious casualties is higher
for vessels that have been inspected and detained six month prior to the casualty then for
101
vessels that have not been detained. The effect of detention will be closer looked at in
Chapter 7 of this thesis but Figure 57 gives a further split up of the seriousness of the
casualties and detention. Detained ships show a considerable higher amount of very
serious and casualties then not detained ships.
Figure 56: Ships Inspected in Relation to Ships with Casualties (1999-2004)
inspected 6
months prior
9%,
(2,321 ships)
not inspected 6
months prior
6.3%,
(1,635 ships)
inspected w/o
casualty
84.7%,
(21,880 ships)
detained
162 of 2,321
ships (7%)
not
detained
2,159 of 2,321
ships (93%)
3,956 ships
Total Ships: 25,836
Note: compiled by author
Room for improvement to target vessels can be identified in the area of vessels that had a
casualty but were not inspected and in the portion of vessels that were inspected but
without any time related to the casualty. Another area of improvement for the inspections
itself versus the targeting could be within the portion of ships that was inspected and/or
detained six months prior to a casualty.
Figure 57: Ships Detention and Seriousness of Casualty (1999-2004)
not detained
very
serious
7%
serious
56%
less
serious
37%
detained very
serious
13%
serious
65%
less
serious
22%
Note: based on ships that were inspected six month prior to casualty
Figure 58 gives an overview of the mean amount of deficiencies found six month prior to a
casualty per flag state group while Figure 59 shows the split up for IACS recognized
102
classification societies and non IACS recognized classification societies. Black listed flag
states have an average of 4.3 deficiencies versus 1.7 deficiencies for white listed flag
states. Ships of Non-IACS classification societies have an average of 6.5 deficiencies
versus 2.4 for ships with ICAS classification in an inspection at least six months prior to a
casualty.
Figure 58: Mean Amount of Deficiencies per Flag State: 6 months prior to casualty
4.3
2.7
1.7
2.2
2.8
0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
4.00
4.50
5.00
Black Grey White Undefined Average
Mean amount of deficiencies
Note: based on ships that were inspected six month prior to casualty
Figure 60 shows the mean amount of deficiencies that are found previous to a casualty
per seriousness of casualty and detention. Ships that have been detained show a
significant higher amount of deficiencies than ships that have not been detained prior to a
casualty.
Figure 59: Mean Amount of Deficiencies per Class: 6 months prior to casualty
6.5
2.4
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
Non IACS IACS
Mean amount of deficiencies
Note: based on ships that were inspected six month prior to casualty
103
Figure 60: Mean Amount of Deficiencies per Seriousness of Casualty
16.9
14.9
16.1
10.0
8.3
9.0
3.0
1.7 1.9
0.00
2.00
4.00
6.00
8.00
10.00
12.00
14.00
16.00
18.00
very serious serious less serious
Mean Amount of Deficiencies
detained average not detained
Note: based on ships that were inspected six month prior to casualty
5.4. Overview of Dataset Combinations for Regressions
This chapter will give an overview of the selection of datasets for the two types of casualty
regressions that are used in the two chapters that will follow. Figure 61 shows the
combination of datasets that were used.
Figure 61: Description of Dataset Combinations for Casualty Regressions
Ships without
Casualty Ships with
Casualty
Port StateControlled Ships
(matched twins)
Regression 1: Casualty
Normal Models
Based on all PSC eligible ships
above 400 gt
Variables:
Age, Size, Ship Type, Flag, Class,
Owner, Ship Yard,
Class/Owner/Flag Changed, Class
Withdrawn,
Legal Instruments ratified,
PSC inspected
Regression 2: Twin Models
Based on all PSC ships
with/without casualty matched as
twins
Variables:
Same as above plus:
PSC inspected
Industry Inspections
Detention
Deficiencies Found
Time in-between inspections
Refined View
Overall View
SSSpppllliiittt pppeeerrr ssseeerrriiiooouuusssnnneeessssss:::
vvveeerrryyy ssseeerrriiiooouuusss,,, ssseeerrriiiooouuusss,,, llleeessssss ssseeerrriiiooouuusss
Wiiittthhh aaaddddddeeeddd tttiiimeee fffrrraaameee
pppeeerrr ssseeerrriiiooouuusssnnneeessssss
pppeeerrr cccaaasssuuuaaallltttyyy fffiiirrrsssttt eeevvveeennntttsss
104
The first type of regressions (Chapter 6) is based on the total dataset of PSC relevant
casualties and all ships with no casualties in the time frame 1999 to 2004. Therefore, it is
based on the total world fleet. This type of regression should give an insight of the overall
effect of inspections in comparison to ships that were not inspected by the respective
regimes. It also gives some insight into the target factor to see if the correct ships are
chosen for inspections by comparing the probability of detention provided by Part II of
this thesis with the probability of casualty.
The second type of regression (Chapter 7) then only takes port state controlled ships into
account and compares ships that had a casualty with a ship that did not have a casualty.
For the ships with casualties, twins are matched according to certain criteria which will
be explained in detail in the respective chapter of the model itself.
The reason for the matching twins (sister ships) is to refine the comparison and to allow
the filtering out of the variables that are of interest in these regressions – the variables
that describe the quality of an inspection such as detention, the deficiencies that were
found, the regime who inspected the vessel, vetting inspections and the time in-between
PSC inspections. This regression further introduces a time frame and only looks at ships
with an inspection six months prior to the casualty. In addition, a series of separate
regressions based on casualty first events is introduced in order to gain more insight
between the deficiencies that are found prior to a casualty and the casualty first events.
It is important to notice that ships appear only once in each of the datasets meaning that
if a ship had several casualties, its casualty history is reflected in the record but the ship
does not appear several times as a separate observation. The same applies for the
inspection history. The numbers reflected in the variables are therefore averages of the
ship’s total history of inspection and casualties. The only time a ship can re-appear in the
datasets that are used for either type of regression is because of the seriousness of a
casualty or if it is a multiple twin.
105
Chapter 6: Probability of Casualty – Overall View
This chapter tries to identify the difference of the probability of having a casualty (very
serious, serious, and less serious) between inspected ships and not inspected ships to see
what the effect is of inspections on casualties. It further links the findings of the
probability of casualty with the probability of detention to see if the correct ships are
targeted for inspections. The chapter will first start with an explanation of the dataset
that is used for the regression analysis.
6.1. Preparation of Datasets and Sample Sizes
Figure 62 gives an overview of the steps that were performed to prepare the relevant
datasets. The first step was the selection of casualties and their re-classification according
to the IMO guidelines. The second step was the selection of the relevant datasets and the
third step the selection of variables used in the regressions.
Figure 62: Description of Methodology Used
6.1.1. The Selection of Relevant Casualties
The selection of relevant casualties is covered at the beginning of Chapter 5 and is not
repeated here. The data used for the regressions to follow is based on 6005 ships (6169
ships including the fishing fleet) with a casualty. The classification of seriousness was
also explained previously but is repeated here in summarized form89 for easier
understanding:
o Very serious casualties: casualties to ships which involve total loss of the ship,
loss of life or severe pollution
o Serious casualties are casualties to ships which do not qualify as “very serious
casualties” and which involve fire, explosion, collision, grounding, contact, heavy
weather damage, ice damage, hull cracking, or suspected hull defect, etc. resulting
in: immobilization of main engines, extensive accommodation damage, severe
structural damage, such as penetration of the hull under water, etc. rendering the
ship unfit to proceed, or pollution (regardless of quantity); and/or a breakdown
necessitation towage or shore Helpance.
89 as per IMO (MSC Circular 953, 14th December 2000)
Step 1
Step 2
Step 3
The selection of PSC relevant casualties and the
classification of the cases by seriousness according to
IMO guidelines
The selection of relevant datasets of ships with
casualties and ships without casualties and their
respective % to the world fleet
The selection of the variables that are used in the
regressions and preparation of the variables
106
o Less serious casualties are casualties to ships which do not qualify as “very
serious casualties” or “serious casualties” and for the purpose of recording useful
information also include “marine incidents” which themselves include “hazardous
incidents” and “near misses”.
6.1.2. The Selection of Relevant Datasets
The second step was the construction of the dataset and is based on the total casualty
dataset (a combination of three sources), the inspection dataset (a combination of six
sources and data from the industry), and the general ship dataset including ship’s
particulars and ship history data. Table 24 lists the split up of the total dataset between
ships that have been inspected and ships that have not been inspected for the time period
in question. The fact that the ships have not been inspected does not necessarily mean
that they have not been inspected by any of the other regimes where data could not be
obtained such as the Tokyo MoU, the Black Sea MoU, or the Mediterranean MoU.
Therefore, the dataset measures ships that have been inspected by the combined PSC
dataset with ships that have not been inspected at all or inspected by another regime.
This will be taken into consideration when interpreting the results. It is difficult to
anticipate how the results would change when repeating the regressions with a full
inspection dataset.
Table 24: Split up of Ships with casualties versus non-casualties (1999 to 2004)
Summary
Ships not
inspected
Ships
inspected
Total
Ships
% not
inspected
%
inspected
Casualties 2213 3956 6169 35.9% 64.1%
No Casualties 22061 21880 43941 50.2% 49.8%
Total 24274 25836 50110
Note: the table includes the fishing fleet used in a separate regression (ships above 400gt)
The total dataset for casualties is based on 6169 ships with a casualty (6,005 without the
fishing fleet) and 43,941 (37,812 without the fishing fleet) ships without casualties where
the split up of the ships represent a sample similar to the world fleet for ships above
400gt as can be seen in Table 25 .
Table 25: Split up of Ship Types of Sample versus World Fleet (1999 to 2004)
Casualties Commercial > 400gt*)
No Casualty Casualty Sample Sample Total Ships
Ship Type Sum Sum Total % to Total % to Total World Fleet
General Cargo 12539 2665 15204 43.20% 34.7% 33.0%
Dry Bulk 5684 1017 6701 16.49% 15.3% 14.0%
Container 3571 492 4063 7.98% 9.3% 12.0%
Tanker 10184 1033 11217 16.75% 25.6% 25.0%
Passenger 1241 296 1537 4.80% 3.5% 3.0%
Other Ship Types 4593 502 5095 8.14% 11.6% 13.0%
Total w/o fishing 37812 6005 43817 n/a 100.0% 100.0%
Fishing (>400gt) 6129 164 6293 2.66% n/a n/a
Total w. fishing 43941 6169 50110 100%
*) Note: the figures on the last column link up to Figure 21: Ships Eligible for Inspection
The fishing vessels are not split into seriousness but only one regression is performed on
the total cases of fishing vessels above 400gt. The ideal situation would have been to
107
include all fishing vessels of which the majority is under 400 gt but due to lack of data,
this could not be done. In total four separate regressions are performed: one for each type
of seriousness and a separate for all the cases of the fishing fleet. Multiple casualties were
not taken into account. A ship with a casualty can therefore appear in each of the
datasets if a ship had multiple combinations of casualties. If a ship had the same type of
seriousness of casualty more than once (e.g. 2 serious casualties), the ship is only taken
into consideration once. Table 26 gives an overview of the amount of observations in each
model.
Table 26: Number of Observations in the End Model
Nr. of Observations Very Serious Serious Less Serious Fishing Fleet
Total Observations 38076 41009 39929 6289
No Casualties 37354 37811 37811 6129
Casualties 722 3198 2118 160
Remarks Without
Caribbean MoU
Note: Figures are final figures used in the models and without outliers
One could argue that the number of ships with casualties (the 1’s in the regressions) is
over-represented in the dataset in comparison to the number of ships without casualties
(the 0’s) since the casualties are an accumulated figure over a time frame of six years
where the ships with no casualties are of the same time period but are not counted six
times. This raises the question if increasing the dataset accordingly adds more
explanatory power to the regression and if the estimations would be different? This
subject was investigated by Cramer, Franses, and Slagter (1999)90 for various sizes of a
reduced dataset of zeros and no significant change was found. Therefore, it is assumed
that the difference in the sample sizes does not have a serious effect on the coefficients
and that by adding more data on the side of the zeros will not add much explanatory
power to the models. The resulting probability represents the probability of having at
least one casualty in a six year time period and will be converted to a yearly probability
for the visualization part. The next section will explain the selection of variables that are
used in the regressions as well as give an explanation of the model itself.
6.2. Selection of Variables and Model Explanation
The binary logistic model was explained in detail in Chapter 4 of this thesis and will not
be repeated here. It can be written in the form of Equation 4 where the term xiβ can
change accordingly to the casualty model and is shown in detail in Equation 5 and its
variables are listed in Table 27 for further reference. The choice of the variables is based
on experience gained from observing inspections and common knowledge of the shipping
industry. The probabilities produced are for any individual ship (i) and the rest of the
notation is defined as follows: ℓ represents the variable groups, nℓ is the total number of
variables within each group of ℓ and k is an index from 1 to nℓ.
Equation 4: Probability of Casualty Standard Model
x β)
x β)
i
i
P (
(
i 1 e
e
+
=
90 Cramer, Franses and Slagter, (1999) Econometric Institute Research Report, 9939/A
108
Four different casualty models are created: very serious, serious, less serious and fishing
fleet. For estimation, QML (Huber/White) standard errors and covariance is used which is
a standard option in Eviews91. The reason for using this option was explained already
under Chapter 4 and in an effort to keep the methods similar across the different types of
regressions; it was decided to also use QML for all types of regressions.
Equation 5: Definition of term xiβ of Casualty Standard Model
Σ PSC
LIOWN LIFS DH Σ RS GR
FSInd Σ OWN OWNInd Σ SY
Σ CL CLInd CLWdr Σ FS
ln(AGE ) ln(SIZE ) Σ ST STInd
1 18
16 17
1
13 14 15 1
12
1
10 11 1
1
9 1
8
1
5 6 7 1
1
1
1
1
18
16
10 12
5 8
0 1 2 3 4
,k k,i
n
k
,k k,i i
n
i i i k
,k k,i
n
,k k,i i k
n
i k
,k k,i
n
,k k,i i i k
n
k
k,i i
n
i i i k
β
β β β β β
β β β β
β β β β
x β β β β β β ,k
3
=
−
=
−
=
−
=
−
=
−
=
−
=
+
+ + + + +
+ + + +
+ + + +
= + + + +
Table 27: List of Variables Used in Casualty Normal Models
Variable Type Total
Total Number of Variables nℓ
Variable
ℓ Casualty (Very Serious, Serious, Less Serious) plus a
separate regression for the fishing fleet 0/1 1
Ln(AGE) 1 Vessel Age at the time of casualty (or inspection) C 1
Ln(SIZE) 2 Vessel Size in gross tonnage C 1
ST 3 Ship Type (including fishing vessels) D 7
STInd 4 Indicates if ship type changed since construction D 1
CL 5 Classification Societies at time of casualty (or inspection) D 42
CLInd 6 Indicates if classification society changed over time D 1
CLWdr 7 Indicates if classification society withdrew D 1
FS 8 Flag State at the time of casualty/inspection D 130
FSInd 9 Indicator if flag changed over time D 1
OWN 10 Ship Owner Countries D 6
OWNInd 11 Indicates if ownership was changed over time D 1
SY 12 Country where ship was primarily built (>100 ships) D 37
LIOWN 13 Number of legal instruments owner country rectified C 1
LIFS 14 Number of legal instruments flag state has rectified C 1
DH 15 Double Hull D 1
RS 16 Ships inspected by Rightship (Vetting Inspection) D 5
GR 17 Ship certified by Greenaward D 1
PSC 18 Indicated total # of inspections per ship per regime (Sum) C 6
Total for the whole dataset (split into seriousness) 244
C = continuous, D = dummy of categorical variables
91 software used to perform the regressions
109
The variables used in the regressions are similar to the variables used for targeting ships
such as age, size, flag, class and owner. The difference to the probability of detention is
that it is based on an aggregated ship history versus separate observations.
The variables which indicate change over time such as ship type, class, flag, or ownership
are based on information obtained by Lloyd’s Register Fairplay and go back in time to
either the time the vessel was constructed or at least the last five to six years of the ship
being in operation. Three types of inspection variables are introduced. First, a variable
which describes if a ship has been inspected by one of the industry vetting inspection
systems (Rightship), second, if a ship is certified by Greenaward and third, if a ship has
been inspected by one of the respective regimes. Figure 63 shows the variable structure of
the two types of ships that are used in this dataset. Ship type 1 reflects ships that have
been inspected and ship type 2 are the ships that have not been inspected. Both types can
either have a casualty or no casualty as already indicated in Table 26.
Figure 63: Visualization of Variable Structure: Normal Models
Initial Event:
Ship Built
Age, Size, DH, ST
Ship Yard
Life Events:
Change of ST
Change of Flag
Change of Owner
Change of Class
Class Withdrawal
No Inspections
Particulars over Time
Average Class
Average Flag
Average Owner
Legal Framework:
Number of LI ratified
(Flag and Owner)
Initial Event:
Ship Built
Age, Size, DH, ST
Ship Yard
Life Events:
Change of ST
Change of Flag
Change of Owner
Change of Class
Class Withdrawal
Vetting Inspection
PSC Inspection (Sum)
Greenaward certified
Particulars over Time
Average Class
Average Flag
Average Owner
Legal Framework:
Number of LI ratified
(Flag and Owner)
Ship Type 1 – Inspected Ship Type 2 –Not Inspected
Casualty/No Casualty Casualty/No Casualty
110
6.3. Model Assessment and Final Results
The model for very serious casualty was tested for presence of heteroscedasticity using
the LM test as described by Davidson and McKinnon (1993)92. Only the very serious
casualty models and the casualty first event models were tested and not the probability of
detention models since the author felt that it was more important to investigate
heteroscedasticity for the casualty models due to the sensitivity of the topic in question.
The probability of detention models only shows the differences across the regimes while
the present models use similar variables to show a certain amount of causal relationship.
The null hypothesis (Ho) assumes homoscedasticity and the alternative hypothesis
assumes heteroscedasticity in the following form where γ is unknown and z are a number
of variables which are assumed to be the cause of heteroscedasticity:
Variance = exp (2z’γ)
The test was performed separately for two variables, namely tonnage and age where the
test results can be found in the table below and in Appendix 18 for further reference.
Presence of heteroscedasticity was found with both variables.
Table 28: LM Test for Tonnage and Age
Variable LM Statistic p-value
Tonnage 23.06 0.00000 – reject ho
Age 13.60 0.00022 – reject ho
Note: 1% significant level used
To find out if the presence of heteroscedasticity has an effect on the estimators and the
significance of the estimators in the model, a standard program which is part of Eviews
(also refer to Appendix 19 for further reference on the adapted version) and which was
developed by Greene93 and based on Harvey (1976) was used. This program allows
estimation in the presence of the form of heteroscedasticity defined earlier. The
corresponding probabilities are calculated based on Equation 6 where z depicts tonnage
alone or tonnage and age and γ the coefficient for tonnage or age obtained by the Greene
program.
Equation 6: Probability of Casualty allowing for heteroscedasticity
)
( γz )
x β
)
( γz )
x β
i
i
exp
(
exp
(
i
1 e
P e
+
=
The results of the estimation of the model are given in Appendix 19. First, the program
was modified for two variables – tonnage and age. In this first trial, age and tonnage were
used in z but age comes out not to be significant. Therefore, the procedure was applied a
second time without age and the results show that three variables come out not to be
significant in comparison to the original model based on Equation 4.
92 Davidson and McKinnon (1993), Estimation and Inference in Econometrics, New York: Oxford
University Press, 1993, page 526ff
93 Greene H.W. (2000), Econometric Analysis, Fourth Edition, Econometric Analysis, Prentice Hall,
New Jersey; page 518ff; Furthermore, recognition is to be given to Richard Paap from the
Econometric Institute for pointing this program out to the author and for making it available.
111
To see whether the probabilities differ from the original model and if heteroscedasticity,
although present, has a serious effect on the estimation results, the corresponding
probabilities are computed for Equation 6 versus Equation 4. The probabilities are then
grouped according to tonnage groups, age groups, flag state groups, classification groups
and ownership groups and the respective probabilities were calculated based on both
models and visualized in Appendix 20. The results further show little difference between
the two estimation processes. One can therefore conclude that although some presence of
heteroscedasticity is present in the model, it does not have a serious effect on the
estimation process with reference to the coefficients and the resulting probabilities and
the standard model is used and applied to all models in this section. Table 29 lists the key
statistics of the final types of models reduced to 1% significance level for logit and probit
estimation and the regression results and be seen in Appendix 21 to Appendix 24
respectively.
In comparing logit with probit, not much difference can be seen in the results other than
that the HL-statistic suggests a better fit for the logit model versus the probit model. The
results are acceptable for the amount of data in each of the models. The hit rate lies above
74% for all logit models except for the models of the fishing vessels. Outliers were
identified and eliminated to improve the fit of the models. The outliers were not analyzed
in detail since it was not found that eliminating them will significantly change the results
of the regression analysis. This can be left as recommendation for future research. For
visualization of the results in the next chapter, the logit model is used.
Table 29: Key Statistics of Final Models: Probability of Casualty
Very Serious Serious Less Serious Fishing Fleet
0 = 37354 0 = 37811 0 = 37811 0 = 6129
# observations in 1 = 722 1 = 3198 1 = 2118 1 = 160
final model Total = 38076 Total = 41009 Total = 39929 Total = 6289
# of outliers 122 97 35 4
Cut Off 0.0189 0.0780 0.0530 0.0250
LOG PRO LOG PRO LOG PRO LOG PRO
Mc Fadden R2 0.634 0.632 0.281 0.278 0.250 0.224 0.400 0.393
% Hit Rate y=0 93.53 92.60 79.87 78.05 76.09 81.93 89.98 85.51
% Hit Rate y=1 88.64 90.39 74.39 76.92 74.41 68.41 71.25 76.25
% Hit Rate Tot 93.43 92.56 79.44 77.97 76.00 81.21 89.51 85.28
HL-Stat. (df=8) 14.65 48.79 16.92 43.08 17.28 25.93 15.20 19.74
p-value 0.0663 0.0000 0.0309 0.0000 0.0273 0.0011 0.0552 0.0114
6.4. Visualization of Results: Casualty Normal Models
In order to make the interpretation of the results easier, the probabilities are converted
from a six year time frame to a yearly probability for the reason explained earlier. The
conversion factor is derived as follows.
If p is denoted as a 1 year probability, then the probability of having no casualty in six
years equals 1-p for 1 year and (1-p)6 for 6 years. The probability of having at least one
casualty in 6 years denoted by q is then 1-(1-p)6. While q is the result from the models,
solving for p leads for the following factor which is then applied to all calculated
probabilities in this section to convert the six year probability into a yearly probability:
112
p = 1-(1-q)1/6
For the purpose of this study, it is assumed that the probability of having at least one
casualty in a year is constant across the years which might not completely accurate as
this change in probability can be affected by changes the industry such as commercial
surroundings. For the purpose of visualization of the effect of inspection on casualties, it
is assumed to be accurate enough as the change in probability is not the research
question in this study but primarily the effect of inspections on the probability of
casualty. The probability of detention is an average probability based on a ship with no
deficiencies to reflect the basic ship profile of a vessel in order to bring the two into
relation. In order to combine both probabilities, each estimated probability of the total
dataset was combined using the IMO number of a respective vessel. For the probability of
detention, averages per vessel were taken. The result is a comparison of both probabilities
for about 50,110 ships, depending on the size of the actual dataset (e.g. very serious,
serious, less serious casualties).
6.4.1. Overall View of Inspected versus Non-Inspected Vessels
Figure 64 shows the average estimated probability of having a casualty based on the total
dataset for the commercial fleet (split into seriousness) and the fishing fleet above 400gt.
There is no particular reason based on existing legislation for the choice of size with
reference to the fishing fleet94. The analysis is based on the fishing fleet above 400gt to
bring them in line with larger fishing vessels or the so called factory ships. Technically,
the factory ships are not fishing vessels and are therefore sometimes inspected by port
state control (about 0.3% of the PSC dataset) but they belong to the same industry and
are the only ships that can be compared to the larger vessels of the regular fishing
vessels. In order to keep the size similar to the commercial fleet, 400 gt were used as cut
off point.
Figure 64: Average Probability of Casualty
0.006
0.016
0.010
0.007
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0.016
0.018
very serious serious less serious fishing
(>400gt)
Probability of Casualty
Commercial fleet
Note: Based on average estimated probabilities of approx. 50,000 vessels
94 The Torremolinos International Convention for the Safety of Fishing Vessels, 1977 was adopted
in 1977 but is not yet in force and only has a provision based on meters and not gross tonnage,
https://monkessays.com/write-my-essay/imo.org/home.asp
113
The basic probability of a very serious casualty lies by 0.6% versus 1.6% for serious and
1% for less serious casualty. The fishing fleet lies slightly above the very serious casualty.
In reality, this probability is expected to be much higher for fishing vessels below 400gt
but for this analysis, only ships above 400gt are included as explained previously. The
next set of graphs shows the difference between the fishing fleet of 400gt and above which
is very little inspected by port state control and the commercial fleet. It further shows the
main difference between vessels that are inspected by any of the regimes in question
versus not inspected by any of the regimes. Not inspected does not necessarily mean that
the vessel was not inspected by port state control at all.
Due to the lack of cooperation of some of the port state control regimes, a portion of
vessels that are only inspected in these region is missing from the dataset and given the
fact that the descriptive statistic section showed that the South and North China Sea are
high risk areas with respect to loss of life and collision and contact, there are a portion of
ships that fall into the category of non-inspected vessels in this dataset but might have
been inspected for instance by the Tokyo MoU. The same applies to a certain extent to the
Black Sea MoU and the Mediterranean MoU although the amount of vessels that are
inspected by those regimes are also partly covered by the Paris MoU. The results would
have been refined by incorporating this data but will be left as a recommendation for
future research.
Figure 65 gives an overall picture by comparing the fishing fleet with the rest of the
commercial fleet where the probability of casualty is split into type of seriousness for the
commercial fleet but not for the fishing fleet.
Figure 65: Probability of Casualty (Inspected versus Non-Inspected Ships)
0.009 0.009
0.007 0.007
0.004
0.021
0.013
0.004
0.000
0.005
0.010
0.015
0.020
0.025
0.030
very serious serious less serious fishing (>400gt)
Average Probability of Casualty
Not PSC Inspected PSC Inspected
factory ships
commercial fleet
Note: Based on average estimated probabilities of approx. 50,000 vessels
The graph shows that the average probability of a very serious casualty is lower for
inspected vessels than not inspected vessels and for the fishing vessels while it is
substantially higher for serious and less serious casualties. Overall, it seems to show that
ships that are not inspected have a higher probability of a very serious casualty while this
is not the case for the other two casualty types. The partial effects on inspection will be
shown in the next section and confirm this picture.
114
The next graph looks into the flag state groupings and also links it with the probability of
detention obtained in part II and calculated for a basic ship risk profile without
deficiencies on a ship per ship level. It shows that the average probability of casualty is
highest for black listed flag states followed by white and grey listed flag states for the
commercial fleet while it is highest for the fishing fleet for white listed flag states. This
can be explained by the fact that many traditional maritime nations of which their flags
are on the white list maintain fishing fleets but these ships are not inspected very often
or not at all.
The inspected portions are primarily factory ships. As mentioned earlier, the difference
between the commercial shipping fleet and the fishing fleet above 400gt are expected to
be much higher if one would include the total fishing fleet. Nevertheless, it shows that
inspections have a negative effect on the probability of a very serious casualty as well as
for the inspected fishing fleet in general while this effect cannot be seen with serious and
less serious casualties for the commercial fleet.
Figure 66: Commercial Fleet versus Fishing Fleet-Flag State Grouping
0.016
0.013
0.011
0.014
0.008
0.010
0.011
0.007
0.012
0.004 0.004
0.006
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0.016
0.018
0.020
Black Grey White Undefined
Probability of Casualty/Detention
detention average casualty (commercial fleet) fishing (>400gt)
Note: Based on average estimated probabilities of approx. 50,000 vessels, Detention based on ships
with zero deficiencies
Figure 67 shows the probability of casualty per ship type. General cargo ships show the
highest probability for very serious and serious casualties while tankers and containers
show the lowest. It is around 1% for general cargo ships and passenger vessels and 0.1%
to 0.3% for container vessels and tankers.
In order to show the difference between inspected and not inspected ships, Figure 68 to
Figure 70 provide this information in detail. For very serious casualties, one can clearly
see that inspected ships show a lower probability for all ship types. For dry bulk carriers,
the difference seems to be the strongest followed by general cargo ships. For tankers and
containers, the difference is much less. The smaller difference for tankers can be
explained as a combined effect of vetting inspections and port state control while this is
not applicable for container vessels. The category passenger vessels comprise mainly Ro-
Ro Passenger ships and cruise vessels.
115
Figure 67: Average Probability of Casualty per Ship Type
0.011
0.022
0.013
0.004
0.018
0.013
0.001
0.012
0.011
0.003
0.010
0.006
0.010
0.010
0.006
0.005
0.011
0.006
0.000
0.005
0.010
0.015
0.020
0.025
very serious serious less serious
Probability of Casualty
general cargo dry bulk container tanker passenger other ST
Note: Based on average estimated probabilities of approx. 50,000 vessels
Figure 68: Probability of Very Serious Casualty per Ship Type
Very Serious Casualty
0.018
0.006
0.013
0.002 0.002
0.001
0.004
0.002
0.011
0.008
0.005 0.005
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0.016
0.018
0.020
not inspected inspected
Probability of Casualty/Detention
general cargo dry bulk container tanker passenger other ship types
Note: Based on average estimated probabilities of approx. 50,000 vessels
It is difficult to interpret the relative high probability of a very serious casualty for
passenger vessels compared to container ships or tankers. This could be due to the fact
that a casualty on passenger vessels will most likely include a fatality which would
automatically classify as a very serious casualty. In addition, after consulting some
experts, the commercial characteristic of operating in the Ro-Ro segment of the industry
does not necessarily explain this high probability for European Waters but is more likely
to be accepted for other areas where vessels are moved from a stricter area to a less strict
116
area and still continue to operate. Another explanation could be that there are no vetting
inspections onboard Ro-Ro passenger vessels as can be found with tankers and dry bulk
carriers. Finally, Ro-Ro vessels can carry truck with dangerous cargo and the manifest of
the cargo is not known to the crew. Operation is under a tight schedule which can add to
fatigue related issues with crew.
Figure 69 and Figure 70 look at the serious and less serious casualties per ship type
where the picture changes accordingly. On average, inspected ships show a higher
probability of casualty then not inspected ships. It reflects two dimensions. First, the
correct ships seem to be targeted for inspection but the partial effect of the inspection
variables do not indicate a negative relationship which means that the inspections are
less effective or not effective for serious and less serious casualties.
Figure 69: Probability of Serious Casualty per Ship Type
Serious Casualty
0.011
0.029
0.012
0.019
0.006
0.014
0.005
0.016
0.020
0.028
0.010 0.011
0.000
0.005
0.010
0.015
0.020
0.025
0.030
not inspected inspected
Probability of Casualty/Detention
general cargo dry bulk container tanker passenger other ship types
Note: Based on average estimated probabilities of approx. 50,000 vessels
This might be due to the fact that for these two types of casualties, the human element
plays a more important role and that therefore, it is more difficult to incorporate this into
an inspection. General cargo ships show the highest probability of a serious casualty of
almost 3% for inspected vessels and 1% for not inspected vessels. In this category, tankers
also show a higher percentage which might indicate that despite the heavily inspected
tankers for vetting inspections, the amount of inspections does not have a negative effect
on the probability of having a serious casualty but that over-inspection can have an
adverse effect on the human factor such as increased stress and fatigue.
In addition, Figure 70 for less serious casualties is more difficult to interpret. Overall, the
probability is lower but this might be due to lack of reporting in general. It can be
understood as the probability of having a potential serious or very serious casualty which
is highest for passenger vessels compared to other ship types. The difference between
inspected and non inspected vessels is not very large. The next section will look at the
partial effects of inspections on casualties.
117
Figure 70: Probability of Less Serious Casualty per Ship Type
Less Serious Casualty
0.008
0.016
0.011
0.014
0.008
0.013
0.004
0.008
0.019
0.017
0.006 0.006
0.000
0.005
0.010
0.015
0.020
0.025
0.030
not inspected inspected
Probability of Casualty/Detention
general cargo dry bulk container tanker passenger other ship types
Note: Based on average estimated probabilities of approx. 50,000 vessels
6.4.2. The Partial Effects of Inspections on Casualties
Table 30 gives a short summary of some of the results of the variables of interest with
their respective coefficients and significance across the casualty types. The coefficients
are not to be interpreted as direct effects like in linear regression. It is merely the partial
effect of a particular variable given all other variables remaining the same.
The interesting part is not necessarily the coefficient but its significance and sign which
determines the tendency of the effect towards the probability of casualty. A 1%
significance level was used since the author felt that the topic in question is a sensitive
issue and therefore, a smaller significance level was felt to be more appropriate. The table
summarizes the main findings as follows:
o General cargo vessels seem to show the highest risk although not necessarily the
largest costs with respect to the aftermath of for instance pollution deriving from
an oil tanker as shown previously in Chapter 2 of this thesis. Second in line are
passenger vessels but primarily for serious and less serious casualties.
o Age is only significant for very serious casualties and its effect is positive. A
separate graph is shown below to visualize this effect.
o Tonnage is also only significant for very serious casualties but is negative
indicating that a smaller vessel seems to be at higher risk than larger vessels
which goes in line with the general cargo vessels being more high risk prone.
o The coefficient of the variable indicating if the ship changed its class during its
course of life is significant and negative for all types of casualties. This could mean
that in general, if a class is changed, an inspection is performed which might have
a positive influence on the quality of the vessel. On the other hand, the coefficient
of the variable indicating if class was withdrawn certainly shows a positive effect.
118
o Change of flag does not seem to be significant while change of ownership is
significant for all types of casualties but in particular for very serious casualties.
The coefficient indicating if an owner changed is further positive. This could
indicate that due to the move of a vessel to the second hand ownership market, the
money which is normally spent might be less and decrease the overall safety level
of a particular vessel.
o The variable double hull is not significant for any type of casualty.
o As for vetting inspections as part of the industry data, the coefficient of this
variable clearly indicates that the inspections have a strong negative effect on the
probability of a very serious casualty while it is not significant for the other two.
o The last group of coefficients of the variables showing the partial effects of the
inspections of a particular regime shows that the effect is negative for all regimes
for very serious casualties and varies for serious and less serious casualties. For
very serious casualties, the Caribbean MoU had to be excluded due to lack of data.
o Certain classification societies, flag state, ownership variables and ship yards
remain to be significant in the models and can be found in the Appendix 21 to
Appendix 23 in detail.
Table 30: Summary of Main Variables: Casualty Normal Models
Significance at 1% very serious serious less serious
Variable of Interest Coefficient Coefficient Coefficient
Ship Types
General Cargo 1.2507 0.6684 0.8451
Dry Bulk n/s 0.4078 0.8240
Container n/s n/s 0.7236
Tanker n/s n/s n/s
Passenger n/s 0.5917 0.8217
Other Ship Types n/s n/s n/s
Ship Particulars
Age 0.4059 n/s n/s
Tonnage -0.3717 n/s n/s
Class changed -0.6965 -2.1564 -2.0292
Class withdrawn 0.5802 0.6703 0.4396
Flag changed n/s n/s n/s
Owner changed 5.3686 2.4263 2.1583
Legal Instr. ratified: Flag -0.0543 -0.1244 -0.1804
Legal Instr. ratified: Owner n/s -0.0360 n/s
Double Hull n/s n/s n/s
Rightship Inspected -0.9454 n/s n/s
Greenaward Certified n/s n/s n/s
Port State Control Inspected
Paris MoU -0.5443 0.0459 0.0535
Caribbean MoU*) not in model n/s n/s
Viña del Mar MoU -0.4934 n/s n/s
Indian Ocean MoU -2.1760 n/s n/s
USCG -1.4685 0.0295 0.0467
AMSA -1.5010 -0.1724 -0.1594
Other Variables (indicates number of variables left in model)
Classification Societies 1 8 17
Flag States 4 49 58
Ownership Groups 4 4 4
Ship Yard Countries 3 25 26
Note: n/s = not significant at a 1% level, *) Caribbean MoU is not included in the very serious
model due to lack of sufficient amount of data
119
The coefficients of the variables indicating where the ship was inspected were tested
using the Wald Test for testing restrictions95 in order to see if the mean varies across the
regimes. The null hypothesis (ho) for testing the restrictions states that the means do not
vary across the regimes. The results can be seen in Table 31 below for various
combinations of variables. For all three types of casualties, two groupings can be found.
For very serious casualties, the group containing AMSA, the Indian Ocean MoU and the
USCG are similar versus a group containing the Paris MoU and the Viña del Mar MoU.
For serious and less serious casualties, the Paris MoU and the USCG are similar but
different to AMSA.
Table 31: Testing of Restrictions – Inspection Variables
Very Serious
Restrictions/p-value
Serious
Restrictions/p-value
Less Serious
Restrictions/p-value
AMSA=IMOU=USCG=PMOU
(0.000)- reject ho
AMSA=PMOU=USCG
(0.000)- reject ho
AMSA=PMOU=USCG
(0.000)- reject ho
AMSA-IMOU-USCG
(0.2318)- do not reject ho
PMOU=USCG
(0.1526)- do not reject ho
PMOU=USCG
(0.6063)- do not reject ho
PMOU=VMOU
(0.6845)- do not reject ho n/a n/a
Note: Figure in brackets is the p-value of the test, 1% significance level
Figure 71 shows how the probability of casualty changes with the number of inspections
of a certain vessel on average. The effect is clearly strongest for very serious casualties
and very weak to non-existing or positive for serious and less serious casualties.
Figure 71: Average Effect of Inspection across Regimes on Very Serious Casualties
Effect of Inspections on Probability of Casualty
0.15
0.48
0.20 0.17
0.32
0.25
0.72
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6
Number of Inspections
Probability of Casualty
Not Inspected Average Very Serious
Average Serious Average Less Serious
Ship Type: General Cargo
Age: 15 yrs, Tonnage: 6612 gt
Flag: Belize
Class: Romanian Naval
SY: South Korea, Owner: EMN
av. effect very serious
not inspected very serious
av. effect serious
av. effect less serious
Note: the average effect represents the average effect of all regimes
95 based on Wald Test for Testing Coefficient Restrictions, a standard procedure in Eviews
120
For a vessel having been inspected in one of the regimes, the probability of having a very
serious casualty decreases gradually as the number of inspections increases to a
maximum of six inspections within six years. This probability is not based on a yearly
probability but is left as a probability of having a casualty within a six year period in
order to better visualize the change in the probability. It further shows that the
magnitude of the effect decreases as the number of inspections increases. On average
though, and depending on the overall risk profile of the vessel, an inspection can
potentially decrease the probability of a very serious casualty by approx. 5% per
inspection.
Since the partial effects are in comparison to the benchmark of not inspected vessels, the
positive signs for serious and less serious casualties could be explained with the fact that
there is no significant difference between the ships that are not inspected or inspected in
another regime (such as the Tokyo MoU, the Black Sea MoU or the Mediterranean MoU)
and that there is room for improvement.
Figure 72 gives an overview of the effect of age and is based on one particular ship profile
where age is measured in years of age of the vessel. One can clearly see that as the age of
the vessel increases, the probability of having a very serious casualty increases by about
12% over a 35 year period which translates into about 0.35% per year. The age factor
cannot be measured for serious and less serious casualties.
Figure 72: Effect of Age on the Probability of Casualty
Effect of Age on Probability of Very Serious Casualty
0.04
0.16 0.17
0.15
0.14
0.09
0.13
0.11
0.00
0.05
0.10
0.15
0.20
0 5 10 15 20 25 30 35
Age of Vessel in years
Probability of Casualty
Ship Type: General Cargo
Age: 15 yrs, Tonnage: 6612 gt
Flag: Turkey, Class: DNV
SY: South Korea, Owner: TMN
The next area will try to evaluate if the correct ships are targeted for inspection and will
partly built on the previous section. It compares inspected vessels with non inspected
vessels.
121
6.4.3. Assessment of the Target Factor: Can targeting be improved?
This section builds on 5.3. Overview of Inspections and Casualties and looks further at
some of the variables that are used to target ships for inspections such flag, class and
ownership by comparing the probability of detention derived in part II of this thesis with
the probability of casualty derived in this section. First, an overview of the possible
magnitude of improvement for targeting will be provided by providing a combined graph
of Figure 55 and Figure 56 presented at the end of Chapter 5.
Figure 55 has shown that about 2.4% of the world fleet eligible for port state control are
ships that have not been inspected but had a casualty. In addition, 1.7% of the vessels
had a casualty and was inspected without any related time frame while 2.5% of the
vessels were inspected six months prior to a casualty. What is interesting to notice but
which is not shown in this graph is that based on the individual port state control
inspections, 54% of all inspections were inspections with zero deficiencies while when
aggregated by ship and taken as a summary of all inspections performed on vessels, this
percentage reduces to 16% of all inspections for the time frame 1999 to 2004. On the other
hand, if only looking at the last inspection six months prior to a casualty as shown in the
figure with 2.5% of the total world fleet, 52.3% of these vessels were ships with zero
deficiencies.
The portion of ships which have been inspected can be understood as the ships that have
been targeted for inspections of which a certain portion was assumed to be sub-standard.
About 16% of all inspected vessels had zero deficiencies over the time period in question
and these ships might have been ships which should not have been targeted (4,221 ships).
On the other hand, looking at ships which have been inspected six months prior to a
casualty (2,321 ships) where 52.3% of these vessels had zero deficiencies (1,215 ships) and
the rest had deficiencies. This changes the 4,221 ships which should not have been
targeted into 3,006 vessels or approx. 501 ships per year.
It is further worth noticing that out of the 1,106 vessels (2,321 – 1,215) with deficiencies,
14.6% were detained (162 vessels) and had a casualty. This portion could be understood
as ships that have been targeted correctly and identified as sub-standard vessels but for
some reason, detention was not sufficient to increase the safety standard of the vessel to
prevent a casualty. The remaining amount of the vessels which have been inspected and
where deficiencies were found are the vessels where the effect of inspections decreased
the probability of a casualty which is the partial effect of the regressions shown in this
section of the thesis. In number of vessels, this amounts to approx. 18,87496 vessels or
3,146 ships per year. Figure 73 then visualizes the discussion above and presents a
summary of the magnitude of possible improvement areas for port state control.
The figure is only based on ships that are relevant for port state control (excluding the
fishing fleet > 400gt) and is a summary of the total time frame. The graph shows several
groups out of which group 1 of about 36% of the vessels eligible for inspections are
identified not to have been problematic over the time period and have also not been
targeted by the regimes in question. About 7% of the vessels eligible for port state control
have been targeted over the time frame but did not have a casualty and also no
deficiencies and therefore represent a group of over-inspected vessels (group 2).
Group 3 of 43% of the vessels can be identified to belong to a group where inspections are
effective in decreasing the probability of casualty where this effect can be measured for
96 21,880 total inspected ships with no casualty minus 3,006 ships with no deficiencies
122
very serious casualties and estimated (depending on the basic ship risk profile) to be a 5%
decrease per inspection. This category can also represent further room for improvement
but shows that port state control is effective.
Figure 73: Improvement Areas for PSC eligible ships (1999-2004)
improve
targeting &
inspections
3.7%
improve
targeting
4.7%
improve
inspections
4.9%
group 3:
inspections
with effects
43.2%
group 2:
over
targeted
ships
6.9%
group 1:not
problematic
ships
36.5%
Total Ships: 43,817
group 4:
Note: Based on only PSC relevant ships and based on total time frame (1999-2004)
Group 4 is split into three portions. The first portion is 4.9% of PSC eligible vessels which
are the amount of ships that have been targeted correctly but since they had a casualty
within six month after the inspection, the enforcement could be improved. The second
portion shows 4.7% of ships which had a casualty but were not inspected and where
targeting could be improved. Finally, the last category shows a grey area. In this group,
ships had a casualty but regardless of the time frame. Therefore, inspections and possibly
targeting could be improved. Most improvement to decrease the probability of a casualty
can be achieved by concentrating on the categories in group 4 by shifting the emphasis
from group 2 to group 4.
Figure 74 and Figure 75 both show the average probability of a very serious or serious
casualty of flag state groups such as black, grey, white and undefined for flags that are
not on the Paris MoU list. The Paris MoU classification is used since about half of the
data derives from the Paris MoU dataset. The graphs show the difference between
inspected and non inspected ships which is clearer for very serious casualties than for
serious casualties. Black listed flag states have a higher probability of having a very
serious casualty followed by grey and white listed flag states. The same picture applies to
detention (based on ships with zero deficiencies).
For serious casualties, the picture changes and indicates that these vessels show a higher
probability of having a serious casualty. It seems to indicate that these vessels are
targeted for inspection but that there is less effect of an inspection on decreasing the
probability of casualty (as can be observed for the very serious casualties). In order to look
into this topic further, Figure 76 was produced and shows a list of top 30 flag states and
their associated probability of casualty and detention. Flag states with relative high
probability of detention in relation to casualty are Mongolia, Cambodia and North Korea.
123
Figure 74: Probability of Very Serious Casualty per Flag State Group
Very Serious Casualty
(detention based on ships with zero deficiencies)
0.019
0.005
0.016
0.006
0.003
0.013
0.004
0.002
0.008
0.006
0.008
0.011
0.000
0.005
0.010
0.015
0.020
0.025
0.030
not inspected inspected detention
Probability of Casualty/Detention
Black Grey White Undefined
Note: Based on average estimated probabilities of approx. 50,000 vessels
Figure 75: Probability of Serious Casualty per Flag State Group
Serious Casualty
(detention based on ships with zero deficiencies)
0.008
0.020
0.016
0.007
0.011
0.013
0.010
0.024
0.008
0.011
0.033
0.011
0.000
0.005
0.010
0.015
0.020
0.025
0.030
0.035
0.040
not inspected inspected detention
Probability of Casualty/Detention
Black Grey White Undefined
Note: Based on average estimated probabilities of approx. 50,000 vessels
The graph also gives the probability of detention based on ships with zero deficiencies
which serves as a basic risk profile based on the ships. The probability of detention for the
top seven flags is lower than the probability of a very serious casualty and then increases
above it for a group of flag states. The graph is based on flag states with more than 50
registered vessels.
124
Figure 76: List of Top 30 Flag States: Very Serious Casualty
Top 30 Flag States
Commercial Ships > 400gt (Flag States with > 50 ships)
Probability of Very Serious Casualty/Detention
0.000 0.005 0.010 0.015 0.020 0.025 0.030 0.035 0.040
Syria
Belize
St Vin. & Grenad
Lebanon
Honduras
Barbados
Nigeria
NorthKorea
Cambodia
Albania
Turkey
Bulgaria
Cyprus
Mongolia
Romania
Bermuda
Philippines
Brazil
Thailand
Denmark
Egypt
Panama
Viet Nam
Croatia
Canada
Lithuania
Spain
Kuwait
Malta
Very Serious Casualty
Detention
Note: Based on average estimated probabilities of approx. 50,000 vessels
Figure 77 shows the picture for the fishing fleet above 400gt. The graph shows the
probability of casualty based on the fishing fleet in comparison to the commercial fleet as
well as detention. Flags with more than 50 ships are taken into consideration. For the
probability of casualty, very serious casualties have been chosen since the author felt it
would be more appropriate to compare this casualty with the probability of casualty of the
fishing fleet. Detention in this graph is merely detention based on commercial vessels
with the same flag and not fishing vessels.
125
Figure 77: List of Fishing Fleet (>400gt, more than 50 ships)
Fishing Fleet above 400gt (Flag States with >50 ships)
Probability of Casualty/Detention
0.000 0.010 0.020 0.030 0.040 0.050
France
Canada
NewZealand
Netherlands
Belize
Namibia
UK
Norway
China
Argentina
Denmark
USA
Panama
Spain
RussianFed
Honduras
Iceland
Chile
Japan
SouthKorea
Taiwan
Ukraine
SouthAfrica
Morocco
Vanuatu
Mexico
Italy
FaroeIslands
Portugal
Peru
Ecuador
Philippines
Indonesia
Casualty Fishing Fleet
Casualty Commercial Fleet
Detention Commercial Fleet
Note: Based on average estimated probabilities of approx. 50,000 vessels
126
The graph shows the main difference between the two industries where one is prone to be
inspected and one is not or barely inspected (only factory ships) and that the flag by itself
in this case cannot solely be the indicator of a high risk vessel since significant differences
between the probability of casualty of the fishing fleet versus the commercial fleet under
the same flag can be observed. For most countries, the probability of casualty of the
fishing fleet is higher than for the commercial fleet. In reality this difference is expected
to be higher when one would include the vessels below 400gt.
Figure 78 lists the top 25 classification societies and their probability of a very serious
casualty and detention. In this graph, the difference between the two probabilities is
much higher than with the flag states. Only three classification societies on this list have
a count of below 50 namely, RINAVE, Panama Reg. Corp. and Romanian Naval. These
three should therefore be interpreted with caution.
Figure 78: Probability of Casualty per Classification Society
0.000 0.005 0.010 0.015 0.020 0.025 0.030 0.035
Romanian Naval
Hellenic
China Corp
Turkish Lloyd
RINAVE
Panama Reg. Corp
No Class
Vietnam RS
Bulgarski Koraben
BureauVeritas
RINA
Russian River
Korean South
Russian MS
Croatian RS
Lloyds UK
PolskivReSt
Germanischer Lloyd
DNV
ABS
NKK Japan
Biro KlasIndo
China Class
Inter Nav SurB
OtherClass
Commercial Ships >400gt (Top 25 Classification Societies)
Probability of Very Serious Casualty/Detention
Very Serious Casualty Detention
Note: Based on average estimated probabilities of approx. 50,000 vessels
127
The overall picture is given in Figure 79 which is not split into inspected versus non
inspected groups. For the graphs of the classification societies, the interpretation has to
be careful due to the lack of some data. From the total casualty dataset, about 25% of the
classification societies were missing which is indicated with unknown in the graph. Most
of these observations are assumed to be Non-IACS class versus IACS class since one of
the data providers does not indicate Non-IACS class. In general, ships classified with
IACS class show a lower probability of casualty for all three categories and detention
versus Non-IACS class.
Figure 79: Probability of Casualty and Class Groups
0.005
0.009
0.006
0.009
0.008
0.019
0.012
0.017
0.008
0.005
0.014
0.009
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0.016
0.018
0.020
very serious serious less serious detention
Probability of Casualty/Detention
IACS NIACS Unknown
Note: Based on average estimated probabilities of approx. 50,000 vessels
In addition, the casualty and inspection data did not indicate the various types of
classification societies a vessel can have which could be up to 4. The analysis is only
based on the main class at the time of inspection or casualty which is primarily the
classification society the vessel is classed in and does not necessarily has to be the
classification society of ISM or ISPS97. An analysis which reflects the different types of
classification societies per vessel are left as a recommendation for further research.
Figure 80 and Figure 81 look at the ownership of a vessel and it is interesting to notice
that the highest probability of a very serious casualty lies with owners from open
registries followed by unknown ownership, traditional maritime nations and emerging
maritime nations. The difference between inspected and non-inspected vessels is also
clear while traditional maritime nations show the lowest probability for inspected vessels
which is in line with the probability of detention. For serious casualties, the order
changes as well as the difference between the groups. Appendix 25 shows the probability
of a very serious and serious casualty per DoC company country of residence. Since not
enough data was available in the casualty dataset, it is based on the inspection dataset
and therefore only shows inspected vessels. The graph is only based on countries that
showed more than 100 ships. It gives further an indication on how the probability of
97 ISPS for the purpose of this analysis is indifferent since security is not included in this thesis.
128
casualty can be linked to ownership. Not all regimes have this variable incorporated into
the target factor to target vessels for inspection. More detailed research in this area is
recommended. At this stage, company data was available but not used for the purpose of
this study but the country of residence was used instead.
Figure 80: Probability of Very Serious Casualty per Ownership Group
Very Serious Casualty
(detention based on ships with zero deficiencies)
0.007
0.003
0.014
0.010
0.003
0.008
0.015 0.013
0.020
0.047
0.010
0.015
0.000
0.005
0.010
0.015
0.020
0.025
0.030
0.035
0.040
0.045
0.050
not inspected inspected detained
Probability of Casualty/Detention
Emerging MN Traditional MN Unknown Open Registries
Note: Based on average estimated probabilities of approx. 50,000 vessels
Figure 81: Probability of Serious Casualty per Ownership Group
Serious Casualty
(detention based on ships with zero deficiencies)
0.006
0.013 0.014
0.017
0.025
0.008
0.003
0.006
0.021 0.020
0.036
0.015
0.000
0.005
0.010
0.015
0.020
0.025
0.030
0.035
0.040
0.045
0.050
not inspected inspected detained
Probability of Casualty/Detention
Emerging MN Traditional MN Unknown Open Registries
Note: Based on average estimated probabilities of approx. 50,000 vessels
129
Figure 82 shows the probability of casualty for ships that have not been inspected versus
ships that have been inspected and detained (or not detained). It is based on an average
probability and not split into seriousness. In general, ships not inspected by port state
control show a lower probability of casualty while ships that have been inspected and
detained show a higher probability. Detained vessels show the highest probability of a
casualty. The graph confirms that on average, detained vessels seems to be more risk
prone towards the probability of casualty but it also shows that detention by itself does
not seem to have a negative effect on the probability of casualty. A possible room for
improvement is therefore the follow up on detentions and deficiencies such as maybe the
correct implementation of the safety management onboard. The partial effect of detention
is analyzed in the detailed regressions in Chapter 7: Probability of Casualty-Refined
View.
Figure 82: Probability of Casualty of Detained versus not Detained Vessels
0.011
0.016
0.008
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0.016
0.018
Not Inspected Inspected Not
Detained
Inspected Detained
Probability of Casualty
Note: Based on average estimated probabilities of approx. 50,000 vessels
Figure 83 gives the average probability of all ships that were certified by the Greenaward
Foundation versus ships that were not certified by the Greenaward Foundation and
Figure 84 shows the same picture based on ships that have been inspected by Rightship
(vetting inspections). For Greenaward certified ships, this applies for oil tankers while
Rightship inspected primarily bulk carriers and some oil tankers.
Figure 83: Probability of Casualty: Greenaward Certified (oil tankers)
Very Serious Casualty
0.0061
0.0003
0.000
0.001
0.002
0.003
0.004
0.005
0.006
0.007
Greenaward not
certified
Greenaward
certified
Probability of Casualty
Note: Based on average estimated probabilities of approx. 50,000 vessels
130
Both graphs show a difference in the probability of a very serious casualty for ships that
have not been certified (inspected) versus ships that have been certified (inspected).
While the partial effects will be looked at in the next chapter in detail, this should provide
some indication on possible improvements for targeting vessels since it can easily be
incorporated into targeting matrices of port state control.
Figure 84: Probability of Casualty: Rightship Inspected (bulk carriers and oil tankers)
Very Serious Casualty
0.0133
0.0044
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
RS not inspected RS inspected
Probability of Casualty
Note: Based on average estimated probabilities of approx. 50,000 vessels
Figure 85 brings the probability of a casualty in relation to the number of deficiencies a
vessel can have during an inspection.
Figure 85: Probability of Casualty and Number of Deficiencies
0.013
0.015
0.013
0.008
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.014
0.016
ZeroDef 1 to 5 6 to 10 > 10
Number of Average Deficiencies per inspection (1999 to 2004) Probability of Casualty
Note: Based on average estimated probabilities of approx. 25,800 vessels (all inspected vessels)
131
One can see that overall, the probability of having a casualty increases as the number of
deficiencies increases to a peak of 6 to 10 deficiencies per inspection. This graph is based
on the total time frame (1999 to 2004) and all port state controlled inspected vessels. It
shows that port state control does find deficiencies on high risk vessels and that a vessel
with more deficiencies also has a higher probability of casualty with a maximum of 10
deficiencies per inspection. The probability then decreases slightly which can be
interpreted with the fact that vessels with more deficiencies are more likely to be
detained and will have to deal with the rectification of the deficiencies. The average
probability of detention (based on results from Chapter 4 of this thesis) for ships with
more then 10 deficiencies lies at 0.45 compared to 0.13 for an average of 5 to 10
deficiencies per inspection.
6.5. Summary of Findings: Casualty Overall View
About 32% of all port state control relevant casualties between 1999 to 2004 show signs of
a casualty first event in engine related areas. Per ship type and for the time period 1993
to 2004 in aggregated form and regardless of port state control relevance, general cargo
ships are leading with 37.9 % followed by dry bulk (15.7%), oil tankers (11.4%) and
container ships (7.3%). The fishing fleet shows a higher amount of percentage with 6.9%
to the total. Passenger vessels including hydrofoils and ferries are by 9.9%. With respect
to the casualty locations, a high risk area is West Africa, the Indian Ocean region, the
North Atlantic East region and the South China Sea.
Improvement for port state control has been identified in the area of targeting and
possibly the inspections itself. About 36% of the vessels eligible for inspections are
identified not to have been problematic over the time period in question and have also not
been targeted by the regimes in question. About 7% of the vessels eligible for port state
control have been targeted over the time frame but did not have a casualty and also no
deficiencies and therefore represent a group of over-inspected vessels where intervals of
inspections could be increased.
About 43% of the vessels can be identified to belong to a group where inspections are
effective in decreasing the probability of casualty where this effect is strongest for very
serious casualties and estimated (depending on the basic ship risk profile) to be a 5%
decrease per inspection. This category can also represent further room for improvement
but shows that port state control is effective. Finally, about 4.9% of PSC eligible vessels
have been targeted correctly but since they had a casualty within six month after the
inspection, the enforcement could be improved. Another portion of 4.7% of ships had a
casualty but was not inspected. This is an area where targeting could be improved.
The mean amount of deficiencies found at least six month prior to a casualty lies by 4.3
for black listed flag states versus 2.7 for grey and 1.7 for white listed flag states. Per
seriousness of casualty, detained vessels show significantly higher amount of deficiencies
(17 deficiencies for very serious) versus not detained vessels (3 deficiencies).
Regression analysis based on inspected and not inspected vessels reveals that the average
probabilities of a casualty are by 0.06% (very serious), 1.6% (serious) and 1% (less serious)
compared to 0.07% for the fishing fleet above 400gt. Comparing an industry that is hardly
or not inspected (the fishing fleet) with commercial fishing, one can see that performance
of flag states changes accordingly. Flag states that are normally exposed to inspections
under the port state control regime seem to perform better than if they are only exposed
132
to their own flag state inspections which can be seen with the white listed flag states
showing the highest probability of casualty under the fishing fleet.
With reference to ship types, general cargo vessels seem to show the highest probability of
a very serious and serious casualty of 1% and 2% respectively compared to tankers with
below 0.5% for very serious casualties and 1% for serious casualties. Container vessels
show the lowest risk in all three types of casualties. Comparing ships that have been
inspected with ships that have not been inspected, the strongest difference can be seen
with general cargo vessels and dry bulk carriers for very serious casualties and least of
the effect can be seen with tankers and container vessels. Vetting inspections for certain
ship types such as oil tanker and dry bulk carriers both show a negative effect on the
probability of casualty.
With respect to the target factor, one can confirm that age is significant for very serious
casualties and its effect is positive. One can clearly see that as the age of the vessel
increases, the probability of having a very serious casualty increases by about 12% over a
35 year period which translates into about 0.35% per year. The average probability based
on all ships changes according to ship type and general cargo ships and passenger vessels
seem to show much of the variation with respect to age. Tonnage is also only significant
for very serious casualties but is negative indicating that a smaller vessel seems to be at
higher risk than larger vessels which goes in line with the general cargo vessels being
more high risk prone.
The probability of casualty changes per ship type and confirms that general cargo vessels
are ships with the highest probability of a casualty which is confirmed by the probability
of detention. Black listed flag states or non inspected ships show a higher probability of a
very serious casualty compared to grey and white listed flag states while the same does
not hold for serious and less serious casualties. It might indicate that the target factors
are targeting high risk vessels but are less effective in decreasing the probability of a
serious and less serious casualty. This is confirmed with the partial effects which are only
negative for very serious casualties.
With respect to classification societies, the probability of a casualty for Non-IACS class is
higher than for IACS class and is also confirmed by the probability of detention. For
ownership groups and in comparison of inspected to non–inspected vessels, highest
probability of casualty lies within owners from open registry countries followed by
unknown owners, owners from traditional maritime nations and emerging maritime
nations. This does not follow the probability of detention where owners from traditional
maritime nations show the lowest probability of detention.
Detained vessels show the highest probability of casualty compared to vessels that have
been inspected but not detained and vessel that have not been inspected. The average
probability of a very serious casualty of ships that are Greenaward certified or have been
inspected by Rightship is lower than for non-certified or non-inspected vessels. A variable
indicating both can be easily incorporated into target factors of port state control. In
addition and based on average probabilities, the probability of casualty increases from
about 0.1% to 0.2% as the number of deficiencies increases. This is irregardless of the
time frame of the inspection and is based on the total inspection time frame of about six
years.
The variable indicating if the ship type changed its class during its course of life is
significant and negative for all types of casualties. This could mean that in general, if a
133
class is changed, an inspection is performed which might have a positive influence on the
quality of the vessel. On the other hand, the next variable indicating if class was
withdrawn shows a positive effect which means that these types of vessel have a higher
probability of having a casualty.
Change of flag does not seem to be significant while change of ownership is significant
and positive for all types of casualties but in particular for very serious casualties. This
could indicate that a change of the vessel into second hand ownership or into a segment of
the market where less money is spent on safety increases the probability of a very serious
casualty. The variable double hull is not significant for any type of casualty.
The next chapter will look at a more detailed view of the effect of inspections themselves
on the probability of having a casualty as well as port state control deficiencies in relation
to casualty first events.
134
135
Chapter 7: Probability of Casualty-Refined View
This chapter further refines the general casualty models and tries to measure the actual
effectiveness of some of the inspection variables on the probability of having a casualty by
comparing ships that have been inspected with a casualty with ships without a casualty
and based on their ship history. It will further also look at deficiencies that were found
during inspections in relation to casualty first events. The first part will explain the
criteria and decision processes for the preparation of the dataset and the second part will
explain the model used for the various types of regressions and visualization of the
results.
7.1. Description of Methodology for Data Preparation
Figure 86 provides an overview of the steps that were taken in order to perform the
analysis which will be described in the next chapter. Four steps could be identified and
will be explained in the chapter to come in detail. The graph should merely give an
overview to the reader in order to provide easier guidance to follow through the whole
process.
Figure 86: Description of Methodology Used
The first step is the same as mentioned in the previous chapter and will not be explained
again in detail. The bases for the casualties are all ships with port state control relevant
casualties – a total of 6,005 ships where the fishing fleet is not taken into consideration
here.
7.1.1. The Selection of Relevant Datasets to be Used
The second step was then to link the ships with a casualty to the ships without casualty
and 3,956 ships emerged to show a connection based on the IMO number and were taken
into further consideration. These vessels provide the basis for several merges within a
certain time frame in order to gain a better overview of the connection between casualties
and port state control inspections. The following results are listed in Table 32 below.
Step 1
Step 2
Step 3
The selection of PSC relevant casualties and the
classification of the cases by seriousness according to
IMO classification
The selection of relevant datasets of inspections and
casualties based on a time frame of six months and no
time frame
The selection of the variables that are used to match
ships for the datasets identified under step 2 and the
execution of the matching.
Step 4
The preparation and calculation of the variables used in
the regression and based on the matches found under
step 3.
136
Table 32: Datasets for Port State Control and Casualty Merges Performed
Datasets
Criteria Used for Data
Links
Total
Inspections
Total
Casualties
Total
Records
Total
Ships
Set 1 Inspections with no casualties 148,557 Nil 148,557 21,880
Set 2A Time Frame: 6 months 23,401 2,921 26,322 2,321
Set 2B No Time Frame 33,974 4,737 38,711 3,956
The datasets show three scenarios where set 2B does not contain any particular time
frame between an inspection and a casualty. The remaining set is based on a minimum of
at least one connection of an inspection prior to a casualty of 6 months. After this merge
was performed, the remaining inspections and if applicable casualties of a particular
vessel were added in order to account for the whole inspection history of a vessel.
7.1.2. Methodology Used to Match Ships
Step 3 is the starting point to create a dataset which is then used in the analysis. The
idea is to match ships from the datasets described earlier, and which had casualties, with
ships that did not have casualties. In order to match ships from each of the groups, a set
of variables had to be identified to guarantee the best possible match.
These variables are assumed not to have a direct impact on the seriousness of a casualty
and are listed in order of importance given the fact that the difference of observations in
the datasets is quite large. In doing the match, the first three variables listed in Table 33
are the most important ones followed by the country the ship was constructed and the
owner and then the remaining variables such as class, flag and hull details for tankers.
The match is performed in order to obtain a more refined view of the variables of interest
(the variables which describe the quality of an inspection). In this way, twin ships are
compared and variables which normally would also have an effect towards the probability
of casualty such as e.g. age, ship type, flag, owner or class are cancelled out and are not
interpreted in the end models.
Table 33: List of Variables used to Match Ships
1. Ship Type at the time of construction
2. Year Built (in 11 ranges)
3. Gross Tonnage (in 44 ranges)
4. Country of Owner at time of construction
5. Country where Ship was primarily built
6. Class at construction
7. Flag at construction
8. Double Hull
Ship type is found to be the most important variable for determining the construction
quality and operating environment of a ship. Out of the total 25,836 ships from the
inspection dataset, 1546 ships were converted or alternated since construction and 290
ships changed their ship type completely.
To merge the ships, these changes were taken into consideration and the results can be
seen in Table 34. To facilitate the matching, the year built and gross tonnage was split
into groups which represent ranges and can be seen in Table 35 and with their respective
counts of ships in each of the groups and datasets. For the construction year, 11 groups in
137
ranges of five to ten years where identified and for gross tonnage, 44 groups were created
(ranges of 1000 gt to 5000 gt).
Table 34: Ship Type Groups for Matching
Ship Count/Group
Groups Ship Types Total Ships
Casualties &
Inspections
Inspections
Only
1 container 2049 309 1740
2 dry bulk 4170 575 3595
3 general cargo 9236 1725 7511
4 other 3713 431 3282
5 passenger 865 210 655
6 tanker 5803 706 5097
Total Ships 25836 3956 21880
Table 35: Year Built Ranges for Matching
Ship Count/Group
Groups
Year Built
Ranges Increment Total
Casualties
& Insp.
Inspections
Only Ship Age
1 1879-1939 28 6 22 70
2 1940-1949 10 years 22 3 19 60
3 1950-1959 172 36 136 50
4 1960-1969 1126 187 939 40
5 1970-1974 5 years 1821 290 1531 30
6 1975-1979 3989 723 3266 25
7 1980-1984 4424 800 3624 20
8 1985-1989 3179 504 2675 15
9 1990-1994 2984 472 2512 10
10 1995-1999 4131 615 3516 5
11 2000-2004 3960 320 3640 0
Total Ships 25836 3956 21880
Table 36: Tonnage Ranges for Matching
Count
Groups Ranges From-To Increment Total Casualties Inspections
0 below 1000 plus 1000 2139 217 1922
1 1001 to 2000 2287 444 1843
2 2001 to 3000 2204 363 1841
3 3001 to 4000 1369 215 1154
4 4001 to 6000 plus 2000 1866 295 1571
5 6001 to 8000 1202 207 995
6 8001 to 10000 966 171 795
7 10001 to 13500 plus 3500 1550 243 1307
8 13501 to 17000 1778 301 1477
9 17001 to 20500 1318 213 1105
10 20501 to 24000 1208 213 995
11 24001 to 27500 1204 155 1049
12 27501 to 31000 1012 129 883
13 31001 to 34500 387 70 317
14 34501 to 38000 891 153 738
15 38001 to 42000 plus 4000 1016 130 886
138
Table 36 continued Count
Groups Ranges From-To Increment Total Casualties Inspections
16 42001 to 46000 316 38 278
17 46001 to 50000 261 39 222
18 50001 to 54000 447 84 363
19 54001 to 58000 378 46 332
20 58001 to 62000 154 21 133
21 62001 to 66000 166 13 153
22 66001 to 70000 140 14 126
23 70001 to 74000 86 22 64
24 74001 to 78000 250 38 212
25 78001 to 82000 247 29 218
26 82001 to 86000 145 11 134
27 86001 to 90000 163 11 152
28 90001 to 95000 plus 5000 141 22 119
29 95001 to 100000 29 4 25
30 100001 to 105000 20 3 17
31 105001 to 110000 19 1 18
32 110001 to 115000 36 2 34
33 115001 to 120000 13 2 11
34 120001 to 125000 7 1 6
35 125001 to 130000 12 1 11
36 130001 to 135000 22 2 20
37 135001 to 140000 17 1 16
38 140001 to 145000 18 0 18
39 145001 to 150000 50 5 45
40 150001 to 155000 53 6 47
41 155001 to 160000 149 18 131
42 160001 to 165000 77 3 74
43 165001 to 170000 and above 23 0 23
Total Ships 25836 3956 21880
Using the three first variables and applying it to the datasets explained earlier, the table
below gives an indication on the number of groups of the respective sets (2A and 2B) to be
matched with the groups of set 1 (the inspected ships with no casualty).
Table 37: Overview of Groups used for Matching Ships per Dataset
Dataset Total Records Groups to Match
Set1 (inspected only) 148,557 1133 groups
Set2A (with casualty – 6m) 26,332 610 groups
Set2B (with casualty – total time frame) 38,711 764 groups
Note: based on ship type, age and tonnage groups
The remaining variables derive from data received by Lloyd’s Register Fairplay98 and are
used to further refine the match. The matching was performed between Set 1 and Set2B
using Oracle99 and following the methodology which is visualized in Figure 87. Set 2A is a
subset of Set 2B and is then extracted from the result of the basic match performed on Set
98 A custom made match was performed by Lloyd’s Register Fairplay to match the IMO numbers of
the inspection/casualty dataset with vessel details at the time of construction for owner, class, flag
and ship type.
99 Credit is given here to Ratan Singh Ratore who Helped the author in performing this match by
providing the necessary software (Oracle) and SQL statements to execute the queries.
139
2B. The match was performed in two rounds where double matches for ships with
casualties are allowed out of the dataset of ships with inspections only. The first round
matches the vessels on the three basic criteria listed in the figure namely: ship type, age
and tonnage. From the 21,880 available ships, 17,727 were used to match and 4,153
records were not used.
Figure 87: Visualization of Matching Methodology (per Ship)
This is then followed by a refinement of the matches using the remaining criteria in order
of importance with weighted points. The basic match comes up with 297,532 matches for
3,916 ships with casualties. The refinement match based on ownership (5 points), the
ship yard country (4 points), the class at construction (3 points) and the flag state at
construction (2 points) reduces the basic matches to 122,295 hits with 3,769 ships with
casualties by using a degree 8 to match ships. The decision to allocate points is not based
on any empirical evidence of the impact of the variables on the construction quality. It is
based on the author’s understanding of the shipping industry and partly derived from
interviews100 with surveyors who have experience with new buildings, naval architects
and one of the ship owner’s associations101.
Degree 8 means that the matching ships have both eight points out of the total of 15
points which would be a perfect twin. Several scenarios were run using various degrees of
matching before a decision was made to use degree eight for the analysis and the results
are shown in Table 38. The table lists the degrees of matches (as total points of the point
allocated per additional criteria), the matches of ships with inspections, the number of
ships with casualties that were used and not used and the % of casualties that are lost to
the total casualties (3,956).
The last column indicates which of the scenarios is then accepted for the analysis. Degree
eight was chosen because it provides a balanced result of loosing 4.7% of the cases with
casualties by having a matching degree of 8 points out of 15 total points where eight
100 For detailed list of interviews performed by the author can be found in the Bibliography.
101 Dutch Royal Ship Owner’s Association
+
3 Main Criteria
for Basic Match:
Ship Type
Age Group
Tonnage Group
Round 1: 297,532 matches with 3,916 casualties
Degree 8: 122,295 matches with 3769 casualties: 187 lost
Round 2: 2,491 matches with 147 casualties: 40 lost
Set 1: inspections only
21,880 ships
(17,727 used)
Set 2B: with casualties
3,956 ships
(3,916 used)
5 Secondary Criteria: points Rd1 points Rd2
Owner at construction 5 1
Shipyard Country 4 1
Class at construction 3 1
Flag at construction 2 1
Double Hull 1 1
140
points means that at least three out of the five additional criteria are matched besides the
three basic criteria.
With degree 8, 187 ships did not have a corresponding ship. Since 23 cases are very
serious casualties and 115 cases serious casualties, a second round of matching was
performed on these ships by using a simplified point system for the remaining criteria (1
point instead of the weighted point system). The resulting basic match based on degree 1
(at least two more variables match besides the three basic ones) reveals 2,491 hits with
147 ships with casualties.
Table 38: Summary of Matches by Degrees for Round 1(by Ship)
Degrees
of Match
Ships with
Inspections
Matched
Ships with
Casualties
Used
Ships with
Casualties
Not Used
% Ships Lost
to
Total (3,956)
Dataset
Used for
Analysis
15 6,840 1,757 2,199 55.5% Yes
14 n/a 1,788 2,168 54.8% No
13 n/a 2,829 1,127 28.5% No
12 n/a 3,021 935 23.6% No
11 n/a 3,166 790 19.9% No
10 47,768 3,394 562 14.2% No
9 106,658 3,720 236 5.9% No
8 122,295 3,769 187 4.7% Yes
7 124,172 3,781 175 4.4% No
A total of 40 ships do not have any corresponding vessel out of the inspection dataset.
From these 40 ships, 6 had a very serious casualty, 24 a serious casualty and 10 a less
serious casualty. These 40 ships are not used in the analysis. The final results of the
match are presented in Table 39 and are based on ship counts (a ship can have several
casualties). The column indicating the ships with casualties lost is based on the number
of ships with casualties as listed in Table 32 and the actual number of ships that found
matches. This match provides the basis for the models that will follow. In addition to the
six months time frame, a separate match was performed based on perfect twins instead of
degree 8 in order to see if enough observations are available to use this dataset. From the
total amount of observations of 8,597 cases in comparison to 77,599 cases, one can easily
see that the amount of observations is limited. Therefore, it is decided to only use the
dataset based on degree 8 and with the time frame of six months prior to a casualty.
Table 39: Summary of Matched Datasets (by Ship)
Final Datasets
Ships with
Inspections
Round 1 & 2
Ships with
Casualties
Round 1 & 2
Ships with
Casualties
Lost
Total
Dataset
Cases
No Time Frame 124,786 3,916 40 128,702
6 months 75,302 2,297 24 77,599
Perfect Twins 6,840 1,757 n/a 8,597
Table 40 gives an overview of the types of models that are used. The table lists the types
of models with the total amount of regressions in each series, the datasets on which the
models are based on and the variables that are of interest in each of the regressions. Type
I model is based on the seriousness of casualty and should give an overview of the
effectiveness of an inspection (per regime) on the probability of either a very serious,
serious or less serious casualty. It is a further detailed view on the overall effect that was
141
given by the casualty models in Chapter 6 and in comparison to ships that were not
inspected.
Type II is an aggregated model which uses multiplicative dummy variables for the
deficiencies and ship types in order to see if there are any significant differences with
respect to deficiencies found before an inspection and the ship types. It can be seen as a
further refinement of the type I models but only for very serious and serious casualties.
Table 40: List of Twin Models and their Variables of Interest
Model Name Types of Model/Data based on Variables of interest
Type I Models
(Total Models: 3)
Based on twins of ships inspected six
months prior to a casualty
Types are as follows*):
o Very serious (5,826)
o Serious (45,486)
o Less serious (27,411)
Time in-between inspection
Detention
Deficiencies (less serious)
Regimes (overall view)
Vetting Inspections
Type II Model
(Total Model: 1)
Based on twins of ships inspected six
months prior to a casualty but combined of
very serious and serious casualties in order
to increase the amount of observations. The
deficiencies are multiplied by ship types
and used as multiplicative dummies
Total # of observations: (52,150)
Time in-between inspection
Detention
Deficiencies per ship type
Regimes (overall view)
Type III Models
(Total Models: 5)
Based on casualties of a respective first
event as identified in Chapter 5 in relation
to the deficiencies found during a PSC
inspection. Types are as follows:
o Fire & Explosion (6,218)
o Wrecked/Stranded/Grounded (19,131)
o Collision/Contact (23,254)
o Deck Related First Events (8,357)
o Engine Related First Events (27,079)
Time in-between inspection
Detention
Deficiencies (refined view)
Regimes (overall view)
*) Note: the numbers in brackets are number of observations in the model
The Type III models are based on the casualty first events identified in 5.1. Selection of
Port State Control Relevant Casualties and is a first attempt to link the deficiencies with
casualties. The next section will explain the variables and the base model itself which is
then applied according to the Type I, II or III models.
7.2. Explanation of Variables and Base Model Used
7.2.1. Variables Used for the Regression
The variables listed in Table 41 are a summary of the variables that are used in the
regressions. The variables are split into two blocks where Block 1 contains the variables
which are normally used to target vessels such as the ship type, the classification society,
the flag state and the ownership of a vessel and Block 2 provides a summary of the
inspection history of a particular vessel including information on industry inspections
(vetting inspections of Rightship and Greenaward).
Within Block 1, changes in any of the variables since the construction of the vessel and
during the years of inspection history are identified (e.g. the ship type was converted,
142
flag, class or ownership changed). This block also includes information the number of
legal instruments a certain flag or country of residence of an owner has rectified.
Table 41: Variables Used in the Twin Regressions (Type I, II and III models)
Dependent Variable 1: Casualty:
This can be either per seriousness or by casualty first event 1/0
Independent Variables
Number of
Variable nℓ
Remark on
Variable
Expected
Sign
ℓ Block 1: Ship Particulars: included to account for target factors
Ln(Age) 1 Average age at Inspection 1 C
Ln(SIZE) 2 Gross Tonnage 1 C
ST 3 Ship Type at present 6 D
STInd 4 Ship Type Changed 1 D
CL 5 Classification Society at inspection 33 D
CLInd 6 Classification Society changed 1 D
CLWdr 7 Class Withdrawn 1 D
FS 8 Flag State at inspection 81 D
FSInd 9 Flag State Changed 1 D
OWN 10 Owner of vessel 6 D
OWNInd 11 Ownership changed 1 D
LIOWN 12 Legal Instruments Rectified (Owner) 1 C
LIFS 13 Legal Instruments Rectified (Flag) 1 C
DH 14 Double Hull 1 D
Block 2: Inspection History: variables of interest
RS 15 Rightship Inspected (5 Star Rating or indicator) 5 D neg
GR 16 Greenaward Certified 1 D neg
ln(TIME) 17 Time in between inspections (days) 1 C neg
PSC 18 Inspections Frequency per Regime (Fractions) 6 D neg
DETPS 19 Detention Frequency per Regime 6 D neg
CODE 20 Deficiency main codes (also multiplied by ST) 26 (156) C und
Total Variables*) 181(311)
*) in brackets indicates number of multiplicative dummy variables
C= Continuous, D= Dummy
Since the whole inspection and casualty history of a particular vessel is taken into
consideration, average percentage fractions over all records of one particular vessel
(aggregated by IMO number) are used in the regressions for the inspections and the
detentions while the deficiencies are aggregated and represent a total sum. In addition
and depending on the final model, the deficiency variables can be multiplied by ship types
and increases the amount of variables accordingly which is shown in brackets in the table
above.
Figure 88 visualizes the variable structure of the twin models by following a time line of a
pair of vessels over its course of life. It further shows the difference of the combination of
variables compared to the casualty models in Chapter 6 where ships were either
inspected or not inspected. In these models, all ships are inspected. The variables of
interest are the inspection variables. In addition, the variable indicating where the ship
was inspected is used as a sum in Chapter 6 while in these models, it is a percentage
fraction over all inspections that were performed on a vessel during the time period in
143
question (1999 – 2004). It is believed that this more accurately splits the total inspection
exposure across the regimes versus just adding the inspections up.
Figure 88: Visualization of Variable Structure: Twin Models
Note: Variables of interest are in italic
The model to estimate the effect of inspections on casualties is given in Equation 7 where
the term xiβ can change accordingly to the casualty model in question and is shown in
Equation 7 in detail and based on the variables listed in Table 41.
Equation 7: Detailed Effect of Inspection on Casualties
x β)
x β)
i
i
P (
(
i 1 e
e
+
=
The next section will provide a summary of the datasets and the results of each of Type I,
II and III models. The visualization and interpretation part is then provided in
Different Life Events:
Change of ST
Change of Flag
Change of Owner
Change of Class
Class Withdrawal
Vettings Inspections
PSC Inspections (Fraction)
Greenaward certified
Detention
Deficiencies
Time-inbetween inspections
Particulars over Time
Average Class
Average Flag
Average Owner
Legal Framework:
Number of LI ratified
(Flag and Owner)
Different Life Events:
Change of ST
Change of Flag
Change of Owner
Change of Class
Class Withdrawal
Vetting Inspection
PSC Inspection (Fraction)
Greenaward certified
Detention
Deficiencies
Time-inbetween inspections
Particulars over Time
Average Class
Average Flag
Average Owner
Legal Framework:
Number of LI ratified
(Flag and Owner)
Ship 1 Ship 2 (Twin of Ship 1)
Casualty No Casualty
Same Initial Events:
Age Group, Tonnage Group, Construction Details (Ship Yard
Country, Owner country, Classification Society, DH)
144
summarized form for all models. The model produces probabilities on an individual ship
level (i). The rest of the notation is as follows: ℓ represents the variable groups, nℓ is the
total number of variables within each group of ℓ and k is an index from 1 to nℓ. A separate
model was produced for each of the models listed in table Table 40 for either the
seriousness of casualty or casualty first events.
Equation 8: Definition of term xiβ of Casualty Detailed Model
k k,i
n
k k,i k
n
k k,i k
n
k
k k,i i i
n
i i k
k k,i i i
n
i k
k k,i
n
k k,i i i k
n
k
k k,i i
n
i i k
β β β
β β β β β
β β β β
β β β β
x β β β β β β
Σ PSC Σ DETPS Σ CODE
LIFS DH Σ RS GR ln(TIME )
FSInd Σ OWN OwnInd LIOWN
Σ CL CLInd CLWdr Σ FS
ln(AGE ) ln(SIZE ) Σ ST STInd
19, 1 20,
1
18, 1
1
1
15, 16 17
1
13 14 1
10, 11 12
1
9 1
8,
1
5, 6 7 1
1
1
3, 4
1
i 0 1 2 1
18 19 20
15
10
5 8
3
=
−
=
−
=
−
=
−
=
−
=
−
=
−
=
+ + +
+ + + + +
+ + + +
+ + + +
= + + + +
7.3. Model Assessment and Final Results
7.3.1. Type I & II Model: Casualty Refined View
Table 42 provides a split up of the ships with casualties into their seriousness which is
the basis for the Type I and II models. Type II is based on a combined dataset for very
serious and serious casualties. The figures are based on the number of ships and since a
ship can have multiple casualties, some ships are counted in each of the casualty
categories. The table provides the number of ships with a certain type of casualty and the
corresponding total cases (ships with casualties and inspections).
Table 42: Summary of Matched Dataset by Seriousness of Casualty (by Ship)
Final Datasets Very Serious Serious Less Serious
By Seriousness Casualties
Total
Cases Casualties
Total
Cases Casualties
Total
Cases
No Time Frame 306 10,778 2,345 77,980 1,457 44,882
6 months 167*) 6,007 1,387 46,522 881 28,008
Perfect Twins 156 958 1,033 4,824 641 3,091
*) Note: figures are with passenger vessels and the Caribbean MoU while final models are without
these variables due to lack of data
Type I model (very serious) and type II model were tested for the presence of
heteroscedasticity following the same procedure as explained in section 6.3. Model
Assessment and Final Results for the variables age and tonnage. The null hypothesis (Ho)
assumes homoscedasticity. A summary of the findings can be seen in Table 43 below and
in Appendix 26 where Ho is rejected only for tonnage in the Type II model. Based on the
process and the findings described in Chapter 6 for the very serious casualties, the
presence of heteroscedasticity in these models for the variable tonnage and age are
145
assumed not to have a serious influence on the results of the calculated probabilities and
estimation is therefore not corrected by using Equation 6 but directly by Equation 7. The
remaining statistics of the final models for Type I and II are then presented in Table 44
and in Appendix 27 and Appendix 28 for further detail. The table lists the number of
observations in each model, the number of twin outliers that were identified and
eliminated, the cut off rate and a summary of other statistics.
Table 43: Test Statistics for LM-Test: Type I and II models
Type of Model Variable
Tested
LM-Statistic p-value
6m very serious (type I) Age 4.261 0.0389 – do not reject ho
Tonnage 4.061 0.0438 – do not reject ho
Combined model (type II) Age 5.900 0.0268 – do not reject ho
Tonnage 20.998 0.0000 – reject ho
Note: 1% significance level used
Table 44: Summary of Statistics – Type I and II Model
6 months Time Frame
Type I Models very serious serious less serious
0 = 5665 0 = 44124 0 = 26551
1 = 161 1 = 1362 1 = 860
# observations in
final model
Total = 5826 Total = 45486 Total = 27411
# outliers (twins) none none none
Cut Off 0.0276 0.0299 0.0314
LOG PRO LOG PRO LOG PRO
Mc Fadden R2 0.166 0.162 0.139 0.139 0.077 0.077
% Hit Rate y=0 73.93 72.22 70.00 68.28 66.95 66.00
% Hit Rate y=1 71.43 72.67 73.35 75.18 64.07 65.23
% Hit Rate Tot 73.86 72.23 70.10 68.49 66.86 65.97
HL-Stat. (df=8) 9.41 19.54 3.00 16.60 9.75 9.62
p-value 0.3088 0.0120 0.9343 0.0345 0.2832 0.2927
Remarks
w/o passenger vessels
and Caribbean MoU
with passenger vessels but without Caribbean
MoU
Type II Model (VS and Serious)
0 = 50610
1 = 1540
# observations in
final model
Total = 52150
# outliers (twins) none
Cut Off 0.0295
LOG PRO
Mc Fadden R2 0.130 0.1292
% Hit Rate y=0 68.88 67.09
% Hit Rate y=1 72.53 74.09
% Hit Rate Tot 68.98 67.29
HL-Stat. (df=8) 11.749 24.39
p-value 0.1633 0.0020
Includes Caribbean MoU and all ST
One cannot see any major difference between probit and logit. The models were reduced
using a 1% significance level. Only two variables (deficiency codes) in the less serious
146
casualty model are left in the model at a 5 % significance level. The results are acceptable
for the amount of observations in each model. For the Type I models, the hit rate and the
McFadden R2 is higher for the very serious models compared to the other two model types
and the HL-statistic indicates a better fit. Logit models are used for the visualization
part.
7.3.2. Type III Model: Casualty First Events and Deficiencies
The Type III models are a series of five regressions which link casualty first events with
deficiencies that are found previously. The basis for these regressions is all ships that
were inspected six month prior to a casualty, a total of 2321 ships. The corresponding
datasets are listed in Table 45 below and the key statistics of the final models are given in
Table 46. The actual regression outputs can be seen in Appendix 30 for all five models.
Table 45: Summary of Matched Dataset by Casualty First Events (by Ship)
Casualty First Event Casualties Twins Total Obs
Fire/Explosion 213 6005 6218
Wrecked/Stranded/Grounded 526 18605 19131
Collision/Contact 713 22541 23254
Deck Related 253 8104 8357
Engine Related 819 26260 27079
*) Figures are with the Caribbean MoU and all ship types while final model can vary accordingly
Table 46: Summary of Statistics – Type III Models (6 month time frame)
Fire/Explosion
Wrecked/
Stranded/Grounded Collision/Contact
0 = 5484 0 = 18098 0 = 21641
1 = 191 1 = 502 1 = 688
# observations in final
model
Total = 5675 Total = 18600 Total = 22329
Cut Off 0.0337 0.02698 0.0308
LOG PRO LOG PRO LOG PRO
Mc Fadden R2 0.088 0.088 0.070 0.0698 0.068 0.068
% Hit Rate y=0 66.63 65.24 67.05 66.07 67.89 66.74
% Hit Rate y=1 66.49 67.54 64.34 65.94 61.48 62.35
% Hit Rate Tot 66.63 65.32 66.98 66.06 67.70 66.61
HL-Stat. (df=8) 9.08 5.00 13.78 15.28 6.72 4.74
p-value 0.3360 0.7574 0.0878 0.0539 0.5666 0.7849
Deck Related Engine Related Remarks
0 = 7538 0 = 26260 All Models except
1 = 233 1 = 819 Engine rel. casualties # observations in final
model
Total = 7771 Total = 27079 are w/o the Caribbean
Cut Off 0.0300 0.0310 Fire: w/o container
LOG PRO LOG PRO WSG: w/o passenger
Mc Fadden R2 0.079 0.079 0.084 0.085 and other ST
% Hit Rate y=0 65.46 64.13 69.72 68.74 COCO: w/o other ST
% Hit Rate y=1 67.81 69.96 63.00 64.22 Deck: w/o passenger
% Hit Rate Tot 65.53 64.30 69.52 68.60 and other ST
HL-Stat. (df=8) 4.80 4.15 6.32 3.90
p-value 0.7789 0.8438 0.6111 0.8664
Note: WSG = Wrecked, Stranded, Grounded, COCO = Collision/Contact
147
A ship can have more than one casualty first event and therefore a ship with a casualty
can appear in each of the categories first events. The twins are the ships out of the ships
with no casualty (21,880 ships) of the respective ships with a casualty.
The results of the Type III models are not as good as the models based on the seriousness
of casualty which is expected to be the case. The results are still acceptable and the hit
rate is still above 65% for all models. The models had to be adapted according to the
number of observations and for most models (except Engine related), the Caribbean MoU
had to be taken out as well as the passenger vessels and other ship types due to lack of
data. Not much difference can be seen between the logit and the probit models and logit
models are used for the visualization part.
In addition, the LM test was performed similar to the Type I and II models and the
results can be seen in Table 47 for the variables age and tonnage and in Appendix 29 for
further reference. The null hypothesis (H0) assumes homoscedasticity.
Table 47: Test Statistics for LM-Test: Type III models
Model Variable Tested LMStatistic
p-value
Fire/Explosion Age n/a n/a
Tonnage 1.255 0.2625 – do not reject ho
Wrecked/Stranded/Grounded Age 0.279 0.5967 – do not reject ho
Tonnage 0.398 0.5276 – do not reject ho
Collision/Contact Age 1.019 0.3126 – do not reject ho
Tonnage 2.448 0.1176 – do not reject ho
Deck Related First Events Age 0.194 0.6600 – do not reject ho
Tonnage 0.302 0.5823 – do not reject ho
Engine Related First Events Age 0.176 0.6744 – do not reject ho
Tonnage 0.261 0.6091 – do not reject ho
Note: 1% significance level was used
The results show that there is no presence of heteroscedasticity in the form described
under 6.3. Model Assessment and Final Results for the two variables in question. The next
section will provide the visualization and interpretation part of all models.
7.4. Interpretation and Visualization of Twin Models
This section will summarize the results and visualize them in order to give a refined view
on the effect of port state control inspection variables on the probability of a casualty
(either by seriousness or casualty first event). It will look into the effect of inspections in
general given they have been performed within a certain time frame, detention, the time
in-between inspections and the deficiencies that were found in the inspections.
7.4.1. Refined View on PSC Inspections: Summary of Partial Effects
Table 48 lists a summary of the coefficients of the variables of interest for the Type I, II
and III series of models. The main findings from this table can be summarized as follows
and will be visualized in the sections to come.
Table 48: Summary of Main Variables and their Significance: All Twin Models
Type I Models Type II Model Type III Models
Variables of Interest
very
serious serious less serious
combined model
very serious & serious
Fire
Expl. W/S/G
Collision
Contact
Deck
Related
Engine
Related
Industry Inspections Coef. Coef. Coef. Coef. Coef 2nd ST Coef. Coef. Coef. Coef. Coef.
Rightship 1 star (RS inspected) benchmark benchmark n/s n/s n/s -1.4344 n/s
Rightship 2 star -1.0013 -0.9613 -0.2872 -0.9796 n/a n/a n/a n/a n/a
Rightship 3 star -1.3447 -1.1532 -0.5105 -1.1838 n/a n/a n/a n/a n/a
Rightship 4 star -2.4995 -2.7490 -0.5970 -2.7714 n/a n/a n/a n/a n/a
Rightship 5 star -3.0056 -3.9050 -0.7596 -3.8793 n/a n/a n/a n/a n/a
Greenaward Certified n/s n/s n/s n/s n/s n/s n/s n/s n/s
Port State Control Inspected/Detained
Time in-between inspections n/s 0.1526 0.1169 0.1107 0.1503 n/s n/s n/s 0.0938
Paris MoU inspected n/s n/s n/s n/s n/s n/s n/s n/s 0.8225
Caribbean MoU inspected not in model n/s not in model n/s
Viña del Mar MoU inspected -1.3812 -0.6322 -0.6322 -0.8266 -1.0950 n/s -0.9150 -1.2067 n/s
Indian Ocean MoU inspected n/s -1.5607 -1.5607 -1.4198 -1.5103 -1.2964 -1.1704 -2.3361 n/s
USCG inspected n/s n/s n/s -0.3184 n/s n/s n/s n/s 0.8545
AMSA inspected n/s -0.5662 -0.5662 -0.5182 n/s -1.3373 -0.8534 n/s n/s
Paris MoU detained n/s n/s n/s n/s n/s n/s n/s n/s n/s
Caribbean MoU detained not in model n/s not in model n/s
Viña del Mar MoU detained n/s n/s n/s n/s n/s n/s n/s n/s n/s
Indian Ocean MoU detained n/s n/s n/s n/s n/s n/s n/s n/s n/s
USCG detained n/s n/s n/s n/s n/s n/s n/s n/s -0.3169
AMSA detained n/s n/s 0.7553 n/s n/s 0.6844 n/s n/s 0.5202
Deficiencies Found (only significant ones are listed) per ship type
C0400 Food and catering n/s n/a n/s -0.6149 (ST3) n/s n/s n/s n/s -0.1245
C0500 Working spaces & acc. prev. n/s n/s n/s n/s n/s n/s n/s n/s n/s
C0700 Fire Safety measures n/s 0.0335 n/s 0.0495 (ST1) n/s 0.02376 n/s n/s 0.0432
C0900 Structural Safety n/s n/s 0.0252 0.0657 (ST3) n/s n/s 0.0315 n/s n/s
C1000 Alarm Signals n/s n/s n/s n/s n/s 0.2636 n/s n/s n/s
C1200 Load lines n/s n/s n/s -0.0527 (ST1) 0.1261 (ST4) n/s n/s n/s 0.0701 n/s
C1400 Propulsion & aux. engine n/s 0.0318 0.0241 0.0467 (ST2) 0.0486 0.02231 n/s n/s 0.0645
C1700 MARPOL A. I (Oil Pollution) n/s n/s 0.0573 n/s n/s n/s n/s n/s 0.0449
C1800 Gas and chemical carriers 0.5037 n/s n/s n/s 0.2849 n/s n/s n/s n/s
C1900 MARPOL A.II (Noxious L.) n/s n/s n/s n/s n/s n/s n/s 0.5146 n/s
C2000 SOLAS Operational def. n/s -0.0724 n/s -0.1117 (ST1) n/s n/s n/s n/s -0.0801
C2100 MARPOL relate. oper. def. n/s n/s n/s n/s n/s n/s -0.3094 n/s n/s
C2200 MARPOL A.III n/s n/s n/s n/s n/s n/s n/s n/s n/s
C2300 MARPOL A.V n/s n/s -0.1578 n/s n/s n/s n/s n/s n/s
C2500 ISM related deficiencies n/s 0.0392 n/s 0.04147 (ST1) 0.1193 (ST4) n/s n/s 0.0624 n/s n/s
Note: n/s= not significant, n/a=not applicable, W/S/G=Wrecked, Stranded, Grounded; ST1=general cargo, ST2=dry bulk, ST3=container, ST4=tanker, ST5=passenger
149
Overall Summary
o The coefficients for the variables indicating if a ship has been inspected by one of the
industry vetting inspection regimes (Rightship) for bulk carriers and oil tankers are all
negative and follow the overall ranking of a vessel102 while the coefficient of the
variable indicating if the ship is Greenaward certified is not significant which might
be just due to lack of data103. For the Type III models, this parameter is only
significant for deck related first events. Given the fact that those inspections are
primarily carried out on bulk carriers, this finding is found to be in line with the
general expectation.
o The parameter of the variable indicating time in-between inspections is not significant
for very serious casualties but is positive for all other categories of the Type I and II
models. For the casualty first events, it is only significant for fire & explosion and
engine related first events.
o The coefficients of the variables indicating where the ship was inspected is mostly
negative for serious and less serious casualties and only one regime remains
significant for very serious casualties (Viña del Mar Agreement on PSC). For the
casualty first events, these parameters are mostly negative or not significant with the
exception for engine related first events where it is positive for two regimes.
o The coefficients of the variables indicating if the ship was detained are mostly not
significant with the exception of less serious casualties and for the categories wrecked,
stranded or grounded and engine related first events. For engine related first events,
the parameter is negative for one regime (USCG).
Deficiencies per seriousness and ship type (Type I and II models)
o For very serious casualties, a positive effect can be found with code 1800 (gas and
chemical carriers) and not other code remains significant.
o For serious casualties, a negative effect can be found for codes 2000 (SOLAS
operational deficiencies) while a positive effect can be seen with codes 700 (fire safety
measures), codes 1400 (propulsion and aux. engine) and code 2500 (ISM code).
o For less serious casualties, only one code shows a negative effect – code 2300
(MARPOL Annex V). Codes with positive effect are code 900 (structural safety), code
1400 (propulsion and aux. engines) and code 1700 (MARPOL Annex I).
o With respect to the deficiencies per ship type (Type II model), codes that remain
significant for general cargo vessels are code 700 (fire safety measures: positive), code
1200 (load line: negative), code 2000 (SOLAS oper. safety: negative) and code 2500
(ISM related def.: positive).
o For dry bulk, only one code remains significant and positive which is code 1400
(propulsion and aux. engine). For the container vessel, codes 400 (food and catering:
negative), code 900 (structural safety: positive) remain significant. For tankers, two
codes remain in the model and are code 1200 (load lines: positive) and code 2500 (ISM
related deficiencies: positive).
Deficiencies per casualty first event (Type III models)
o Looking at the deficiency codes and the significance of their parameters with respect
to the casualty first events, for fire and explosion, codes 1400 (propulsion and aux.
engines) and code 1800 (gas and chemical carriers) are significant and positive.
102 For Rightship, the risk associated with a vessel is ranked by stars where a 1 star vessel shows
highest risk and a 5 star vessel shows lowest risk
103 the total amount of Greenaward certified vessels incorporated into the dataset was only about
240 records for the time span 2000 to 2004
150
o For the category wrecked/stranded/grounded, the remaining codes are code 700 (fire
safety measures), code 1000 (alarm signals) and code 1400 (propulsion and aux.
engines) are significant and positive104.
o For collision/contact, two codes are positive as – code 900 (structural safety) and code
2500 (ISM related def.) and one codes is negative – code 2100 (MARPOL related
operational deficiencies).
o For deck related first events, codes 1200 (load lines) and code 1900 (MARPOL Annex
II) are significant and positive.
o For engine related first events, four codes remain of which three are positive and two
are negative. Positive effects are with codes 700 (fire safety measures), 1400
(propulsion and aux. engines) and 1700 (MARPOL Annex I) while negative effects are
with codes 400 (food and catering) and code 2000 (SOLAS operational deficiencies).
Summary of Findings in relation to deficiencies
It is difficult to interpret the significance and the sign of the parameters of the deficiency
codes towards either the seriousness and with respect to the ship types or the casualty
first events since the variable is not based on the last inspection only but also contains
information from previous inspections. Sometimes, this means an accumulation of
inspections and sometimes this means, only the last inspection which was performed at
least six months or less before the casualty. The analysis is therefore only being seen as a
first attempt to look at both aspects closer.
What can be concluded from this portion of the analysis is that some inspections and the
fact that deficiencies are found are effective towards decreasing the probability of having
a casualty. This effect varies across ship types, seriousness of casualty and casualty first
events. Code 1400 (propulsion & aux. engine) seems to be an important deficiency code
which in the probability of detention does not come out to be very important in all
regimes. It has a positive effect for serious and less serious casualties, in particular for
dry bulk carriers and for casualty first events such as fire & explosion,
wrecked/stranded/grounded and engine related casualty first events. This could indicate
that there is room for improvement. It seems that deficiencies are found but due to lack of
enforcement (detention) or follow up on deficiencies, the effect is positive rather than
negative. It is important to notice though that when the vessel is in port and port state
control is performed, the engines are not under full operation and it is difficult to inspect
some aspects of the main engine.
Another code where its parameter shows a positive effect is the ISM code (code 2500)
which captures the whole safety management system onboard a vessel. It is positive for
serious casualties, general cargo vessels and tankers and for casualty first event collision
and contact. On the other hand, code 2000 (SOLAS operational related deficiencies) is
negative for serious casualties and engine related first events for the ship type general
cargo. This could be interpreted as the effectiveness in rectifying the deficiencies or drills
and having a negative effect for general cargo ships.
Another clear example is code 1800 (gas and chemical carriers) which is positive and
significant for very serious casualties and casualty first events fire and explosion. This
could be further identified as an area that should be looked at and has already been
discussed at IMO during MSC (81) in May 2006105. This finding confirms that there is
problem with enforcing the legal conventions on chemical carriers. The next area will
104 code 700 and code 1400 are significant at the 5% level only
105 The author attended as observer MSC (81) in May 2006, IMO, London
151
present some graphs to visualize the findings. It will first look at the time in-between
inspections, the effect of inspections and detention and the deficiencies themselves.
7.4.2. The Effect of Time in-between Inspections
Figure 89 and Figure 90 both show the effect of time in-between inspections on the
probability of a casualty per seriousness and casualty first events for a dry bulk carrier.
As can be seen from the first graph, the time in-between inspections is not important for
very serious casualties versus serious and less serious casualties and is only significant
for fire & explosion and engine related first events. The strongest effect can be seen with
serious casualties and only little effect can be identified with less serious casualties.
Figure 89: Effect on Time in-between Inspections: Seriousness
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
1 2 3 4 5 6 7 8 9 10 11 12
months in-between inspections
Probability of Casualty
serious less serious very serious
serious
less serious
very serious
Age: 14 yrs
Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
On average and regardless of the seriousness of the casualty, the probability of casualty
increases by 2.3% within the time frame of one year. With respect to fire and explosion,
the probability of casualty can increase as much as 2.7% within a year while this figure
changes to barely 0.5% for engine related casualties.
7.4.3. The Effect of Inspections and Detentions
Summary of Partial Effects per Seriousness and First Events
The results given in Table 48 indicate that detention is only significant for less serious
casualties and for casualties related to ships that are wrecked, stranded or grounded and
or have engine related casualty first events. The fact that this variable is mostly not
significant, especially for very serious and serious casualties indicates that detention
cannot be measured of having a negative effect on the probability of casualty. One would
expect to see a negative coefficient associated with this variable.
152
Figure 90: Effect on Time in-between Inspections: First Events
0.00
0.02
0.04
0.06
0.08
0.10
0.12
1 2 3 4 5 6 7 8 9 10 11 12
months in-between inspections
Probability of Casualty
fire & explosion engine related
fire & explosion
engine related
Age: 14 yrs
Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
On the other hand and recalling the overall figures that were presented at the beginning
of Chapter 5 where out of the 25,836 ships with inspections, 2,321 ships were inspected
within six months prior to a casualty and 162 ships were detained (4% of vessels with
casualty) over a six year time period, it might be difficult to interpret this coefficient and
the effect is captured by the inspections itself. The difference between detention and a
regular inspection is the enforcement of the rectification of the deficiencies. While
detained vessels have to rectify them, not detained vessels do not have to do so and
although the deficiency was found during an inspection, the inspection lacks on the
enforcement side.
Table 49 is based on the Type I models where restrictions106 using the Wald Test were
tested for the variables indicating where the vessel was inspected to see if their means
differ. The null hypothesis in this case states that there is no significant difference across
the regimes (Ho= coefficients do not vary). The results indicate that while there is no
significant difference with relation to very serious casualties (only one variable remains
significant), for serious casualties, AMSA and the Viña del Mar MoU are apart from the
Indian Ocean MoU. No difference can be seen for the Paris MoU and the USCG since as a
benchmark, the Paris MoU was used and the USCG does not remain to be significant.
For less serious casualties, AMSA, the Indian Ocean MoU and the Viña del Mar
Agreement do not show a difference (at the 1% significance level) while the USCG and the
Paris MoU are not significant. Figure 91 to Figure 93 both visualize these effects for a
particular ship for a particular vessel but not in a combined format. The variables
indicating where a ship was inspected is constructed as a percentage fraction of each
vessel to the total inspection a vessel had previously and not as a total sum of inspections
which was used in the normal casualty models. What the variables give is a capture of the
106 based on Wald Test for Testing Coefficient Restrictions, a standard procedure in Eviews
153
total inspection fraction of all of the regimes of a particular vessel where the Caribbean
MoU had to be excluded from the models due to lack of data.
Table 49: Testing of Restrictions (Wald Test) – Inspection Variables: Type I Models
Very
Serious
Serious
Restrictions/p-value
Less Serious
Restrictions/p-value
AMSA=IMOU=VMOU
(0.0017) – reject ho
AMSA=IMOU=VMOU
(0.0286) – do not reject ho
AMSA=IMOU
(0.0007)- reject ho
AMSA=IMOU
(0.1054)- do not reject ho
IMOU=VMOU
(0.0013) – reject ho
IMOU=VMOU
(0.0086)- reject ho
Only the VMOU
remains significant
and shows a negative
effect. There is no
significant difference
amongst the other
regimes.
AMSA=VMOU
(0.07607) – do not reject ho
AMSA=VMOU
(0.2040) – do not reject ho
Note: Figure in bracket is the p-value of the test, 1% significance level
Figure 91: Inspection Effect across Regimes: Very Serious
Very Serious Casualty
Inspection Effect – Difference Across Regimes
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.10
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00
Percentage change in Frequency of Inspection
Vina del Mar Average
Ship Type: Dry Bulk
Age: 14 yrs, Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
Vina del Mar
average
To visualize the differences, a particular ship is chosen and its associated probability of
casualty is calculated. To see the effect over time, the variable in question is increased by
certain percentage fractions (10%) which can also be seen as an increase in the frequency
of inspections. The interesting part in these graphs is to see how the regimes differ with
respect to the probability of casualty. As mentioned before, only the Viña del Mar
Agreement on PSC is significantly different from the other regimes for very serious
casualties. The average is to be understood as the average of all regimes.
154
Figure 92: Inspection Effect across Regimes: Serious
Serious Casualty
Inspection Effect – Difference across Regimes
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.10
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00
Vina del Mar Indian MoU
AMSA Average
Ship Type: Dry Bulk
Age: 14 yrs, Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
Percentage Change in Frequency of Inspections
average
Indian Ocean MoU
Vina del Mar
AMSA
Figure 93: Inspection Effect across Regimes: Less Serious
Less Serious Casualty
Inspection Effect – Difference across Regimes
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.10
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00
Vina del Mar Indian MoU
AMSA Average
Ship Type: Dry Bulk
Age: 14 yrs, Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
Percentage Change in Frequency of Inspections
average
AMSA
Vina del Mar
Indian Ocean MoU
155
For serious casualties, AMSA and the Indian Ocean MoU are very close and apart from
the Indian Ocean MoU and the other regimes. For less serious casualties, AMSA, the
Indian Ocean MoU and the Viña del Mar Agreement on PSC are similar (also confirmed
previously) and below the average. They are different from the Paris MoU and the USCG.
At first sight, the order of the regimes is not as one would have expected them to be.
Regimes below the average seem to show a larger effect. On the other hand, it might also
reflect the learning stage of a regime over the time period covered by the inspection data.
By the end of 2004, the Paris MoU has been in existence since 1982 while the Indian
Ocean MoU only exists since 7 years and the Viña del Mar since 13 years.
The Indian Ocean MoU region has been identified as a high risk area based on the
descriptive statistics. This regime has more local trade with ships that might show more
obvious signs of being sub-standard and the effect of inspections are therefore to be
expected to be higher than in other regions with better ships. In addition, the Paris MoU
area has maintained the 25% target factor previously which could have led to the
inspection of good ships in the past in order to fill the quota versus sub-standard ships
when they have not been available in the area needed for inspection for the last six years.
As for the Viña del Mar region, the region might show more of the sub-standard ships
that have been driven out of the Paris MoU or the USCG over the last six years.
The next section will give an overview in relation to the casualty first events based on the
Type III models and similar to the procedure described for the Type II models. The results
can be seen in Table 50. One can see that there are no significant differences for the
variables that are left in the models across the regimes and that the null hypothesis (Ho=
coefficients do not vary) can be rejected in all cases at a 1% significance level. The results
are visualized with a combined graph (Figure 94) which shows the average effect of
inspection on the probability of casualty per first event for a dry bulk carrier. The average
is based on an average of all regimes and is therefore more averaged out in comparison to
each of the individual variables.
Table 50: Testing of Restrictions (Wald Test) – Inspection Variables: Type III Models
Model Type Variables Tested p-value
Fire/Explosion VMOU=IMOU 0.5819 – do not reject ho
Wrecked/Stranded/Grounded AMSA=IMOU 0.9359 – do not reject ho
Collision/Contact AMSA=VMOU=IMOU 0.7131 – do not reject ho
Deck Related First Events VMOU=IMOU 0.1136 – do not reject ho
Engine Related First Events USCG=PMOU 0.0406 – do not reject ho
Note: 1% significance level used
The graph shows that the strongest effect can be seen for deck related first events
followed by collision/contact similar to the category wrecked/stranded/grounded and
fire/explosion. For engine related first events, the effect is slightly positive. Linking this
graph back to the probability of detention and deficiency code 1400 (propulsion and aux.
machinery) where the contribution weight of this code was found not to be very high
across all regimes, one could conclude that there is room for improvement in this area.
156
Figure 94: Average Inspection Effect per Casualty First Event
Average Inspection Effect per Casualty First Event
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00
Percentage Change in Frequency of Inspection
Probability of Casualty
Fire/Explosion Wrecked/Stranded/Grounded
Collision/Contact Deck Related
Engine Related
ST: Dry Bulk
Age: 14 yrs
Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
Fire/Explosion Wrecked/Stranded/Grounded
Engine Related
Collision/Contact
Deck Related
Frequency of Inspection and Detention
The next two graphs give an overview of the probability of casualty per frequency of
inspection and detention given the ship has been inspected at least once within a six
months time period. The probabilities are averages based on all inspected vessels or all
detained vessels.
Figure 95 shows that the probability of detention decreases with the frequency of
inspections while the probability of serious casualty increases from 3% to 7%. Less
serious casualties increase by about 3% while very serious casualties decrease from 4% to
2% over time and with increased frequency of inspections. In essence, on could conclude
that with increased amount of inspections, the probability of casualty does not necessarily
decrease.
Figure 96 then shows the probability of casualty and how it changes with the frequency of
detention versus not detained ships. The graph shows that for ships that are inspected
and detained six months prior to a casualty, the probability decreases from an average of
2.9% to 1.3% for very serious casualties over a time period of six years while it increases
for serious and less serious casualties. For less serious casualties, it then decreases again
after the ship has been detained more than 3 times. The same applies for very serious
casualties but not for serious casualties where the probability increases from 6.1% to
7.7% for ships that have been detained more then five times.
157
Figure 95: Probability of Casualty per Frequency of Inspection (6 months prior)
0.02
0.07
0.04
0.06
0.03
0.04
0.02 0.02
0.05
0.04
0.03
0.03
0.08
0.04
0.000
0.010
0.020
0.030
0.040
0.050
0.060
0.070
0.080
up to 5 times up to 10 times up to 15 times above 16 times
Probability of Casualty/Detention
very serious serious less serious detention
Note: based on a time frame of six years or 4 complete inspection years and average probabilities of
approx. 50,000 vessels
Figure 96: Probability of Casualty per Frequency of Detention (6 months prior)
0.013
0.077
0.051
0.029 0.031
0.027 0.028
0.061
0.053
0.039 0.040
0.035 0.038
0.047
0.056
0.000
0.010
0.020
0.030
0.040
0.050
0.060
0.070
0.080
0.090
not detained detained once detained twice detained 3-5
times
detained > 5
times
Probability of Casualty
very serious serious less serious
Based on casualty normal model and on average probabilities of approx. 50,000 vessels
The fact that the probability of casualty for serious casualties and less serious casualties
increases with either the frequency of inspection and detention could also indicate the
involvement of a certain human factor associated with these casualties. It might be easier
for port state control to identify very substandard vessels and therefore the effect of
158
inspections and detentions are expected to be higher for very serious casualties while this
is not the case for serious and less serious casualties. On the other hand, the increased
probability of casualty for increased inspections and detentions also reflects to a certain
extend that higher risk vessels are targeted for inspection. As third reflection, increased
inspection or over inspection does not necessarily have a negative effect of the probability
of detention. The last chapter of part III will give some more insight into deficiencies and
the probability of casualty either by seriousness or casualty first event.
7.4.4. PSC Deficiencies and the Probability of Casualty
The last chapter of this thesis will provide a closer look at deficiencies in relation to
seriousness and first events of a casualty and is based on the Type I, II and III models. It
visualizes the findings stated in Table 48 previously in order to facilitate the
interpretation of the coefficients.
The first set of graphs is based on the Type II model (the combined model) where the
deficiencies are multiplicative dummy variables of the ship types and combine very
serious and serious casualties. Figure 97 and Figure 98 show the results for codes with
negative effects and codes with positive effects.
Figure 97: Very Serious and Serious Casualties (Negative Effects)
Very Serious and Serious per Ship Type (Negative Effects)
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.10
1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Casualty
Food & Catering (5) – container
Load Lines (6) – general
Solas oper. Safety (4) – general
Age: 14 yrs
Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
Note: based on type II models
The negative effect of food and catering for container vessels cannot really be explained
other than that an improvement in working and living conditions for crew members
onboard have an overall positive effect on the performance of the crew.
159
The significance of the other two codes for general cargo ships are easier to interpret.
According to the Paris MoU Manual for PSC Officers107 deficiencies in the area of load
lines include overloading, freeboard markings, conditions of railings, cargo hatches, doors,
ventilation pipes and lashings. It seems that general cargo vessel seems to show
deficiencies in this area prior to a casualty and that these deficiencies are rectified. For
load line, rectification might not be so easy and immediate while for deficiencies in the
area of SOLAS related deficiencies, it can be rectified easier and therefore have an
immediate effect. SOLAS related operational deficiencies include deficiencies such as the
muster list, fire drills, abandon ship drills, the level of communication onboard, bridge
operations, the operation of GMDSS and cargo operations. It shows that if port state
control identifies these deficiencies and if they are rectified, they can have a negative
effect. In addition, the drills might help in this respect to. The only regime who requires
drills during an annual exam is the USCG.
Figure 98 shows the deficiency codes that have a positive effect towards the probability of
casualty. ISM appears twice (for tankers and general cargo) and codes associated with
stability and structure (load lines and structural safety) are relevant for containers and
tankers. It might be more difficult to rectify deficiencies in this area since it might take
more time to do so. In the case of lack of follow up, it seems that ships with deficiencies in
this area show a higher probability of having a very serious or serious casualty.
Figure 98: Very Serious and Serious Casualties (Positive Effects)
Very Serious and Serious per Ship Type (Positive Effects)
0.00
0.05
0.10
0.15
0.20
0.25
0.30
1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Casualty
Structural Safety (6) – container
Propulsion & Machinery (3) – dry bulk
Age: 14 yrs
Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
ISM (8) – tanker
Load Line (6) – tanker
ISM (8) – general cargo
Fire Safety M. (2) – general cargo
Note: based on type II models
107 Paris MoU, Manual for PSC Officers, Revision 8
160
Fire safety measures are relevant for general cargo and show that general cargo ships
might show more problems with the actual equipment related to fire prevention while the
drills and the performance of drills reflects the operational side which can have a
negative effect. Very little effect can be seen by deficiencies in the area of propulsion and
machinery for dry bulk vessels.
What is not visualized here but clearly shown in Table 48 is the positive effect of code
1800 (gas and chemical carriers) towards the probability of a very serious casualty. This
code is associated with deficiencies in the area of cargo segregation, cargo transfers and
ventilation systems, the cargo pump room, temperature controls, and fire protection of
cargo deck areas, personal protection and emergency towing arrangements. It applies to
chemical tankers, gas carrier and oil tankers and shows that this is an area port state
control can improve in not only detecting the deficiencies but also in ensuring that they
are rectified and that the ISM system onboard is implemented onboard which is further
confirmed by the positive effect of the ISM code with tankers. This is somehow
surprisingly given the fact that tankers undergo a significant amount of vetting
inspections which also looks closely at the implementation of the ISM code.
Figure 99 shows the deficiency codes which are left to be significant for less serious
casualties for all ship types. It is less accurate than the previous models but confirms two
areas – structural safety and propulsion and machinery with a moderate positive effect
and MARPOL Annex I (oil pollution) with a stronger positive effect.
Figure 99: Less Serious Casualties and Deficiencies
Less Serious Casualty-all ship types
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Casualty
Certificates (1) Working Conditions (2)
Safety & Fire (2) Stability & Structure (4)
Equipment & Machinery (5) Navigation & Commun (6)
Ship & Cargo Operations (7) Management (8)
Age: 14 yrs
Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
Structural Safety (4)
Propulsion & Machinery (5)
All other codes
Marpol Annex I (7)
Marpol Annex V (7)
Note: based on type I model – less serious
161
Deficiencies in this area contain for instance the SOPEP (Ship oil emergency plan), the oil
record book, the 15 ppm alarm and oil filtering equipment, the segregation of ballast
tanks, the operation of COW (crude oil washing). The probability of detention is very
strong in this area for the USCG especially for tankers but in aggregated form, for all
ship types (contribution weight is about 21%), this code is not significant for very serious
or less serious casualties but might play a role for less serious casualties and could be an
area of potential problems.
This last section takes a closer look at the various types of casualties and the deficiencies
that were found during an inspection. A separate graph per casualty first event is
produced and is shown in Figure 100 through Figure 104. The deficiency codes are shown
individually in the graphs and not in aggregated format such as the groups that were
used for the visualization part of the probability of detention (see Table 23: Grouping of
Deficiency Codes for Visualization) since few deficiency codes remain to be significant in
the final models. It is therefore more accurate to show the codes individually.
Nevertheless and to keep the coding across the models the same, each graph shows a
legend at the bottom with the deficiency groups used for the probability of detention and
the corresponding number (e.g. 4 for cargo and ship operations) links the actual code
shown in the graph to the deficiency main groups of the probability of detention graphs
produced in Part II of this thesis.
Figure 100 shows the results for fire and explosion identified as first event. Fire and
explosion in this case means a fire and explosion anywhere on the vessel where the main
area of fire has been identified to be in the engine room.
Figure 100: Fire and Explosion and Deficiencies
Fire and Explosion
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.45
1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Casualty
Propulsion and Machinery (3)
Oil, Chemical Tankers &
Gas Carriers (4)
Age: 14 yrs
Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
Note: based on type III model
162
Code 1800 (oil, chemical tankers and gas carriers) and code 1400 (propulsion and
machinery) both show a positive effect towards the probability of having a fire or
explosion. This finding is interesting as it confirms a problem that is already known in
the industry and which has been an agenda item during MSC108 (81) in May 2006 where a
study conducted by an inter-industry workgroup identified 35 cases of fire and explosions
on chemical and product tankers over the last 25 years.
The group concluded that technical failure could not be identified but that the prime
contributor was lack of following the proper operational guidelines which is partly
reflected in the ISM system (onboard and shore side) and might also explain the positive
contribution of the ISM code for tankers in Figure 98. From the port state control point of
view, it shows that a certain lack of compliance has been detected but that the system
lacks in enforcement and implementation. The same applies to propulsion and machinery
where the effect is much less.
Figure 101 shows the probability of engine related first events and Figure 102 gives an
insight into the probability of deck related first events in relation to deficiencies
previously found in port state control inspections.
Figure 101: Engine Related First Events and Deficiencies
Engine Related First Events
0.000
0.005
0.010
0.015
0.020
0.025
0.030
0.035
1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Casualty
Food & Catering (5)
SOLAS oper. def. (4)
Fire Safety Meas. (2)
Marpol Annex I (4)
Propulsion & Machinery (3)
Age: 14 yrs
Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
Note: based on type III model
Engine related first events contain engine breakdowns, black outs, steering gear failure
and propulsion failure. It is therefore not surprising that deficiencies associated with code
1400 (Propulsion and machinery) shows the strongest positive contribution followed by
MARPOL Annex I (code 1700) and fire safety measures (code 700). MARPOL Annex I has
108 Maritime Safety Committee Meeting at IMO (10th to 19th May 2006)
163
also been identified with a positive effect for less serious casualties which might also be
reflected here. The code is not unrelated to the engine room but somehow not directly
related to the events listed above since it deals with environmental issues (oil pollution)
and all procedures connected to it. The fact that this code is positive can also just indicate
the lack of the implementation of operational procedures in the engine room and that
ships that do have a high probability of engine related casualties, also do have a problem
in the area of pollution prevention and fire & safety measures.
On the other hand, two codes show a negative effect which are code 2000 (SOLAS
operational related deficiencies) and code 400 (food and catering). Both codes also show a
negative effect for very serious and serious casualties for general cargo ships and
container vessels. It seems that drills and other operational related items do have a
negative effect on the probability of having an engine related casualty. The code food and
catering might just reflect the human factor such as living and working conditions in
general which are also associated with food.
Figure 102 shows the probability of deck related first events and two deficiency codes
which remain significant – MARPOL Annex II (code 1900) and load lines (code 1200).
Deck related first events contain items such as deck maintenance and stability related
items (capsizing, listing, cargo shifts and flooding).
Figure 102: Deck Related First Events and Deficiencies
Deck Related First Events
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Casualty
Load Lines (6)
Marpol Annex II (4)
Age: 14 yrs
Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
Note: based on type III model
According to the Paris MoU PSC Manual for PSC Officers109, deficiencies associated with
MARPOL Annex II (Noxious Liquids in Bulk) are deficiencies such as the cargo record
109 Paris MoU, Manual for PSC Officers, Revision 8
164
book, the P&A (Procedure & Arrangement Manual) manual, stripping and tank washing
equipment, cargo heating systems and ventilation equipment. At first sight, this code
does not seem to be directly associated with deck related first events but by taking a
closer look, one can identify a connection, especially when it comes to cargo handling
which might also be reflected in deficiencies associated with load lines where cargo shifts
or flooding might be more relevant. What is interesting to see is that for instance the ISM
code is not relevant which is somehow unexpected. Overall, one can conclude that the lack
of following proper cargo operation procedures (in what form ever) do have a positive
influence on the probability of having a casualty. For port state control, this could mean
that deficiencies in this area have been identified but that there is lack of ability to
ensure that these procedures are followed in the future.
Figure 103 shows the effect of deficiencies on casualty first events associated with ships
that were wrecked, stranded or grounded. This category is dominated by stranded and
grounded ships versus wrecked ships. Three codes are significant and show a positive
effect – code 1000 (Alarm Signals), code 700 (fire safety measures) and code 1400
(propulsion and machinery) where the last two are only slightly significant. The more
interesting group of deficiencies are the groups of alarm signals which contains
deficiencies related to the general alarm, crew and fire alarm, steering gear alarm,
engineer’s and other machinery alarms, inert gas alarm, UMS (unmanned machinery
spaces) and boiler alarms. The types of alarms which can be brought into relation with
the first event are probably the alarms associated with the steering gear and other
machinery alarm. It seems that deficiencies are identified in this area by port state
control and that the positive effect is rather strong.
Figure 103: Wrecked/Stranded/Grounded and Deficiencies
Wrecked/Stranded/Grounded
0.00
0.10
0.20
0.30
0.40
0.50
0.60
1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Casualty
Fire Safety Meas. (2) and
Alarm Signals (2)
Propulsion & Machinery (3)
Age: 14 yrs
Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
Note: based on type III model
165
Figure 104 shows the last graph in this series and shows that deficiencies found in the
area of ISM (code 2500) have a positive effect on the probability of having a collision or
contact. The same applies for structural safety (code 900). Deficiencies in the 900 range
contain closing devices (such as watertight doors), stability and loading information and
instruments, steering gear, hull damage, the condition of ballast tanks, any kind of hull
and bulkhead corrosion and cracking.
Figure 104: Collision and Contact and Deficiencies
Collision and Contact
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Casualty
Marpol related oper. Def. (4)
ISM (8)
Structural Safety (6)
Age: 14 yrs
Tonnage: 30419 gt
Flag: Panama
Class: GL
Owner: TMN
Note: based on type III model
MARPOL related operational deficiencies (code 2100) are associated with oily mixtures in
cargo spaces and other related items for cargo spaces of tankers. This code is not related
to collisions and contacts and therefore cannot be interpreted. For code 900, items related
to the steering gear might be very relevant in this category.
The ISM code is certainly relevant as it might show lack of enforcement of safety
procedures. ISM related deficiencies contain items such the safety and environmental
policy, the definition of company responsibilities and the master’s responsibility,
deficiencies in the area of shipboard operations, emergency preparedness, reporting and
analysis of non-conformities, accidents and near misses, the maintenance of the vessel
and company audits. Another area deals with resources and personnel. Violations against
working and resting hours are not a separate code in the ISM group but are included in
the deficiency group 200 – crew certificates.
It is difficult to interpret this graph with certainty but the strong positive effect of ISM
related deficiencies shows that vessels in this category have a higher probability of
having a collision or contact. It might reflect fatigue or lack of bridge procedures as well
as lack of overall onboard maintenance (as reflected in code 900). From a port state
166
control perspective, this might also mean that the deficiencies are identified but that the
enforcement onboard is lacking as is the rectification of such deficiencies. Especially ISM
audits (if required) cannot be done immediately unless the ship is detained. When under
pressure to keep schedules, ships might proceed and ignore some of the recommendations
from port state control.
On a short notice on ISM, the code’s origins go back as far as the late 1980’s when more
concern arose as to the poor safety management of the industry. It was adopted in 1993110
and had to be implemented by 1 July 2002. The last Maritime Safety Committee (MSC
81) in December 2005 presented an impact assessment of the ISM code on the industry in
which several areas of improvement could be identified as follows111:
o More systematic training
o Having an ISM Code performance measurement scheme
o More monitoring of compliance
o Integrating into employment requirements; and
o Involving more people, especially seafarers in writing ISM manuals.
Given the findings in this section of this thesis, the author can fully support these
recommendations. Especially the last recommendation is very relevant. From the 25
inspections and one ISM audit the author could observe during the course of this project,
it has become apparent that very few management companies allow proper input from
seafarers on the design and continuous improvement of the safety management system
which is of direct impact of daily shipboard operations.
On oil and chemical tankers, it has been observed that on many occasions, the system has
been designed to only serve one purpose – which is to meet the requirements of the
vetting inspection questionnaires and not the overall perspective which is to improve the
safety level onboard a ship by taking into consideration the particular working
environment onboard a vessel. This reduces the ISM code to a paper exercise rather than
a workable system for the industry. The latest addition of the oil industry’s Tanker
Management Self Assessment (TMSA) system in addition to ISM further proofs that ISM
has reduced to a paper exercise. In theory, one safety management system should be
sufficient and adaptable to the various industries within the shipping industry. TMSA
allows compliance to four levels where the first level is seen to be the minimum
requirement and meets the requirements of ISM.
7.5. Summary of Major Findings: Refined View
The parameters of the variables indicating if a ship has been inspected by one of the
industry vetting inspection regimes are all negative. The coefficients of the variable
indicating time in-between inspections are not significant for very serious casualties but
are positive for all other categories. For the casualty first events, it is only significant for
fire & explosion and engine related first events.
110 MSC 81/17/1, Role of the Human Element, Assessment of the impact and effectiveness of
implementation of the ISM Code, 21 December 2005, page 2
111 MSC 81/17/1/ Role fo the Human Element, Assessment of the impact and effectiveness of
implementation of the ISM Code, 21 December 2005, page 14
167
The variables indicating where the ship was inspected is mostly negative for serious and
less serious casualties and only one regime remains significant for very serious casualties
(Viña del Mar Agreement on PSC) while several other regimes appear to be significant for
serious and less serious casualties.
For the casualty first events, the parameters are mostly negative or not significant with
the exception for engine related first events. Testing of restrictions shows that there is no
significant difference with respect to the coefficients of the variables indicating where the
ship was inspected and casualty first events at the 1% significance level. The strongest
negative effect can be found on the probability of deck related first events (about 3%) and
a slightly positive effect can be found for engine related casualty first events.
The time span in-between inspection is not significant for very serious casualties but is
for less serious and serious casualties. On average and regardless of the seriousness of
casualty, the probability increases by 2.3% within the time frame of one year. For fire and
explosion, this increase can be 2.7% and 0.5% for engine related casualties.
With respect to the probability of casualty and frequency of inspection and detention, the
probability of a casualty decreases on average while on the contrary, the probability of
serious and less serious casualties increases with the frequency of inspection. The picture
is similar for multiple detentions. The coefficients of the variables indicating if the ship
has been detained are mostly not significant with the exception of less serious casualties
and for the categories wrecked, stranded or grounded and engine related first events. For
engine related first events, this variable is negative for one regime (USCG).
It is difficult to interpret the significance and the signs of the parameters of the deficiency
codes towards either the seriousness and with respect to the ship types or the casualty
first events since the variable is not based on the last inspection only but is a summary of
all inspections that were performed prior to a casualty. Sometimes, this means an
accumulation of inspections and sometimes this means, only the last inspection which
was performed at least six months or less before the casualty. The analysis is therefore
only being seen as a first attempt to look at both aspects closer.
What can be concluded from this portion of the analysis is that some inspections and the
fact that deficiencies are found are effective towards decreasing the probability of having
a casualty. This effect varies across ship types, seriousness of casualty and casualty first
events. Code 1400 (propulsion & aux. engine) seems to be an important deficiency code
while in the probability of detention does not come out to be very important in all
regimes. It has a positive effect for serious and less serious casualties, in particular for
dry bulk carriers and for casualty first events such as fire & explosion,
wrecked/stranded/grounded and engine related casualty first events. This could indicate
that there is room for improvement. It seems that deficiencies are found but due to lack of
enforcement (detention) or follow up on deficiencies (detentions), the effect is positive
rather than negative.
The same applies for another important code – the ISM code (code 2500) which captures
the whole safety management system onboard a vessel. The parameter of this variable is
positive for serious casualties, general cargo vessels and tankers and for casualty first
event collision and contact. The ISM code captures the whole safety management onboard
and its effect is only positive. It could mean that even though port state control detects
deficiencies in the implementation of the ISM code onboard, the rectification or follow up
on the deficiencies it not very successful and the lack of proper implementation onboard
168
leads to an increase in the probability of having a casualty which can be very serious and
serious and is more likely to be associated with a collision.
On the other hand, code 2000 (SOLAS operational related deficiencies) is negative for
serious casualties and engine related first events for the ship type general cargo. This
could be interpreted as the effectiveness in rectifying the deficiencies and therefore
having a negative effect for general cargo ships. It can also be interpreted that for
instance increased drills can help in decreasing the probability of having a very serious
casualty.
Another clear example is code 1800 (gas and chemical carriers) which is positive and
significant for very serious casualties and casualty first events fire and explosion. This
could be further identified as an area that should be looked at and has already been on
the agenda of MSC (81) in May 2006. This finding confirms that there is problem with
enforcing the legal conventions on chemical carriers but that the main contributor was
identified to be of human error by not following the proper procedures.
169
PART IV
This final chapter of the thesis will combine all the various aspects which have been
looked at in this thesis in order to provide the answer to the research questions and to
give recommendations. It therefore provides a condensed summary of the major findings
and gives recommendations on how the safety regime can be improved. It will end with
suggestions on further research in this area.
170
171
Chapter 8: Conclusions and Recommendations
The author would like to stress that this thesis is an independent study and should be
understood as a first attempt to look at various aspects of inspections and their link to
casualties and is by no means complete due to some barriers which could not be overcome
for this study and access to some PSC data could not be obtained. The political dimension
of flag state compliance and port state control is not treated by the author in this study
but emphasis is given on the technical aspects of this topic only.
Before presenting the main findings and recommendation of this thesis, it is perhaps
useful to re-state briefly the original research questions which are to be kept in mind: 1)
What is the present state of the safety regime? 2) Can targeting of ships for inspections be
improved? 3) What is the effect of inspections on casualties? and 4) How can inspections
be improved? The areas of these questions are overlapping and are therefore not treated
separately but in order to put some structure into this last chapter in addressing these
questions, four overall sections were identified as follows: general overview of the safety
regimes, the overall magnitude of improvement possibilities, the effects of inspections on
casualties and recommendations on how to improve the target factor and finally,
identified areas to improve inspections.
8.1. General Overview of the Safety Regime
Many players are part of the safety regime consisting of a statutory part and an industry
driven part. The lack of trust in the industry between flag states, port states,
classification societies, insurance companies and cargo owners has created a playground
for many inspections which are performed on certain ship types (oil tankers, chemical
tankers and dry bulk carriers) in the name of safety. The areas that are inspected in all of
these inspections show a considerable amount of overlapping although the overall
emphasis of industry driven inspections versus statutory inspections is different. The
estimated inspection costs of a port state control inspection is USD 747 per inspection and
port state control inspections associated with zero deficiencies (which accounted for 54%
of all inspections of the dataset used) are estimated to be at USD 12.5 million per year for
the regimes used in this study. The figures are not completely accurate as they do not
take the differences of administrative costs of all countries present in this study into
consideration. Total inspection costs of mandatory and non-mandatory inspections per
vessel per year are estimated to vary from USD 47,000 for tankers to USD 17,500 for
other ship types while the frequency of all inspections performed in the name of safety is
estimated to be at 11 inspections per year for tankers, 6 for dry bulk carriers and 5 for all
other ship types.
In addition, the inspection regimes do not accept each others inspections and in
particular, port state control inspections that are performed in another regime are not
taken into consideration by another port state control regime. The same applies for the
industry driven inspections where no information is exchanged between the vetting
inspection systems112. This leaves certain ship types to be exposed to a relatively large
amount of inspections where the inspections are performed sometimes during critical port
operations and increase the working hours of crew which further adds pressure and
fatigue to the overall work load in ports due to shortened port stays. The underlying
question is how the functioning of the safety regimes can be improved and how the money
112 CDI and OCIMF
172
allocated to port state control (which is understood to be the second line of defense in
eliminating substandard ships) can be better used.
One suggestion would be to improve the target factor of vessels by introducing a system
which spans across the regimes and which also enhances cooperation between the port
state control regimes. With the development of GISIS113 as a centralized database of port
state control data and casualty data, the combined data can provide a very useful and
overall picture of vessels. In addition, comments on inspections from flag states can also
be incorporated into the system and another group that could provide valuable feedback
would be the classification societies.
The role of classification societies is disputed due to the conflict of interest and its split
role of both public and commercial functions. Nevertheless and in particular with respect
to the statutory functions that are performed by class on behalf of the flag states, their
comments on inspections, and in particular follow up inspections of detained vessels can
all be valuable sources of information. The reason for allowing flag states or classification
societies to comment is to create a system that allows for follow up on inspections which is
lacking in the industry.
The overall aim is to decrease inspections on complying vessels and to increase the
frequency and time of inspections spent on substandard vessels in the areas where the
inspections are needed most and show the best effect on decreasing the probability of a
casualty. With respect to the casualty locations, high risk areas for the time frame 1999
to 2004 were West Africa, the Indian Ocean region, the North Atlantic East region and
the South China Sea. Regardless of the political dimension of port state control and the
underlying question of flag enforcement, increased cooperation in standardizing the
inspection procedures and providing training to emerging regimes should further speed
up the process. Developments into this direction are underway with the latest decisions
made at FSI (14)114 at IMO in June 2006.
In addition, about 47% of the world fleet is eligible for port a state control inspections
which excludes the fishing fleet. Descriptive statistics show a high amount of casualties
related to fishing vessels (approx. 7% of the world fleet). Incorporating the fishing fleet
into the safety regimes in an adapted inspection version to port state control should
enhance the safety culture onboard the fishing fleet and increase the overall safety level
of this industry. Comparison between the probability of casualty based on a certain flag of
the commercial fleet versus the fishing fleet has further shown that flag by itself is not
the sole indicator of sub-standardness but that many other variables play a role.
An interesting finding which further quantifies the effect of the relevant legislative
instruments on the probability of casualty is reflected by the result in relation to the
number of legal instruments a flag state has ratified and the number of legal instruments
a country of residence of the owner has ratified. While the first variable is significant for
all types of casualties (very serious, serious and less serious), the latter is only significant
for serious casualties. The associated effect is negative meaning that ratification has a
decreasing effect on the probability of casualty. To a certain extent, this means that
ratification has led to more enforcement and that this effect can be measured. On the
other hand, some other results of the analysis provide evidence that more improvement
113 Global Integrated Shipping Information System (IMO)
114 During FSI 14 (Flag State Implementation Sub-Committee meeting in June 2006, steps to
harmonize port state control have been indentified and a working group has been established to
deal with all PSC related issues in the future.
173
can be achieved. It will be interesting to see how the implementation of the IMO
Voluntary Member State Audit will enhance implementation of the conventions.
8.2. The Overall Magnitude for Improvement Possibilities
The variables which are used by the port state control regimes to target vessels for
inspection are similar variables that were used in the regressions to determine the
probability of detention and the probability of casualty. It can be confirmed, that the
classic variables such as ship type and size, age, flag, classification societies, the number
of deficiencies previously found in an inspection are all relevant and valid variables to be
used to target vessels for inspections. What remains to be seen, is how targeting can be
improved. First, an overall view of the picture will be presented and then the main
findings based on the probabilities will be shown.
To provide an overall view of the picture on inspections and casualties, Figure 105 shows
the amount of ships that were exposed to inspections and how they relate to casualties.
Figure 105: The Overall View on Inspections and Casualties (1999 to 2004)
inspected
with casualty
w/o time
connection
1.7%
inspected
6 m prior to
casualty
2.5%
not
inspected
with casualty
2.4%
not
inspected
without
casualty
23.5%
inspected
without
casualty
23.3%
not PSC
eligible fleet:
47% with
fishing
(53% w/o
fishing)
Total Ships: 93,719
Note: Casualties are over a time frame (1999 to 2004) and only PSC relevant casualties are shown
in this graph plus the fishing fleet (>400gt)
The fishing vessels (>400gt) of which the casualties are also considered are also
incorporated into this figure. The graph gives an overview of the total fleet where the
portion on top shows the portion not exposed to port state control minus a portion of the
fishing fleet (above 400gt) which is also considered in this study. The right hand lower
part of the graph represents inspected ships and the left hand side of the graph
174
represents ships that have not been inspected by the respective regimes115 or not
inspected at all. The lower middle portion summarizes the vessels that had casualties.
One can see that about an equal amount of ships that had not been inspected by the
regimes in question (they might have been inspected by another regime only) did not have
a casualty – about 46% of the world fleet including fishing vessels above 400gt. Not
inspected ships with casualties accounted for 2.4% of the world fleet versus 1.7% of the
vessels which had a casualty and were inspected without any related time frame and
2.5% of the vessels were inspected six months prior to a casualty.
What is interesting to notice but which is not shown in this graph is that based on the
individual port state control inspections, 54% of all inspections were inspections with zero
deficiencies while when aggregated by ship and taken as a summary of all inspections
performed on vessels, this percentage reduces to 16% of all inspections for the time frame
1999 to 2004. On the other hand, if only looking at the last inspection six months prior to
a casualty as shown in the figure with 2.5% of the total world fleet, 52.3% of these vessels
were ships with zero deficiencies.
It is difficult to give a conclusion on the target factor based on the percentages above. The
portion of ships which have been inspected can be understood as the ships that have been
targeted for inspections of which a certain portion was assumed to be sub-standard.
About 16% of all inspected vessels had zero deficiencies over the time period in question
and these ships might have been ships which should not have been targeted (4,221 ships).
On the other hand, looking at ships which have been inspected six months prior to a
casualty (2,321 ships) where 52.3% of these vessels had zero deficiencies (1,215 ships) and
the rest had deficiencies. This changes the 4,221 ships which should not have been
targeted into 3,006 vessels or approx. 501 ships per year.
It is further worth noticing that out of the 1,106 vessels (2,321 – 1,215) with deficiencies,
14.6% were detained (162 vessels) and had a casualty. The mean amounts of deficiencies
of these vessels are by 4.3 deficiencies for black listed flag states versus 2.7 deficiencies
for grey and 1.7 deficiencies for white listed flag states. Per seriousness of casualty,
detained vessels show significantly higher amount of deficiencies. This portion could be
understood as ships that have been targeted correctly and identified as sub-standard
vessels but for some reason, detention was not sufficient to increase the safety standard
of the vessel to prevent a casualty. The remaining portion of the vessels which have been
inspected and were deficiencies were found are the vessels where the effect of inspections
decreased the probability of a casualty which is the partial effect of the regressions in
part III of this thesis. In number of vessels, this amounts to approx. 18,874116 vessels or
3,146 ships per year. The left side of Figure 105 shows then the not inspected portion of
which 2.4% of the vessels had a casualty. In number of ships, this accounts for 2,213
vessels or approx. 369 ships per year. This is an area of improvement for targeting.
Figure 106 then visualizes the discussion above and presents a summary of the
magnitude of possible improvement areas for port state control. The figure is only based
on ships that are relevant for port state control (excluding the fishing fleet > 400gt) and is
a summary of the total time frame. The graph shows several groups out of which group 1
of about 36% of the vessels eligible for inspections are identified not to have been
115 As explained previously, some Port State Control regimes decided not to participate in this
study such as the Tokyo MoU, the Black Sea MoU or the Mediterranean MoU while others did not
have any data available yet.
116 21,880 total inspected ships with no casualty minus 3,006 ships with no deficiencies
175
problematic over the time period and have also not been targeted by the regimes in
question. About 7% of the vessels eligible for port state control have been targeted over
the time frame but did not have a casualty and also no deficiencies and therefore
represent a group of over-inspected vessels (group 2).
Figure 106: Improvement Areas for PSC eligible ships (1999-2004)
improve
targeting &
inspections
3.7%
improve
targeting
4.7%
improve
inspections
4.9%
group 3:
inspections
with effects
43.2%
group 2:
over
targeted
ships
6.9%
group 1:not
problematic
ships
36.5%
Total Ships: 43,817
group 4:
Note: Based on only PSC relevant ships and based on total time frame (1999-2004)
Group 3 consists of 43% of the vessels that can be identified to belong to a group where
inspections are effective in decreasing the probability of casualty where this effect is
strongest for very serious casualties and estimated (depending on the basic ship risk
profile) to be a 5% decrease per inspection. This category can also represent further room
for improvement but shows that port state control is effective.
Group 4 is split into three portions. The first portion is 4.9% of PSC eligible vessels which
are the amount of ships that have been targeted correctly but since they had a casualty
within six month after the inspection, the enforcement could be improved. The second
portion shows 4.7% of ships which had a casualty but were not inspected and where
targeting could be improved. Finally, the last category shows a grey area. In this group,
ships had a casualty but regardless of the time frame. Therefore, inspections and possibly
targeting could be improved. Most improvement to decrease the probability of a casualty
can be achieved by concentrating on the categories in group 4 by shifting the emphasis
from group 2 to group 4.
Taking the probabilities of casualty into account, the probability of casualty changes per
ship type and confirms that general cargo vessels are ships with the highest probability of
a casualty which is confirmed by the probability of detention. Black listed flag states or
non inspected ships show a higher probability of a very serious casualty compared to grey
and white listed flag states while the same does not hold for serious and less serious
casualties. It confirms that the target factors are targeting high risk vessels but are less
effective in decreasing the probability of a serious and less serious casualty.
176
Overall, one can identify areas for improvement of either the targeting of sub-standard
ships or the inspections itself. The next chapter will give recommendations on how to
improve the target factor but will also summarize the effects of inspections on the
probability of casualty before the last chapter will give recommendations on how to
improve the inspections.
8.3. Effect of Inspections and Improvement for Targeting
One could argue that port state control works best as a regional agreement. The basic
ship profiles given by age, size, flag, class and ownership do not vary significantly across
the regimes with respect to the probability of detention. Most differences across the
regimes with respect to the probability of detention can be found within the use of the
deficiency codes towards detention and the port states. When combined by ship types, the
differences average out but looking at the ship types individually, one can see that certain
codes show higher contributions compared to each other within each of the regimes.
The basic ship risk profile based on the probability of detention (without taking the effect
of deficiencies into account) for all regimes is between 0.5% to 1.5% while the basic risk
profile based on the probability of casualty reveals that the average probabilities of a
casualty are by 0.06% (very serious), 1.6% (serious) and 1% (less serious) respectively.
With reference to ship types, general cargo vessels seem to show the highest probability of
a very serious and serious casualty of 1% and 2% respectively followed by passenger
vessels for very serious casualties. Tankers and container vessels are below 0.5% for very
serious casualties and 1% for serious casualties. Container vessels show the lowest risk in
all three types of casualties. Comparing ships that have been inspected with ships that
have not been inspected by the respective regimes, the strongest difference can be seen
with general cargo vessels and dry bulk carriers for very serious casualties and least of
the effect can be seen with tankers and container vessels which is not surprisingly for
tankers due to the amount of vetting inspections that are performed on oil and chemical
tankers.
With respect to average insurance claims, tankers and passenger vessels show the
highest average claim amount while general cargo and containers show the lowest for the
time period 1999 to 2004. The probability itself describes the risk associated with a
particular ship type and for very serious casualties, this means loss of life, complete loss
of the vessel or substantial pollution. According to the claim costs above the deductible,
tankers still remain a high risk vessel due to the higher costs associated with a casualty
despite the fact that the probability is rather low. One can further see a significant
difference between insurance claim figures of port state control inspected vessels versus
not inspected vessels and that the difference is in particular significant for tankers and
passenger vessels. In addition, the regressions of the probability of casualty has shown
that some parameters of the variables used in the analysis show a positive effect towards
the probability of a very serious, serious or less serious casualty, some show a negative
effect and some are not significant.
If a ship changed its classification society during its course of life, the partial effect is
negative on the probability of casualty for all casualty types. This could mean that in
general, if a class is changed, an inspection is performed which might have a positive
influence on the quality of the vessel. On the other hand, a ship whose class is withdrawn
shows a higher probability of all types of casualties. Both variables can relatively easy be
incorporated into the targeting matrixes of port state control regimes. With respect to
177
classification societies, the probability of a casualty for Non-IACS class is higher than for
IACS class and is also confirmed by the probability of detention.
On the other hand, change of flag does not seem to be significant while change of
ownership shows a strong positive effect towards the probability of casualty for all types.
This could indicate a change of a vessel into second hand ownership or into a segment of
the market where less money is spent on safety. Again, this information besides the
actual information to incorporate the beneficial owner or DoC117 Company into a
targeting matrix for inspection can easily be obtained and incorporated. For ownership
groups and in comparison of inspected to non–inspected vessels, highest probability of
casualty lies within owners from open registry countries followed by unknown owners,
owners from traditional maritime nations and emerging maritime nations. This does not
follow the probability of detention where owners from traditional maritime nations show
the lowest probability of detention.
The coefficient for the variable double hull is not significant for any type of casualty.
Some classification societies, flag states, ownership groups and ship yard countries
remain to be significant with either a positive or negative effect on the probability of
casualty. Age only remains significant for very serious casualties and as the age of the
vessel increases, the probability of having a very serious casualty increases by about 12%
over a 35 year period which translates into about 0.35% per year. Tonnage is also only
relevant for very serious casualties but is negative indicating that a smaller vessel seems
to be at higher risk than larger vessels which goes in line with the general cargo vessels
being more high risk prone.
Easily incorporated into the target factor can be a variable indicating if a ship has been
Greenaward certified or inspected by Rightship. Although partial effects for Greenaward
certified ships cannot be measured which might be due to lack of data, the average
probabilities of a very serious casualty is clearly lower than for non certified vessels.
Based on the inspection data, the mean amount of deficiencies is lower for Greenaward
certified vessels than for non certified vessels. As for ships inspected by Rightship, the
partial effect can be measured and the star ranking can be confirmed.
The parameter for detention is not significant towards the probability of casualty but
inspections are. The fact that it is not significant and not negative (as expected) shows
that its effect cannot be measured and might be overruled by the effect of inspections in
general since only very few vessels are detained within a certain time frame and prior to
a casualty (162 vessel). In addition and based on average probabilities, the probability of
casualty increases by 0.1% as the number of deficiencies increases. This is irregardless of
the time frame of the inspection and is based on the total inspection time frame of about
six years.
The parameter obtained for the variable indicating the time span in-between inspections
is not significant for very serious casualties but is for less serious and serious casualties.
On average and regardless of the seriousness of casualty, the probability increases by
2.3% within the time frame of one year. For fire and explosion, this increase can be 2.7%
and 0.5% for engine related casualties. Time span might therefore be important to further
improve the efficiency of port state control with respect to serious and less serious
casualties.
117 Document of Compliance Company (the company which is according to ISM, the responsible
company for the safety management onboard a vessel)
178
Overall, one can conclude that the classic variables such as ship type, age, size, flag, the
classification society, deficiencies found in prior inspections and detention are all valid
variables for targeting sub-standard ships for inspections. However, the targeting of ships
can further be improved in order to capture a certain amount of vessels which are not
inspected but have a casualty. Improvement can be made by adding the variable
indicating the ownership of a vessel and certain data on ship history such as change of
class, class withdrawal and change of ownership.
Another possibility would be to include where the ship was built and if possible, if a ship
had undergone vetting inspections. The ship profile can change over time and targeting
factor therefore needs to be dynamic and adaptable from time to time. At first sight, it
seems that too many ships with zero deficiencies are targeted but when aggregated by
ship over the whole time period, the percentage of ships with zero deficiencies reduces
significantly but a certain group of vessels which are over-inspected are still identified
and efforts to inspect those vessels should be shifted towards the group 4 identified in
Figure 106.
Targeting on combined data as described in the previous chapter by making inspections
available to all regimes can further enhance the overall targeting and concentrate on high
risk vessels versus ships that comply. It allows gaining a better picture of a vessel
inspection history over time and can help to identify sub-standard vessels, even if there
are differences in the inspections which were confirmed by the probability of detention
and the parameters that were obtained for the deficiency codes and the port states. It can
also help in identifying if performances of flag or classification societies vary significantly
across the globe which this study in its results cannot support as no significant difference
with respect to the probability of detention could be identified in part II of this thesis.
8.4. Identified Areas for Improvement of Inspections
The last area will summarize the findings with respect to deficiencies for either the
probability of detention or casualty which could be identified in Figure 106 since some
ships are targeted correctly for inspections but the effect of the inspection cannot be
measured as being effective enough. The two groups of casualties are primarily serious
and less serious casualties where serious casualties are seen to be more relevant for port
state control.
The probability of detention models revealed the highest contribution of deficiencies in
the areas of certificates, ship and cargo operations, the ISM118 code and safety & fire
appliances while lowest contribution is found for machinery and equipment. The
probability of casualty either per seriousness of casualty or casualty first event also
revealed three areas of interest – the ISM code, ship and cargo operations and machinery
and equipment. Those are the main areas which have been identified were room for
improvement exists.
What is interesting to notice is the relative high contribution of ISM related deficiencies
for some ship types and regimes for the probability of detention and the positive effect of
ISM related deficiencies towards the probability of casualty. ISM related deficiencies
contain items such the safety and environmental policy, the definition of company
responsibilities and the master’s responsibility, deficiencies in the area of shipboard
operations, emergency preparedness, reporting and analysis of non-conformities,
118 International Safety Manamagent Code
179
accidents and near misses, the maintenance of the vessel and company audits. Another
area deals with resources and personnel. Its effect is positive for very serious and serious
casualties on general cargo vessels and tankers and for casualty first event collision and
contact. It could mean that even though port state control detects deficiencies in the
implementation of the ISM code onboard, the rectification or follow up on the deficiencies
is not very successful. Besides ISM, structural safety also shows a positive effect for very
serious and serious casualties on container vessels and for less serious casualties on all
vessels. This could further reflect poor implementation of maintenance programs which
are part of the ISM system.
The author can fully support the recommendations given in MSC 81/17/1 which includes
a more systematic training, increase monitoring of compliance as well as involving
seafarers into the development of ISM manuals. In addition, Argentina presented at the
last Sub Committee on Flag State Implementation (FSI 14)119 in June 2006 a proposed
IMO model course for training of ISM auditors and has established an inter-sessional
group to further develop the course. The author finds that such model course should also
be made available to port state control officers in the future for additional training. Lack
of recognition by port state control does not seem to be the real problem. The
improvement with respect to decreasing the probability of casualty would be by improving
enforcement of the code. More detailed audits should be requested and information should
be made available to flag state and recognized societies for all inspections and deficiencies
related to ISM and not only detentions. In addition, further research in this area
especially with reference to audits performed by recognized organizations and on DoC
companies can be beneficial in this respect. With this respect, the revision of the port
state control deficiencies is a very welcoming development which will hopefully in the
future allow better analysis in this area.
While ISM is more related to the system’s side of the safety management onboard a
vessel, deficiencies in the area of ship and cargo operations include all the relevant
operational aspects of the ISM code. For the probability of detention, ship and cargo
operations seem to be more important for tankers while the effect of deficiencies of
stability and structure are highest for dry bulk carriers and containers. With respect to
the probability of casualty, this group of codes can vary considerably. Negative effects can
be found with code 2000 (SOLAS operational related deficiencies) for general cargo ships
and for engine related first events for all ship types as well as with code 2300 (MARPOL
Annex V: Garbage) for less serious casualties and with code 2100 (MARPOL related
operational deficiencies) for first events collision & contact – again for all ship types. On
the other hand, positive effects can be found with code 1700 (MARPOL Annex I: Oil
Pollution) for less serious casualties and engine related first events, code 1800 (oil,
chemical tankers & gas carriers) for first event fire & explosion and with code 1900
(MARPOL Annex II: Noxious Liquids) for deck related first events.
If the effect is negative, effective implementation and rectification can be found on board
such as deficiencies in the area of code 2000 which also includes drills. Therefore, in
shifting the emphasis of inspections from over-inspected ships to more substandard ships,
inspections could be expanded to include drills also which is for instance the case with the
USCG but is not done on a regular basis in other regimes. Rooms for improvement can
also be found in rectifying or following up deficiencies in the area of MARPOL Annex I,
code 1800 (oil, chemical tankers & gas carriers) and in the area of MARPOL Annex II. As
for fire and explosions on chemical tankers, the area has already been identified and been
119 FSI 14/3/1, agenda item 3, Assessment of model training courses, Model Course: Safety
Management System Auditor (ISM Code), 21 February 2006
180
dealt with during MSC (81) in May 2006. This finding confirms that there is problem with
enforcing the legal conventions on chemical carriers but that the main contributor was
identified to be human error by not following the proper procedures which is also reflected
in the positive contribution of the ISM related deficiencies for tankers.
Linking this with the findings of the probability of detention, one could further conclude
that there is room for improvement in the area of ISM related deficiencies as well as in
the area of ship and cargo related deficiencies and in particular, the proper
implementation of procedures onboard as well as the follow up by flag states, recognized
organizations who perform ISM audits and the DoC companies. Port state control can
further improve the system by making information available to all inspectors.
The last identified area for improvement for inspections is machinery related items such
as propulsion and aux. engines. In this area, lowest contribution towards the probability
of detention is found for the deficiency codes associated with machinery and equipment
while about 32% of all port state control relevant casualties between 1999 to 2004 show
signs of a casualty first event in engine related areas. With respect to the probability of
casualty, Code 1400 (propulsion & aux. engine) has a positive effect for serious and less
serious casualties, in particular for dry bulk carriers and for casualty first events such as
fire & explosion, wrecked, stranded or grounded and engine related casualty first events.
On could argue that it is not easy to find deficiencies in this area during port state control
inspections since during port time, the engine is not up and running. However, it seems
that deficiencies are found which might indicate a possible problem but due to limited
testing possibilities in port and possibly due to lack of enforcement or follow up on
deficiencies, the effect of such deficiencies is positive rather than negative towards the
probability of casualty.
For casualty first event wrecked, stranded or grounded, code 1000 (alarm signal) shows a
relative strong positive effect on the probability of having a casualty. With respect to the
probability of detention, the deficiency group navigation and communication which
includes code 1500 but also code 1600 show the highest variety across the regimes while
the deficiency group safety & fire which includes the alarms does not vary much across
the regimes. A simple recommendation can be to perform a concentrated inspection
campaign on this topic to see if there is any other latent problem.
As a last word of recommendation, the author feels that emphasis on working and living
conditions including working and rest hours onboard vessels should be increased. While
the only group of codes related to this group of deficiencies is code 400 (food and catering),
the parameter of this code shows a negative effect on the probability of a very serious
casualty as well as towards the probability of an engine related first event. This study did
not concentrate on the human element but can support any development in this area. The
effect of inspections is only negative for very serious casualties and slightly positive or
none existing for serious and less serious casualties and a higher portion of human
related errors are associated with them which are more difficult to detect through port
state control normal practices. The same applies for vetting inspections.
8.5. Suggestions for Further Research
In organizing this project and writing this thesis, the author came across several barriers
which could not be eliminated but provided an opportunity for new ideas to conduct
research in the future.
181
One of the difficulties which provided limitations to this study was the inability to obtain
data from some of the port state control regimes, namely the Tokyo MoU, the Black Sea
MoU and the Mediterranean MoU. A very valid suggestion for future research which can
easily be done through GISIS120 is to include port state control data from all regimes
which will further refine the findings of part III of this thesis. Another barrier to this
study was the inability to obtain data from two of the vetting inspection regimes, namely
CDI and OCIMF. Future research in this area could be the incorporation of this data and
relate it to port state control data.
A further refinement to the analysis could be to split up the variable indicating the
classification society into their respective areas of responsibility onboard since a vessel
can have up to three different classification societies. By doing so, a more refined view on
the performance of classification societies can be obtained. While this information might
not be possible to receive for all casualty data, it is available for port state control data.
With respect to change of class, one could further indicate whether the change was
between two IACS class member or not.
With this in mind, ISM audit reports and timings or time elapsed between inspections
and annual or periodical surveys could also be incorporated into the analysis in order to
obtain the effect of the timings of the surveys performed onboard. Finally, a separate
analysis can be performed based on data from classification societies which provides more
detailed information then port state control inspection data. Any future research should
try to measure the actual impact of ISM on the probability of casualty. This study could
not provide a negative relationship – on the contrary, it has shown that there is lack of
implementation and enforcement of ISM in the industry.
With respect to ship types and if the amount of data can be increased, separate
regressions could be performed per ship type and seriousness or casualty first events
which will further refine the results, especially with respect to the deficiency codes. A
separate analysis for all passenger vessels including smaller ferries and fishing vessel
including smaller ships (below 400gt) could also be performed in the future to obtain a
better view of these two areas since both ship types are of special interest.
Finally, one could use duration models based on a more extended time frame and look
into changes of probabilities over time for either the probability of detention or the
probability of casualty. Another area would be to find the optimum global targeting factor
while allowing for some regional differences. Ideally, the target factor would be
constructed on combined data and would take a vessel’s history into account where maybe
changes in probabilities are also considered. The optimum amount of inspections could
also be a further topic of interest.
In the area of port state control, further research in order to try harmonizing port state
control procedures, the impact of an inspector background of cultural surroundings would
be beneficial to bring some more insight into this area. While this study has tried to look
at it briefly and could identify some differences with respect to the probability of
detention, it did not provide a detailed insight into the background of inspectors across
regimes. The same could be performed for vetting inspectors or class surveyors with the
respective datasets. Furthermore, with increased amount of data such as casualties and
inspections over a longer time period, deficiency codes could further be broken down into
the sub-codes. In addition, more research could be conducted with respect to follow up
120 Global Integrated Shipping Information System (IMO)
182
actions to see how harmonization can benefit the follow up on inspections and decrease
the probability of casualty.
Last, not least but not minder important, more research should be conducted which takes
the human factor into account which should be possible once the correct type of data is
available.
183
Summary in Dutch (Nederlandse Samenvatting)
Dit proefschrift moet gezien en begrepen worden als een eerste poging om te kijken naar
port state control (PSC) op mondiale schaal door het effect te meten van inspecties op de
waarschijnlijkheid van ongelukken en door gebieden voor verbetering te onderkennen.
Wat nieuw is in dit proefschrift is de combinatie van PSC-gegevens van verschillende
regimes en gegevens van ongelukken afkomstig uit drie verschillende bronnen in dezelfde
tijdsspanne. De corresponderende onderzoeksvragen zijn: 1) Wat is de huidige toestand
van het veiligheidsregime? 2) Kan de selectie voor inspectie verbeterd worden? 3) Wat is
het effect van inspecties op ongevallen? 4) Hoe kunnen inspecties verbeterd worden? Het
onderzoek is gebaseerd op ongeveer 183.000 PSC inspecties en 11.000 ongevallen tussen
1999 en 2004. Het zal hopelijk tot een nieuw hoofdstuk leiden in het onderzoek op het
gebied van maritieme veiligheid. Maritieme veiligheid waarvan de toekomstige
mogelijkheden, wanneer politieke barrières zijn geslecht en meer transparantie
geaccepteerd is, niet genegeerd zouden moeten worden door zowel de bedrijfssector als de
regulerende instanties. Verscheidene econometrische technieken worden gebruikt om een
kansberekening van aanhouding en ongelukken te maken naar ernst of “eerste incident” –
ongevallen. De auteur kijkt niet naar politieke dimensie van vlagstaat-implementatie of
PSC maar concentreert zich alleen op de technische aspecten van het onderwerp in
kwestie.
De maritieme sector wordt gekenmerkt door een uitgebreid juridisch kader gebaseerd op
internationale wetten die beperkte wettelijke bevoegdheden voor handhaving bieden in
geval van overtreding. Dit creëert mazen in de regelgeving en concurrentievervalsing
door het bestaan van een markt met inferieure schepen. Vanuit publiek oogpunt gezien,
is de gewenste situatie er één waarbij veilig en milieuvriendelijk vervoer en vermindering
van het aantal inferieure schepen gestimuleerd wordt. Vlagstaten moeten worden
beschouwd als de eerste verdedigingslinie bij het elimineren van inferieure schepen,
gevolgd door de tweede verdedigingslinie, de havenstaten. Het gebrek aan vertrouwen in
de sector heeft speelruimte gecreëerd voor inspecties van bepaalde scheepstypen inclusief
een aanzienlijke hoeveelheid door de sector geëntameerde inspecties, waarin het totaal
aan inspecties geschat wordt op 11 inspecties per jaar voor tankschepen, 6 voor drogebulk
schepen en 5 voor alle andere scheepstypen.
Twee gebieden waarin mogelijk een verbeteringsslag kan worden gemaakt zijn
onderkend: 1) de selectie van inspecties en 2) de inspecties en het opvolgen van gebreken.
Op het eerste gezicht lijkt het dat teveel schepen met nul gebreken worden uitgekozen
maar na samenvoeging van schepen en regimes, neemt het percentage schepen met nul
gebreken aanzienlijk af over de gegeven tijdsperiode. Desondanks is bij een bepaalde
groep schepen (ongeveer 7 % van schepen die voor PSC in aanmerking komen) onderkend
dat deze teveel geïnspecteerd worden, en zullen er pogingen moeten worden ondernomen
om inspectie inspanningen te verschuiven naar de groepen van schepen die baat hebben
bij een inspectie. Dit wordt geschat op ongeveer 14 % van alle schepen die voor PSC in
aanmerking komen gebaseerd op de tijdsperiode die gebruikt is voor deze analyse. Het
effect van PSC-inspecties op de kans op ongevallen kan gemeten worden voor “zeer
ernstige” ongevallen maar niet voor “ernstige” en “minder ernstige” ongevallen.
Afhankelijk van het algehele risicoprofiel van een schip, kan een inspectie de kans op een
“zeer ernstig ongeval” mogelijk met ongeveer 5 % per inspectie verminderen, waarbij het
effect zelfs 10 % kan zijn voor schepen met een hoge risicofactor.
184
Hoewel de sleutelgetallen over tekortkomingen en aanhoudingen variëren per regime,
worden de verschillen in de kans op aanhouding alleen maar weerspiegeld in de
verschillen in havenstaten en de behandeling van tekortkomingen en niet
noodzakelijkerwijs door ouderdom, grootte, vlag, klasse of eigenaar. De selectiefactor kan
verbeterd worden door ontwikkeling van een factor die gebaseerd is op gecombineerde
inspectie datasets die rekening houden met de gehele inspectie geschiedenis van schepen
en door rekening te houden met regionale verschillen. In de periode van 1999 tot 2004
werden West Afrika, het gebied rond de Indische Oceaan, het oostelijk deel van het
Noord-Atlantische gebied en de Zuid-Chinese Zee, aangemerkt als gebieden met een hoog
risico. Het verschil in het effect van inspecties op de kans op “ongevallen naar ernst” of
“eerste incident”- ongevallen bevestigt bovendien een verschuiving van inferieure schepen
van gebieden zoals het Paris MoU en de USCG naar andere gebieden van de wereld, zoals
het Zuid-Amerikaanse gebied, het gebied rond de Indische Oceaan of Australië. Inspecties
van deze gebieden blijken de kans op een ongeval te doen afnemen bij een significantie
niveau van 1%.
De klassieke variabelen zoals scheepstype, leeftijd, grootte, vlag, de classificatie-indeling,
tekortkomingen gevonden in eerdere inspecties en aanhoudingen zijn allemaal geldige
variabelen voor het doelbewust selecteren van inferieure schepen. Vlag is maar een
variabele uit de vele die kunnen worden gebruikt voor de selectie op inferieure schepen.
Leeftijd blijft alleen significant voor “zeer ernstige” ongevallen. Naarmate het schip ouder
wordt, stijgt de kans op een “zeer ernstig” ongeval met ongeveer 12 % over een periode
van 35 jaar, hetgeen zich vertaalt in ongeveer 0.35% per jaar. De waarschijnlijkheid van
een ongeval bevestigt bovendien dat general-cargo schepen, schepen zijn met de hoogste
kans op een ongeval, hetgeen tevens bevestigd wordt door de kans op aanhouding.
Vlagstaten op de zwarte lijst of niet geïnspecteerde schepen laten een hogere kans zien op
een “zeer ernstig” ongeval in vergelijking met de vlagstaten op de grijze en witte lijst. Dit
geldt niet voor “ernstige” en “minder ernstige” ongevallen. Daarentegen wijzen de
gemiddelde kosten van een verzekeringsclaim uit dat de hoogste claimkosten gerelateerd
zijn aan tank- en passagiersschepen
Verdere verbeteringen voor de selectie van inferieure schepen kunnen gemaakt worden
door het toevoegen van de variabele die het eigendom of “DoC bedrijf”121 van een schip
weergeeft en specifieke data van de geschiedenis van een schip zoals verandering of
intrekking van classificatie, verandering van eigenaar over een zekere periode of de
locatie waar een schip gebouwd is. Al deze variabelen hebben een negatief of positief
effect op de kans op een ongeval. Een andere mogelijkheid zou zijn om een variabele toe te
voegen die aangeeft of het schip geïnspecteerd is door één van de vetting inspectie
regimes (in geval van droge bulk) of gecertificeerd is door de Greenaward Foundation.
Nauwkeuriger onderzoek van de effectiviteit van inspecties wijst uit dat aanhoudingen
niet significant lijken te zijn voor de kans op een ongeval, hetgeen een verassend
resultaat is. Dit betekent niet persé dat een aanhouding irrelevant is, maar eerder dat
misschien de inspectie zelf het effect met zich meebrengt. Tevens is de tijdspanne tussen
inspecties niet significant voor “zeer ernstige” ongevallen maar wel voor “minder ernstige”
en “ernstige” ongevallen. Gemiddeld en onafhankelijk van de ernst van het ongeluk stijgt
de kans binnen een jaar met 2,3%.
121 “Document of Compliance Company”, is het aangewezen bedrijf dat verantwoordelijk is voor
veiligheidsmanagement aan boord van schepen aan de hand van de “International Safety Management
Code (ISM)”
185
Het risicoprofiel van schepen over alle regimes ligt tussen 0.5% en 1.5% voor de meeste
scheepstypen terwijl de gemiddelde kans op een ongeval, bij elkaar voor alle
scheepstypen, 0.06% is voor “zeer ernstige”,1.6% voor “ernstige-” en 1% voor “minder
ernstige-” ongevallen.
De waarschijnlijkheid van aanhoudingsmodellen wijst uit dat het grootste aandeel aan
gebreken die tot aanhoudingen leiden ligt op het gebied van certificaten, scheeps- en
ladingsoperaties, de ISM122-code en veiligheids- en blusapparatuur, terwijl het kleinste
aandeel op het gebied van machines en apparatuur blijkt te liggen.
De kans op een ongeval ofwel op basis van “ernst van ongeval” danwel “eerste incident
ongeval” heeft ook drie aandachtsgebieden blootgelegd – de ISM-code, scheeps- en
ladingsoperaties en machines en apparatuur. Dit zijn de belangrijkste gebieden die
onderkend zijn, waar ruimte voor verbetering van PSC-inspecties bestaat zodat de kans
op een ongeval afneemt.
122 International Safety Management Code
186
187
Summary in Spanish (Resumen en Español)
La presente tesis debe ser vista y comprendida como un primer intento por analizar el
control por el Estado rector del puerto (ERP) a escala global, midiendo los efectos de las
inspecciones sobre la probabilidad de siniestros e identificar áreas para su mejora. Lo
novedoso de esta tesis es la combinación de datos sobre control por el Estado rector de
varios regímenes por un lado y datos de siniestros tomados de tres fuentes diferentes
dentro de un mismo marco cronológico por el otro. Los planteos de investigación
correspondientes son: 1) ¿En qué estado se encuentra el actual régimen de seguridad?, 2)
¿Puede mejorarse el factor de prioridad para las inspecciones de los buques?, 3) ¿Cuál es
la repercusión de las inspecciones en los siniestros? y 4) ¿Cómo pueden mejorarse las
inspecciones? Para el análisis se han utilizado tres grupos de datos de aproximadamente
183.000 inspecciones de varios Memorandos de Entendimiento (MoU123) para el período
comprendido entre enero de 1999 y diciembre de 2004 y de aproximadamente 11.000
registros para el período 1993 – 2004.
La industria marítima se caracteriza por tener un marco legal muy engorroso basado en
el derecho internacional con limitadas posibilidades de aplicación de la ley en caso de
incumplimientos. Esto crea vacíos legales y distorsión de la competencia debido a la
existencia de un mercado de buques subestandar. Desde una perspectiva pública, la
situación deseada es promover un transporte marítimo seguro, protegido y compatible con
el medio ambiente y reducir la cantidad de buques subestandar. Los Estados de
abanderamiento deben ser vistos como la primera línea de defensa en la eliminación de
este tipo de buques, seguidos por la segunda línea de defensa, los Estados rectores del
puerto. La falta de confianza en la industria ha creado un campo propicio para las
inspecciones a ciertos tipos de buques incluyendo una considerable cantidad realizadas
por la propia industria estimándose un total de 11 por año para los buques tanque, 6 para
los graneleros y 5 para los restantes tipos de buques.
Se han identificado dos áreas de mejora potencial: 1) el factor de prioridad de los buques
para inspección y 2) las inspecciones y el seguimiento de las deficiencias mismas. A
primera vista, parece que se priorizan muchos buques con cero deficiencias pero al
sumarlos por buque y cruzarlos con los regímenes, el porcentaje de buques con cero
deficiencias se reduce significativamente a lo largo del período dado. Sin embargo, se ha
detectado que cierto grupo de buques (alrededor del 7% de los buques elegibles por el
ERP) está sobre inspeccionado y deberían dirigirse los esfuerzos hacia los grupos de
buques que pueden resultar beneficiados de una inspección. Basado en el período usado
para este análisis el grupo que puede resultar beneficiado se estima en 14% de todos los
buques elegidos para control por el Estado rector. La repercusión de las inspecciones por
el Estado rector en las probabilidades de siniestros puede ser medida para los casos de
siniestros muy graves pero no para los graves o menos graves. Dependiendo del perfil de
riesgo general de un buque, una inspección puede potencialmente reducir la probabilidad
de tener un siniestro grave en un 5% aproximadamente, mientras que la repercusión
puede alcanzar un 10% para buques de alto riesgo.
Si bien las principales cifras sobre deficiencias y detenciones varían según cada régimen,
las diferencias respecto de la probabilidad de detención se ven condicionadas por
diferencias entre los Estados rectores y el trato de las deficiencias y no necesariamente
123 Un memorando de entendimiento (MOU) es un documento legal que describe un acuerdo entre
partes pero es menos formal que un contrato.
188
por la edad, las dimensiones, la bandera, la clase o el propietario. El factor de prioridad
puede verse mejorado al desarrollar un factor de prioridad sobre datos de inspección
combinados tomando en cuenta el historial total de inspección del buque y considerando
las diferencias regionales dado que se han identificado zonas de alto riesgo para el período
1999 – 2004, como África Occidental, la región del Océano Índico, la del este del Océano
Atlántico Norte, y el Mar del Sur de China. La diferencia en la repercusión de las
inspecciones sobre la probabilidad de siniestros por casos graves de siniestros confirma un
cambio en el rumbo de los buques subestandar de zonas como las del MOU de París y la
Guardia Costera de los EE.UU. a otras zonas del globo, como las regiones de Sudamérica,
el Océano Índico o Australia. Las inspecciones de estos regímenes han mostrado reducir
la probabilidad de sufrir un siniestro en un nivel de importancia del 1%.
Las variables clásicas como tipo de buque, edad, dimensiones, bandera, sociedad de
clasificación, deficiencias encontradas en inspecciones previas y detenciones son todas
válidas para priorizar buques subestandar. El abanderamiento es solo una variable de las
tantas que pueden ser usadas para priorizar buques subestandar. La edad solo sigue
siendo relevante para los siniestros muy serios y a medida que la edad sube, la
probabilidad de sufrir siniestros muy serios se incrementa en un 12% en un período de 35
años, el cual se traduce en un 0,35% por año. La probabilidad de siniestros también
confirma que los buques de carga general son los que muestran la mayor probabilidad de
siniestro, lo cual es confirmado por la probabilidad de detención. Los buques no
inspeccionados o cuyas banderas figuran en listas negras muestran una mayor
probabilidad de sufrir un siniestro muy grave comparados con los buques de banderas que
aparecen en listas grises o blancas, aunque no se aplica lo mismo para siniestros graves o
menos graves.
Otras mejoras en la priorización de buques subestandar pueden conseguirse al agregar la
variable que indica el propietario o la Compañía que figura en el DoC124 del buque y
ciertos datos sobre el historial del buque, como cambios de clase, retiro de clase y cambio
de propiedad en el tiempo o el lugar donde el buque fue mayormente construido, todo lo
cual ha demostrado tener un efecto positivo o negativo sobre la probabilidad de siniestros.
Otra posibilidad sería incluir si un buque ha sido inspeccionado por uno de los regímenes
de exámenes previos (para cargas secas a granel) o certificado por la Greenaward
Foundation.
Una visión más detallada sobre la efectividad de las inspecciones revela que la detención
no parece ser importante para la probabilidad de sufrir un siniestro, lo cual es un
resultado sorprendente. Ello no significa necesariamente que la detención no sea
relevante sino que el efecto quizá pueda ser percibido por la inspección. Asimismo, el
lapso entre inspecciones no resulta de importancia para siniestros muy graves, pero sí lo
es para los graves o menos graves. En promedio y sin tomar en cuenta la gravedad del
siniestro, la probabilidad se incrementa en 2,3% en el plazo de un año. El perfil de riesgo
básico de los buques en todos los regímenes se halla entre 0,5% y 1,5% para la mayoría de
los tipos de buques y regímenes, mientras que la probabilidad promedio de siniestros
sumada para todos los tipos de buques se encontró en 0,06% para los siniestros muy
graves, 1,6% para los graves y 1% para los menos graves.
Los modelos de probabilidad de detención develaron la más alta contribución de
deficiencias en las detenciones en las áreas de certificados, operaciones de buque y de
124 Compañía que figura en el Documento de Cumplimiento, la compañía designada responsable de
la gestión de la seguridad a bordo del buque de acuerdo con el Código Internacional de Gestión de
la Seguridad (IGS)
189
carga, el código IGS125 y seguridad y lucha contra incendios, mientras que la menor
contribución se encuentra en máquinas y equipos. La probabilidad de siniestros por
gravedad del mismo o por ocurrir por vez primera, también ha mostrado tres áreas de
interés – el código IGS, las operaciones de buques y de carga y maquinas y equipo. Esas
son las principales áreas identificadas donde existe lugar para una mejora a fin que las
inspecciones por el Estado rector disminuyan la probabilidad de siniestros.
125 Código Internacional de Gestión de la Seguridad
190
191
References
Most of the information which has contributed to this thesis derived from interviews,
observations during ships inspections and attendance at sub-committee and committee
meetings at IMO (International Maritime Organization) where some papers used are not
accessible to the public or only at a later stage in the legislative process. In order to bring
some structure to the different types of information sources used in this thesis, the list of
references is split into different sections. The order of appearance in the list is given
based on public accessibility of the resources.
Books and Articles
Alderton T. and Winchester N (2002). “Flag States and Safety: 1997-1999”. Maritime
Policy and Management, Vol 29, No. 2, pp 151-162
Clausen, S. E. (1998). Applied Correspondence Analysis – An Introduction (Sage
University Papers Series on Quantitative Applications in the Social Sciences, series no.
07-121). Thousand Oaks, CA: Sage
Cramer, M. , Franses P.H. and Slagter, E. (1999). “Censored Regression Analysis in Large
Samples with Many Zero Observations”, Econometric Institute Research Report 9939/A
Davidson and McKinnon (1993). “Estimation and Inference in Econometrics”, New York:
Oxford University Press, 1993
Franses, P.H. and Paap, R. (2001). Quantitative Models in Marketing Research.
Cambridge University Press, Cambridge
Greene H.W. (2000). Econometric Analysis, Fourth Edition, Econometric Analysis,
Prentice Hall, New Jersey
Goodwin, S. (2006), “Does Work Keep you Awake at night ?” Learning from Marine
Incidents III, Conference, The Royal Institute of Naval Architects (January 2006: London,
UK)
Harvey, A. (1976) “Estimating Regression Models with Multiplicative Heteroscedasticity.”
Econometrica, 44, pp.461-465
Hill, C., Griffiths, W. and Judge G. (2001). Undergraduate Econometrics, 2nd Edition, New
York: John Wiley & Sons, Inc.
Hosmer, D. and Lemeshow S. (1989). Applied Logistic Regression. New York: John Wiley
& Sons
Keller, G. and Warrack, B. (2003). Statistics for Management and Economics, 6th Edition,
USA: Brooks/Cole – Thomson Learning
Knapp, S. (2004), Analysis of the Maritime Safety Regime – Risk Improvement
Possibilities of the Target Factor (Paris MoU), Master Thesis, Erasmus University,
Rotterdam
192
Soma, T. (2004), Blue-Chip or Sub-Standard, Doctoral Thesis, Norwegian University of
Science and Technology, Trondheim
Spence, L. and Vanden, E. (1990). Calculus with Applications to the Management, Life
and Social Sciences. Illinois: Scott, Foresman/Little, Brown Higher Education (A Division
of Scott, Foresman and Company)
Talley, WK., Jin D and Kite-Powell, H. (2001). “Vessel accident oil-spillage: post US OPA
90”. Transportation Research Part D, 6: 405-415
Talley, WK. (2004). “Post OPA-90 vessel oil spill differentials: transfers versus vessel
accidents”. Maritime Policy and Management, Vol.31, No.3: 225-240
Talley, WK (2002). “Vessel Damage Cost Differentials: Bulk, Container and Tanker
Accidents”. International Journal of Maritime Economics, 4: 307-322
Talley, WK (1999). “Determinants of the property damage costs of tanker accidents”
Transportation Research Part D, 4: 412-436
Talley, WK (1999). “The safety of sea transport: determinants of crew injuries”. Applied
Economics, 31: 1365-1372
Talley, WK (2002). “Non-Seaworthy Risks of Bulk Ship Accidents”. International Journal
of Transport Economics, Vol XXIX-No 1
Talley, WK (2002). “The safety of ferries: an accident injury perspective” Maritime Policy
and Management, Vol 29, No 3: 331-338
Thomas, R.L. (1997). Modern Econometrics – An introduction. Harlow: Addison Wesley
White, H. “Maximum Likelihood Estimation of Misspecified Models.” Econometrica, 53,
pp:1-16
Conference Attendances
Mare Forum: Shipping in a Responsible Society, Quo Vadis? 12th and 13th Sept. 2005,
Rome, Italy
Royal Institution of Naval Architects: Learning from Marine Incidents III, 25th and 26th
January 2006, London, UK
Connecticut Maritime Association: Shipping 2006: 20th to 22nd March 2006, Stamford,
Connecticut, USA
IMO Legislative Resources including IMO Proceedings (as Observer)
IMO Proceedings (Attendance as Observer):
Sub-Committee Meeting on Flag State Implementation – FSI (13), 7th to 11th March 2005,
IMO, London
193
Committee Meeting on Maritime Safety – MSC (80), 17th to 20th May 2005, IMO, London
General Assembly – 24th Session, 21st November to 2nd December 2005, IMO, London
Sub-Committee on Standards of Training and Watchkeeping – STCW (37), 23rd January
to 27th January 2006, IMO, London
Committee Meeting on Maritime Safety – MSC (81), 9th to 15th May 2006, IMO, London
Sub-Committee Meeting on Flag State Implementation – FSI (14), 6th to 9th June 2006,
IMO, London
Conventions:
International Convention on Load Lines (LL), 1966 and Protocol 1988, adopted 5th April
1966, IMO, London
International Convention on Civil Liability for Oil Pollution Damage (CLC),1969 and
Protocols 1976/1992, adopted 1969, IMO, London
International Convention of 1996 on Liability and Compensation for Damage in
Connection with the Carriage of Hazardous and Noxious Substances by Sea (not in force
yet), IMO, London
International Convention on Civil Liability for Bunker Oil Pollution Damage, 2001
(Bunker Oil Convention), adopted March 2001, IMO, London
International Maritime Dangerous Goods Code, IMO Publication, London, 2004
International Convention for the Prevention of Pollution from Ships with Annexes I to VI,
1973/1978 (MARPOL), Consolidated Edition, IMO Publication, London, 2001
International Convention for the Safety of Life at Sea (SOLAS), 1974 and Protocols 1978
and 1988, Consolidated Edition, IMO Publication, London, 2004
International Convention on Standards of Training, Certification and Watchkeeping for
Seafarers (STCW), 1978, IMO Publication, London
The Torremolinos International Convention for the Safety of Fishing Vessels, adopted
1977 and superseded by the 1993 Protocol, adopted April 1993 (not yet in force)
Codes and Resolutions:
Code of safe practice for solid bulk cargoes (BC Code) and Code for the construction and
equipment of ships carrying dangerous chemicals in bulk (IBC Code) adopted December
85 by resolution MEPC 19(22), IMO, London
Code of practice for the safe loading and unloading of bulk carriers, adopted November
1997 by Assembly Resolution A862 (20), IMO, London
Code of safe practice for cargo stowage and securing, adopted November 1991 by
Assembly Resolution A 714(17), IMO, London
194
Fire Safety Systems Code, adopted December 2000 by MSC 98(73), IMO, London
International Code of safety for high-speed craft, adopted May 1994 at MSC 36 (63), IMO,
London
International Code for the construction and Equipment of Ships Carrying Liquefied Gas
in Bulk and Old Gas Carrier Code for ships constructed before 1st October 1994 as per
resolution MSC 5 (48), IMO, London
International Code for the Construction and Equipment of ships carrying dangerous
chemicals in bulk, adopted 1993, IMO, London
International Code for security of ships and of port facilities, adopted 2002, IMO, London
International Life Saving Appliance Code, adopted June 1996 by MSC 48(66), IMO,
London
International Safety Management Code and Guidelines on Implementation of the ISM
Code, IMO Publication, London, 2002
IMO Resolution A.682 (17): Regional Co-Operation in the Control of Ships and
Discharges, adopted November 1991, IMO, London
IMO Resolution A.746 (18): Survey Guidelines under the Harmonized System of Survey &
Certification, IMO, London
IMO Resolution A.973 (24): Code for the Implementation of Mandatory IMO Instruments,
IMO, London
IMO Resolution A.974 (24): Framework and Procedures for the Voluntary IMO Member
State Audit Scheme, IMO, London
IMO Resolution A.739 (18): Guidelines for the Authorization of Organizations acting on
behalf of the Administration, IMO, London
IMO Resolution A.787(19) and A.822(21), Procedures for Port State Control, IMO
Publication, London, 2001
Circulars, Committee Working Papers and any Other Documents:
FSI/14/WP.3, Harmonization of Port State Control Activities, PSC on Seafarer’s Working
Hours, Development of Guidelines on Port State Control Under the 2004 BWM
Convention, Report of the Working Group, Flag State Implementation Sub-Committee
Meeting, IMO, London, 8th June 2006
FSI/14/WP.5 Add.1, Draft Report to the Maritime Safety Committee and the Marine
Environment Protection Committee, Sub-Committee on Flag State Implementation,
IMO, London, 9th June 2006
MSC/Circ. 1023, MEPC/Circ. 392, Guidelines for Formal Safety Assessment (FSA) for use
in the IMO Rule-Making Process, IMO, London, 4th April 2002
195
MSC81/18, Formal Safety Assessment, Report of the correspondence group, submitted by
the Netherlands, MSC 81/18, Maritime Safety Committee, IMO, London, 7th February
2006
MSC 81/17/1, The Role of the Human Element, Assessment of the impact and
effectiveness of implementation of the ISM Code, IMO Secretariat, Maritime Safety
Committee Meeting 81, London, 21st December 2005
MSC/Circ. 953, MEPC/Circ. 372, Reports on Marine Casualties and Incidents, Revised
harmonized reporting procedures, 14th December 2000, IMO, London
MSC/Circ. 1014, Guidance on Fatigue Mitigation and Management, IMO, London, 12th
June 2001
National Legislative Resources including EU Law
Australian Navigation Act of 1912, The Power of Inspections of Surveyors and Detentions
of Ships not registered in Australia:
http://scaletext.law.gov.au/html/pasteact/1/516/top.htm
Bahamas Maritime Authority, Flag State Inspection Report, received from the Bahamas
Maritime Authority directly, London, March 2006
Canadian Marine Insurance Act of 1993, c.22 – Loss and Abandonment,
https://monkessays.com/write-my-essay/canlii.org/ca/sta/m-0.6/sec57.html
Code of Safety for Caribbean Cargo Ships (CCSS Code), Cargo Ships less than 500 gt,
adopted 1996, Barbados
https://monkessays.com/write-my-essay/uscg.mil/hq/g-m/pscweb/code_of_safety_for_caribbean_car.htm
Directive 106/2001/EC of the European Parliament and of the Council of 19th December
2001 amending Council Directive 95/21/EC concerning the enforcement, in respect of
shipping using Community ports and sailing in the waters under the jurisdiction of the
Member States, of international standards for ship safety, pollution prevention and
shipboard living and working conditions (port State control)
http://europa.eu.int/eur-lex/en/index.html)
Proposal for a Directive on Port State Control, COM (2005), 588 final version of 23rd
November 2005
http://europa.eu.int/comm/transport/maritime/safety/2005_package_3_en.htm
Proposal for a Directive on the civil liability and financial securities of shipowners, COM
(2005), 593 final version of 23rd November 2005
http://europa.eu.int/comm/transport/maritime/safety/2005_package_3_en.htm
Proposal for a Directive establishing the fundamental principles governing the
investigation of accidents in the maritime transport sector and amending Directive
1999/35/EC and 2002/59/EC, COM (2005). 590 final version of 23rd November 2005
http://europa.eu.int/comm/transport/maritime/safety/2005_package_3_en.htm
196
Council Framework Decision 2005/667/JHA of 12th July 2005 to strengthen the criminallaw
framework for the enforcement of the law against ship-source pollution.
http://europa.eu.int/comm/transport/maritime/safety/2005_package_3_en.htm
Directive 2005/35/EC of 7th September 2005 on ship-source pollution and on the
introduction of penalties for infringements.
http://europa.eu.int/comm/transport/maritime/safety/2005_package_3_en.htm
Final Regulatory Impact Assessment – Draft Merchant Shipping (Port State Control
Amendment) Regulations 2003, MCA
International Labor Organization, Maritime Labor Convention 2006,
https://monkessays.com/write-my-essay/ilo.org/public/english/dialogue/sector/papers/maritime/consolcd/overview.htm
Malta Maritime Authority, Flag State Inspection Report, received from Inspector directly
Memorandum of Understanding on Port State Control in the Caribbean Region, received
from IMO Regional Maritime Adviser (Caribbean) at IMO, London.
Memorandum of Understanding on Port State Control for the Indian Ocean Region as of
1st Oct. 2000, https://monkessays.com/write-my-essay/indianmou.org
Memorandum of Understanding on Port State Control in the Asia-Pacific Region
containing 8th Amendment, 23rd Nov. 2004, https://monkessays.com/write-my-essay/tokyomou.org
Paris Memorandum of Understanding on Port State Control Including 27th Amendment,
adopted 13th May 2005, https://monkessays.com/write-my-essay/parismou.org
Paris Memorandum of Understanding on Port State Control, Annual Reports 2002, 2003
and 2004, https://monkessays.com/write-my-essay/parismou.org
Paris Memorandum of Understanding on Port State Control, Manual for PSC Officers,
Revision 8
USCG Marine Safety Manual, Vol. II, Section D: Port State Control
https://monkessays.com/write-my-essay/uscg.mil/hq/g-m/pscweb/Publication.htm
USCG Port State Control Speech
https://monkessays.com/write-my-essay/uscg.mil/hq/g-m/pscweb/psc_speech.pdf
United Nations Conventions on the Law of the Sea:
https://monkessays.com/write-my-essay/un.org/Depts/los/convention_agreements/texts/unclos/closindx.htm
Wikipedia, Legal Definitions:
http://en.wikipedia.org/wiki/List_of_legal_terms
http://en.wikipedia.org/wiki/General_average
Other Accessible Resources
BIMCO/ISF, Manpower 2005 Update, The worldwide demand for and supply of seafarers,
Institute for Employment Research, Coventry, 2005
197
Chemical Distribution Institute (CDI), Ship Inspection Report, Chemical Tanker, 5th
Edition, 2003, London
Fearnley’s Review 2004, Fearnresearch, Oslo, 2005
Fearnley’s World Bulk Trades 2003, Fearnresearch, Oslo, 2003
IACS, Class – What, Why and How, www.iacs.org.uk/_pdf/Class_WhatWhy&How.PDF
International Transport Worker’s Federation, Seafarer fatigue: Wake up to the dangers,
ITF, London
ISL Shipping Statistics and Market Review (SSMR), Volume 49 (2005), Institute of
Shipping Economics and Logistics, Bremen
ISM Code Training Manual for DNV Auditors, DNV ISM Code Auditor Course, Part I-V
Lloyd’s Maritime Atlas of World Ports and Shipping Places, Lloyd’s Marine Intelligence
Unit, T&F Informa, UK, 2004
Marine Accident Investigation Branch, Bridge Watchkeeping Safety Study, July 2004,
Southampton
Main Characteristics of CAP and CAS Compared, DNV Presentation, 2005
OCIMF (Oil Companies International Marine Forum), Vessel Inspection Questionnaire for
Bulk Oil Tankers, Combination Carriers and Shuttle Tankers, 3rd Edition, May 2005
OCIMF (Oil Companies International Marine Forum), Tanker Management and Self
Assessment, A Best-Practice Guide for Ship Operators, First Edition 2004, London
Reyner, L. and Baulk S, (1998), Fatigue in Ferry Crews: A Pilot Study, Seafarers
International Research Center, Cardiff, 1998
Rules and Regulations for the Classification of Ships, Part 1, Edition July 2003, Lloyd’s
Register, London 2003
Rules for Classification and Construction/Ship Technology, Edition 2005, Germanischer
Lloyd, Hamburg, 2005
Selection & Accreditation of Rightship Tanker Inspectors, received from Rightship
directly, Australia
SeaCure for Operations 2004, 9th Edition, Rev. O, Greenaward Foundation, Rotterdam,
The Netherlands
Ship Inspection System, Final Inspection Report for Dry Cargo Ships, received from
Rightship directly, Australia
United Nations Conference on Trade and Development, Review of Maritime Transport,
2004, https://monkessays.com/write-my-essay/unctad.org/Templates/StartPage.asp?intItemID=2614&lang=1
198
Interviews
Bergot, G., and Barbeira-Gordon, S. (2004). Interview by author, European Commission,
Directorate-General for Energy and Transport, Brussels, June 2004
Bergot, G. and Gonzalez-Gil, J. (2005), Interview by author, European Commission,
Directorate-General for Energy and Transport, Brussels, October 2005
Castex, B. M. (2005 & 2006), Interview by Author, IMO Secretariat, during proceedings of
General Assembly, STCW and MSC, IMO, London, 22nd -23rd November 2005, 23rd
January 2006 and 14th May 2006
Davidson, C. and Rimington D. (2005), Interview by Author, Australian Maritime Safety
Authority, during proceedings of General Assembly, IMO, London, 22nd -23rd November
2005
De Graeve, W. (2004), Interview by Author, Federal Public Service Mobility and
Transport, Maritime Transport, Maritime Inspectorate, Antwerp, July 2004
Dudley, J. Capt. (2005), Interview by Author, Koch Supply & Trading Ltd, Rotterdam,
October 2005
Downs Tim J. Capt. and George, D. Capt.,(2005) Interview by Author, Shell Trading and
Shipping Company Ltd, London, November 2005
Fransen, J. and Capt. Den Heijer, R. (2004). Interview by Author, Green Award
Foundation, Rotterdam
Gardiner, C.R. D. (2005), Interview by Author, Office of the Permanent Representative to
IMO (Antigua & Barbuda), during proceedings of MSC, IMO, London, 11th -20th May 2005
Groves, B. (2006), Interview by Author, Australian Maritime Safety Authority, during
proceedings of STCW, IMO, London, 23rd – 27th January 2006
Harts, P. (2004). Interview by Author, Inspectorate Transport and Water Management,
Netherlands Shipping Inspectorate, Rotterdam
Hassing, S. (2006), Interview by Author, Dutch Directorate for Transport Safety, during
proceedings of STCW, IMO, London, 23rd – 27th January 2006
Huisink, G.J. (2005), Interview by Author, Royal Association of Netherlands’ Shipowners,
Rotterdam, October 2005
Hutchinson, D. Capt. (2005), Interview by Author, Bahamas Maritime Authority, during
proceedings of General Assembly, IMO, London, 22nd -23rd November 2005
Jansen, P. Capt. (2005), Interview by Author, Ministry of Transport and Infrastructure,
Antwerp, November 2005
Kamstra, P.C. (2004), Interview by Author, Inspectorate Transport and Water
Management, Netherlands Shipping Inspectorate, Rotterdam.
199
Kinley, M. and Evans, B. (2005), Interview by Author, Australian Maritime Safety
Authority, during proceedings of FSI , IMO, London, 7th -11th March 2005
Koorneef, C.W. (2004). Interview by Author, Department of Noxious and Dangerous
Goods, Port of Rotterdam, Rotterdam
Koert, C. (2004), Interview by Author, Department of Noxious and Dangerous Goods, Port
of Rotterdam, Rotterdam
Mansell, J. (2006), Interview by Author, New Zealand Maritime Authority, during
proceedings of FSI, IMO, London, 6th June 2006
Monzon, A.M. (2005), Interview by Author, Prefectura Naval Argentina, during
proceedings of FSI, IMO, London, 7th -11th March 2005
Morton, L. (2006), Interview by Author, Exxon Mobile, June 2006, Rotterdam
Norman, W. Capt. (2005), Interview by Author, RightShip, Mare Forum Conference,
Rome, 12th -13th September 2005
Parr, P. and Dolby, P. (2006), Interview by Author, UK Shipping Policy Unit and MCA,
London, January 2006
Pas, D. (2005), Interview by Author, Directorate for Transport Safety (former Senior
Policy Advisor), Erasmus University, Rotterdam, November 2005
Salwegter, A. (2004), Interview by Author, Inspectorate Transport and Water
Management, Netherlands Shipping Inspectorate, Rotterdam
Sakurada, Y. (2005), Interview by Author, DNV Senior Surveyor, Rotterdam, October
2005
Scheres, G. (2004), Interview by Author, Inspectorate Transport and Water Management,
Netherlands Shipping Inspectorate, Rotterdam
Schiferli, R. (2005), Interview by Author, Paris MoU Secretariat, Den Hague, November
2005
Snow, G. (2006), Interview by Author, Oil Companies International Marine Forum,
London, May 2006
Thorne Paul Cdr. and E.J. Terminella Cdr. (2005), Interview by Author, USCG Foreign &
Offshore Compliance Division, during proceedings of FSI, IMO, London, 7th -11th March
2005
Turenhout, H., van der Veer G.J. and Kreuze, A. (2006), Interview by Author, Jo Tankers,
Rotterdam, June 2006
Whittle, M. A., Interview by Author, Chemical Distribution Institute, London, November
2005
200
Wright, C. (2006), Interview by Author, Permanent Secretary of IACS, London, January
2006
Zecchin, L. A. (2005), Interview by Author, Prefectura Naval Argentina, during
proceedings at General Assembly, IMO, London, 22nd -23rd November 2005
Ship Visits, Inspections, Surveys
The ship names and IMO numbers are not disclosed as per the request of some of the ship
owners/operators.
PSC Inspection: Flag: Luxembourg, Ship Type: Containership, Surveyor: Aarnout
Salwegter, Rotterdam, June 2004
PSC Inspection: Flag: Syria, Ship Type: General Cargo, Surveyor: Walter De Graeve,
Antwerp, July 2004
PSC Inspection: Flag: Cyprus, Ship Type: General Cargo, Surveyor: Walter De Graeve,
Antwerp, July 2004
PSC Expanded Inspection: Flag: Grand Caymans, Ship Type: Bulk Carrier, Surveyor:
Aarnout Salwegter, Amsterdam, August 2005
PSC inspection/Detention: Flag: Ukraine, Ship Type: General Cargo, Inspector: J. P. Van
Byten, Antwerp, October 2005.
PSC safety inspection: Flag: Hong Kong, Ship Type: Dry Bulk, Inspector in charge: Ralph
Savercool, New York, March 2006
PSC security inspection: Flag: Liberia, Ship Type: Container, Inspector in charge: Diane
R. Semmling, New York, March 2006
PSC security inspection: Flag: Panama, Ship Type: Container, Inspector in charge: Diane
R. Semmling, New York, March 2006
Flag State Inspection: Flag: Malta, Ship Type: Container, Surveyor: Henk Engelsman,
Rotterdam, August 2005
Flag State Inspection: Flag: Malta, Ship Types: Bulk Carrier, Surveyor: Henk
Engelsman, Rotterdam, October 2005,
Class Annual Survey and Underwater Diving Inspection: Flag: Norwegian International
Register, Ship Type: Oil/Bulk Carrier, Surveyor: Yuri Sakurada, DNV, Rotterdam, March
2005
Class Annual Survey: Flag: Norwegian International Register, Ship Type: Chemical
Tanker, Surveyor: Yuri Sakurada, DNV, Rotterdam, May 2005
Class Annual Survey: Flag: Malta, Ship Type: Crude Oil Tanker, Surveyor: Rob Pijper,
Lloyd’s Register, Rotterdam, November 2005.
201
Class Annual Survey: Flag: Barbados, Ship Type: General Cargo Ship, Surveyor: Pieter
Andringa, Germanischer Lloyd, Rotterdam, October 2005.
Class Renewal Survey: Ship Name: Flag: Dutch, Ship Type: Chemical/Oil Product
Tanker, Surveyor: Rob Pijper, Lloyd’s Register, Rotterdam Damen Shipyard, August 2005
Class Follow Up: Flag: Cyprus, Ship Type: Bulk Carrier, Surveyor: Rob Pijper, Lloyd’s
Register, Rotterdam, September 2005
ISM Audit: Flag: Liberia, Ship Type: Juice Carrier, Surveyor: Rob Pijper, Lloyd’s
Register, Rotterdam, October 2005
Vetting Inspection (CDI): Flag: Dutch, Ship Type: Chemical Tanker, Inspector (CDI):
Henk Engelsman, Rotterdam, August 2005
Vetting Inspection (CDI): Flag: Bahamas, Ship Type: Chemical/Oil Tanker, Inspector
(CDI): Henk Engelsman, Rotterdam, October 2005;
Vetting Inspection (SIRE, Kuwait Oil): Flag: Sweden, Ship Type: Oil Tanker, Inspector
(OCIMF): Henk Engelsman, Rotterdam, September 2005
Vetting Inspection (SIRE, Eni Oil): Flag: Saudi Arabia, Ship Type: Chemical Tanker,
Inspector (OCIMF): Henk Engelsman, Rotterdam, October 2005;
Vetting Inspection (SIRE, Statoil): Flag: Sweden, Ship Type: Tanker, Inspector (OCIMF):
Henk Engelsman, Rotterdam, June 2006
Vetting Inspection (SIRE, Statoil): Flag: Liberia, Ship Type: Oil Tanker, Inspector
(OCIMF): Henk Engelsman, Rotterdam, June 2006
Vetting Inspection (Rightship): Flag: Hong Kong, Ship Type: Dry Bulk Carrier, Inspector
(Rightship): Dennis Barber, Ijmuiden, March 2006
P&I Club Inspection: Flag: Greece, Ship Type: Bulk Carrier, Inspector: Walter
Vervloesem, Ghent, October 2005;
MARPOL Inspection: Flag: Norway, Ship Type: Oil Tanker, Port Superintendent: Mr.
Cees-Willem Koorneef, Rotterdam, August 2004
MARPOL Inspection: Flag: Panama, Ship Type: OBO, Port Superintendent: Mr. Cees-
Willem Koorneef, Rotterdam, August 2004
Ship Visit (VLCC): Flag: Bahamas, Ship Type: Oil Tanker, Class: ABS, Rotterdam,
October 2005
202
203
Biography
Despite the author’s origin from a non maritime nation, her life has been highly
influenced by the seven seas. She received her first Master of Science degree from Maine
Maritime Academy (USA) and served at sea for approximately 10 years of her life on
cruise ships as hotel officer including onboard S.Y. Sea Cloud, one of the last historic tall
ships still afloat and still in service. Sailing areas during her seagoing career on various
other ships included most countries with access to the sea except parts of Africa, the
North China Sea, the Bering Sea and Antarctica. Land based working experience
includes primarily countries in North, Central and South America as well as the
Caribbean where she worked in Tortola (British Virgin Islands) as port agent.
Her last land-based position was the establishment of the fleet auditor position for
Carlson Companies, USA where she was responsible for six vessels. Carlson Companies is
one of the largest privately owned companies in the USA and owns Regency Cruises
(former Radisson Seven Seas Cruises). The fleet auditor position entailed constant travel
across the globe but also permitted to be part of the management team of two new
buildings and experiencing the maiden voyage of one of the vessels, the Seven Seas
Mariner. Working in the cruise industry has trained her to function in a five star
operation including working in a multicultural and multi-corporate environment where
many nationalities have to cooperate to make the complex operation of a cruise vessel
possible. Besides German and English, she is fluent in French and Spanish.
In autumn 2002, she recalled her European roots and returned back to the European
Union and to university. She first studied Advanced European Studies with an emphasis
on European Union law at Danube University, Austria and then moved to Rotterdam.
She has been working on her PhD as a guest at the Econometric Institute of Erasmus
University in the area of maritime safety from November 2004 onwards. She will
continue her career in the commercial shipping industry where her particular interest
lies within the regulatory perspective of shipping including further applications of
Econometrics to other segments of this industry and the increase of transparency and
improvement of working and living conditions of seafarers.
204
205
Appendices
Appendix 1: List of Member States of each MoU
Paris MoU Caribbean MoU Viña del Mar Indian MoU
AMSA
(Tokyo MoU)
Belgium Anguilla Argentina Australia Australia
Canada (1994) Antigua & Barbuda Bolivia (1999) Bangladesh Canada
Croatia (1997) Aruba Brazil Djibouti Chile (2002)
Denmark Bahamas Chile Eritrea China
Estonia (2005) Barbados Colombia Ethiopia (Obs.) Fiji (1996)
Finland Bermuda Cuba (1995) India Hong Kong
France British Virgin Islands Ecuador Iran Indonesia (1996)
Germany Cayman Islands Honduras (2001) Kenya Japan
Greece Dominica Mexico Maldives Republic of Korea
Iceland (2000) Grenada Panama Mauritius Malaysia
Ireland Guyana Peru Mozambique New Zealand
Italy Jamaica Uruguay Myanmar Papua New Guinea
Latvia (2005) Montserrat Venezuela Oman Philippines (1997)
Netherlands Netherlands Antilles Seychelles
Russian Fed.
(1995)
Norway St. Kitts & Nevis South Africa Singapore
Poland (1992) St. Lucia Sri Lanka Solomon Islands *)
Portugal St. Vinc.& Grenadines Sudan Thailand (1996)
Russian Fed. (1996) Suriname Tanzania Vanuatu
Slovenia (2003) Trinidad & Tobago Yemen Viet Nam (1999)
Spain Turks & Caicos Islands
Sweden Cuba *) not yet accepted
UK Dominican Republic
206
Appendix 2: List of Detainable Deficiencies126
The following is a list of detainable deficiencies as per the IMO PSC Guidelines and are
split up into the relevant legal bases:
SOLAS (all ships)
1. improper operation of propulsion or essential machinery
2. insufficient cleanliness in engines room, excessive dirty bilges, insulation of piping
and contamination of oil, improper operation of bilge pumping arrangements
3. failure of proper operation of emergency generator, lighting, batteries and switches
4. failure of proper operation of main and auxiliary steering gear
5. absence or insufficient life savings appliances, survival craft and launching
arrangements or serious deterioration thereof
6. non-functional fire fighting detection or fighting system or equipment including fire
dampers and ventilation valves
7. absence or serious deterioration of fire fighting equipment of the cargo deck area for
tankers.
8. absence or serious deterioration of lights, shapes or sound signals
9. absence or serious deterioration of radio equipment for distress and safety
communication
10. absence of serious deterioration of navigational equipment
11. absence of corrected navigational charts or publications
12. absence of non-sparking exhaust ventilation for cargo pump rooms
13. number of crew does not match the safe manning certificate
14. non-implementation of the enhanced survey program when applicable
IBC and IGC Code (ships carrying dangerous cargo and gas carriers)
1. ship is carrying cargo not mentioned in the certificate of fitness or missing cargo
information
2. missing or damaged high pressure safety devices
3. electrical installation not in compliance with IBC Code
4. sources of ignition in hazardous locations
5. exceeding of maximum allowable cargo quantity per tank
6. insufficient heat protection for sensitive products
7. missing closing devices for accommodation or service spaces
8. bulkhead not gastight
9. defective air locks
10. missing of defective quick-closing valves or safety valves
11. ventilation in cargo area not operable
12. pressure alarms for cargo tanks not operable
13. gas detection plant and/or toxic gas detection plant defective
Load Line Convention
1. significant damage or corrosion effecting seaworthiness of vessel
2. insufficient stability
3. absence of sufficient and reliable information for loading and ballasting of the vessel
to maintain stability
4. absence or substantial deterioration of closing devices, hatch covers and watertight
and weather tight doors
5. overloading
6. absence or impossible to read draught marks or load line marks
126 as per IMO guidelines on PSC, Chapter 2.3 and Appendix I (detainable deficiencies)
207
MARPOL Annex I, Annex II
1. malfunction of oily water separator
2. remaining capacity of slop/sludge tanks insufficient for the intended voyage
3. Oil Record Book not available
4. unauthorized discharge bypass fitted
5. failure to meet requirements of 13G(4) or 13G(7) – Crude Oil Washing
6. Absence of P&A Manual
7. Cargo is not categorized
8. No Cargo Record Book
9. Transport of oil-like substances without satisfying the requirement
STCW Convention
1. invalid certificate of competence or no endorsement by the flag state
2. failure to comply to safe manning requirements by the flag state
3. failure of navigational or engineering watch arrangements to conform to flag state
requirements
4. absences in a watch of a qualified person to operate equipment for safe navigation,
safety radio communications or the prevention of marine pollution
5. non compliance to sufficient rest periods
208
Appendix 3: IMO Definitions of Selected Major Ship Types127
Passenger Ship: ship that carries more than twelve passengers.
Cargo Ship: any ship which is not a passenger ship.
Tanker: cargo ship constructed or adapted for the carriage in bulk of liquid cargoes of an
inflammable nature.
Oil Tanker: a ship constructed or adapted primarily to carry oil in bulk in its cargo spaces
and includes the combination carriers and any “chemical tanker” when it is carrying a
cargo or part cargo of oil in bulk. For the purpose of the Condition Assessment Scheme, oil
tankers are divided into three categories as follows:
Category 1: oil tanker of 20,000 tons deadweight and above carrying crude oil, fuel
oil, heavy diesel oil or lubricating oil as cargo, and of 30,000 tons deadweight and
above carrying oil other than the above, which does not comply with the
requirements for new oil tankers as defined in regulation 1(26) of MARPOL Annex
I.128
Category 2: oil tanker of 20,000 tons deadweight and above carrying crude oil, fuel
oil, heavy diesel oil or lubricating oil as cargo, and of 30,000 tons deadweight and
above carrying oil other than the above, which complies with the requirements for
new oil tankers as defined in regulation 1(26) of MARPOL Annex I.
Category 3: oil tanker of 5,000 tons deadweight and above but less than that
specified for category 1 and 2.
Combination Carrier: a ship designed to carry either oil or solid cargoes in bulk.
Chemical Tanker: is a cargo ship constructed or adapted and used for the carriage in bulk
of any liquid product listed in either:
• Chapter 17 of the International Code for the Construction and Equipment of Ships
Carrying Dangerous Chemicals in Bulk (IBC Code)
• Chapter VI of the Code for the Construction and Equipment of Ships Carrying
Dangerous Chemicals in Bulk adopted by resolution A.212(VII).
Gas Carrier: a cargo ship constructed or adapted and used for the carriage in bulk of any
liquefied gas or other products listed in either:
• Chapter 19 of the International Code for the Construction and Equipment of Ships
Carrying Liquefied Gases in Bulk (IGC Code)
• Chapter XIX of the Code for Construction and Equipment of Ships Carrying
Liquefied Gases in Bulk adopted by resolution A.328 (IX).
Bulk Carrier: a ship which is constructed generally with single deck, top-side tanks and
hopper side tanks in cargo spaces, and is intended primarily to carry dry cargo in bulk,
and includes such types as ore carriers and combination carriers
Fishing Vessel: a vessel used for catching fish, whales, seals, walrus or other living
resources of the sea.
Ro-Ro Passenger: a passenger ship with ro-ro cargo spaces or special category spaces as
defined in regulation II-2/3. (SOLAS).
127 All sources of definitions are taken from SOLAS and MARPOL
128 Refers to the date the vessel was built
209
Appendix 4: Variable List and Respective Coding for Regressions
The following list of variables shows a list of variables which have been used either in the
probability of detention or probability of casualty models. Some variables are the same in
both models while some variables only appear in one model. Due to the large size of
variables in the detention models, only one ship type is shown below as an example. The
detention models also use the following coding:
*) Variable+ Ship Type (e.g. GC) + PSC Regime (1 for Paris MoU)
Type of Models – Codes Used
Detention*) Casualty Explanation of Variable
IMO IMO IMO number of vessel
n/a Casualty ship had casualty (split into seriousness)
n/a VerySerious very serious casualty
n/a Serious serious casualty
n/a LessSerious less serious casualty
detained detained ship was detained
AgebyGC1 lnAge Age at time of inspection/casualty
SizeGC1 lnTon Gross Tonnage
n/a ST1_max general cargo ship
n/a ST2_max dry bulk carrier
n/a ST3_max container
n/a ST4_max tanker
n/a ST5_max passenger
n/a ST6_max other ship type
n/a ST7_max fishing vessel
CG n/a general cargo ship (middle portion of code)
DB n/a dry bulk carrier (middle portion of code)
CO n/a container (middle portion of code)
TA n/a tanker (middle portion of code)
PA n/a passenger (middle portion of code)
OT n/a other ship type (middle portion of code)
n/a RS_Ins ship inspected by Rightship
n/a RS_1S Rightship 1 star vessel
n/a RS_2S Rightship 2 star vessel
n/a RS_3S Rightship 3 star vessel
n/a RS_4S Rightship 4 star vessel
n/a RS_5S Rightship 5 star vessel
n/a Green ship is Greenaward certified
n/a DH double hull
n/a STChgd ship type changed
n/a FLChgd flag changed
n/a CLChgd classification society changed
n/a CLWd_d classification society withdrawn
n/a OWChgd ownership changed
1 n/a Paris MoU (end portion of code)
2 n/a Caribbean MoU (end portion of code)
3 n/a Viña del Mar (end portion of code)
4 n/a Indian Ocean MoU (end portion of code)
5 n/a USCG (end portion of code)
6 n/a AMSA (end portion of code)
n/a PMOU_av average inspection fraction Paris MoU
210
Type of Models – Codes Used
Detention*) Casualty Explanation of Variable
n/a CMOU_av average inspection fraction Caribbean MoU
n/a VMOU_av average inspection fraction Viña del Mar MoU
n/a IMOU_av average inspection fraction Indian Ocean MoU
n/a USCG_av average inspection fraction USCG
n/a AMSA_av average inspection fraction AMSA
n/a PMOU_s number inspection fraction Paris MoU
n/a CMOU_s number inspection fraction Caribbean MoU
n/a IMOU_s number inspection fraction Indian Ocean MoU
n/a VMOU_s number inspection fraction Viña del Mar MoU
n/a USCG_s number inspection fraction USCG
n/a AMSA_s number inspection fraction AMSA
n/a det_PMOU detained by Paris MoU
n/a det_CMOU detained by Caribbean MoU
n/a det_VMOU detained by Viña del Mar MoU
n/a det_IMOU detained by Indian Ocean MoU
n/a det_USCG detained by USCG
n/a det_AMSA detained by AMSA
n/a lnTimebw time inbetween inspections
OWEMNGC1 OW_EMN Owner from Emerging Maritime Nation
OWTMNGC1 OW_TMN Owner from Traditional Maritime Nation
OWOORGC1 OW_OOR Owner from Old Open Registry
OWNORGC1 OW_NOR Owner from New Open Registry
OWIORGC1 OW_IOR Owner from Intern. Open Registry
OWUNKGC1 OW_UNKn Owner Unknown
n/a LI_FLRet Number of Legal Instruments Ratified by Flag
n/a LI_OWRet Number of Legal Instr. Ratified by Owner Country
FL_AFGC1 FL_AF FL_Afghanistan
FL_ALGC1 FL_AL FL_Albania
FL_DZGC1 FL_DZ FL_Algeria
FL_AGGC1 FL_AG FL_Antigua
FL_ANGC1 FL_AN FL_AntillesNetherland
FL_ARGC1 FL_AR FL_Argentina
FL_AUGC1 FL_AU FL_Australia
FL_ATGC1 FL_AT FL_Austria
FL_AZGC1 FL_AZ FL_Azerbaijan
FL_BSGC1 FL_BS FL_Bahamas
FL_DHGC1 FL_DH FL_Bahrain
FL_BDGC1 FL_BD FL_Bangladesh
FL_BBGC1 FL_BB FL_Barbados
FL_BEGC1 FL_BE FL_Belgium
FL_BZGC1 FL_BZ FL_Belize
FL_BMGC1 FL_BM FL_Bermuda
FL_BOGC1 FL_BO FL_Bolivia
FL_BRGC1 FL_BR FL_Brazil
FL_BGGC1 FL_BG FL_Bulgaria
FL_BVIGC1 FL_BVI FL_BVI
FL_KHGC1 FL_KH FL_Cambodia
FL_CAGC1 FL_CA FL_Canada
FL_KYGC1 FL_KY FL_CaymanIslands
FL_CLGC1 FL_CL FL_Chile
FL_CNGC1 FL_CN FL_China
FL_COGC1 FL_CO FL_Colombia
211
Type of Models – Codes Used
Detention*) Casualty Explanation of Variable
FL_KMGC1 FL_KM FL_Comoros
FL_HRGC1 FL_HR FL_Croatia
FL_CUGC1 FL_CU FL_Cuba
FL_CYGC1 FL_CY FL_Cyprus
FL_DKGC1 FL_DK FL_Denmark
FL_DMGC1 FL_DM FL_Dominica
FL_DOGC1 FL_DO FL_DominicanRepublic
FL_ECGC1 FL_EC FL_Ecuador
FL_EGGC1 FL_EG FL_Egypt
FL_ERGC1 FL_ER FL_Eritrea
FL_EEGC1 FL_EE FL_Estonia
FL_ETGC1 FL_ET FL_Ethiopia
FL_FOGC1 FL_FO FL_FaroeIslands
FL_FJGC1 FL_FJ FL_Fiji
FL_FIGC1 FL_FI FL_Finland
FL_FRGC1 FL_FR FL_France
FL_GEGC1 FL_GE FL_Georgia
FL_DEGC1 FL_DE FL_Germany
FL_GIGC1 FL_GI FL_Gibraltar
FL_GRGC1 FL_GR FL_Greece
FL_GYGC1 FL_GY FL_Guyana
FL_HTGC1 FL_HT FL_Haiti
FL_HNGC1 FL_HN FL_Honduras
FL_HKGC1 FL_HK FL_HongKong
FL_ISGC1 FL_IS FL_Iceland
FL_INGC1 FL_IN FL_India
FL_IDGC1 FL_ID FL_Indonesia
FL_IRGC1 FL_IR FL_Iran
FL_IQGC1 FL_IQ FL_Iraq
FL_IEGC1 FL_IE FL_Ireland
FL_IMGC1 FL_IM FL_IselofMan
FL_ILGC1 FL_IL FL_Israel
FL_ITGC1 FL_IT FL_Italy
FL_JMGC1 FL_JM FL_Jamaica
FL_JPGC1 FL_JP FL_Japan
FL_JOGC1 FL_JO FL_Jordan
FL_KIGC1 FL_KI FL_Kiribati
FL_KWGC1 FL_KW FL_Kuwait
FL_LVGC1 FL_LV FL_Latvia
FL_LBGC1 FL_LB FL_Lebanon
FL_LRGC1 FL_LR FL_Liberia
FL_LYGC1 FL_LY FL_Libya
FL_LTGC1 FL_LT FL_Lithuania
FL_LUGC1 FL_LU FL_Luxembourg
FL_MYGC1 FL_MY FL_Malaysia
FL_MVGC1 FL_MV FL_Maldives
FL_MTGC1 FL_MT FL_Malta
FL_MHGC1 FL_MH FL_MarshallIslands
FL_MUGC1 FL_MU FL_Mauritius
FL_MXGC1 FL_MX FL_Mexico
FL_MDGC1 FL_MD FL_Moldovia
FL_MNGC1 FL_MN FL_Mongolia
212
Type of Models – Codes Used
Detention*) Casualty Explanation of Variable
FL_MAGC1 FL_MA FL_Morocco
FL_MMGC1 FL_MM FL_Myanmar
FL_NAGC1 FL_NA FL_Namibia
FL_NLGC1 FL_NL FL_Netherlands
FL_NZGC1 FL_NZ FL_NewZealand
FL_NIGC1 FL_NI FL_Nicaragua
FL_NGGC1 FL_NG FL_Nigeria
FL_NISGC1 FL_NIS FL_NIS
FL_KPGC1 FL_KP FL_NorthKorea
FL_NOGC1 FL_NO FL_Norway
FL_OTGC1 FL_OT FL_Other
FL_PKGC1 FL_PK FL_Pakistan
FL_PAGC1 FL_PA FL_Panama
FL_PYGC1 FL_PY FL_Paraguay
FL_PEGC1 FL_PE FL_Peru
FL_PHGC1 FL_PH FL_Philippines
FL_PLGC1 FL_PL FL_Poland
FL_PTGC1 FL_PT FL_Portugal
FL_QAGC1 FL_QA FL_Quatar
FL_ROGC1 FL_RO FL_Romania
FL_RUGC1 FL_RU FL_RussianFed
FL_ASGC1 FL_AS FL_Samoa
FL_STGC1 FL_ST FL_SaoTomePrin
FL_SAGC1 FL_SA FL_SaudiaArabia
FL_SCGC1 FL_SC FL_Seychelles
FL_SGGC1 FL_SG FL_Singapore
FL_SKGC1 FL_SK FL_Slovakia
FL_ZAGC1 FL_ZA FL_SouthAfrica
FL_KRGC1 FL_KR FL_SouthKorea
FL_ESGC1 FL_ES FL_Spain
FL_LKGC1 FL_LK FL_SriLanka
FL_VCGC1 FL_VC FL_StVincentGrenad
FL_SDGC1 FL_SD FL_Sudan
FL_SEGC1 FL_SE FL_Sweden
FL_CHGC1 FL_CH FL_Switzerland
FL_SYGC1 FL_SY FL_Syria
FL_TWGC1 FL_TW FL_Taiwan
FL_THGC1 FL_TH FL_Thailand
FL_TOGC1 FL_TO FL_Tonga
FL_TTGC1 FL_TT FL_TrinidadTobago
FL_TNGC1 FL_TN FL_Tunisia
FL_TRGC1 FL_TR FL_Turkey
FL_TMGC1 FL_TM FL_Turkmenistan
FL_TVGC1 FL_TV FL_Tuvalu
FL_AEGC1 FL_AE FL_UAE
FL_UKGC1 FL_UK FL_UK
FL_UAGC1 FL_UA FL_Ukraine
FL_UNGC1 FL_UN FL_Unknown
FL_USGC1 FL_US FL_USA
FL_VUGC1 FL_VU FL_Vanuatu
FL_VEGC1 FL_VE FL_Venezuela
FL_VNGC1 FL_VN FL_VietNam
213
Type of Models – Codes Used
Detention*) Casualty Explanation of Variable
CLABSGC1 CL_ABS CL_ABS
CLBKGC1 CL_BKI CL_BiroKlasIndo
CLBUKGC1 CL_BUK CL_BulgarskiKoraben
CLBVGC1 CL_BV CL_BureauVeritas
CLCCGC1 CL_CCS CL_ChinaClass
CLCOOGC1 CL_CCO CL_ChinaCorp
CLCRRGC1 CL_CRR CL_CroatianRS
CLDNVGC1 CL_DNV CL_DNV
CLGLGC1 CL_GL CL_GermanischerLloyd
CLGBKGC1 CL_GBS CL_GuardianBS
CLHELGC1 CL_HEL CL_Hellenic
CLHINGC1 CL_HIN CL_HondurasInterNav
CLINCGC1 CL_INC CL_Inclamar
CLINRGC1 CL_INR CL_IndianRegister
CLINSGC1 CL_INS CL_InterNavSurB
CLIRSGC1 CL_IRS CL_InterRegShipping
CLIBSGC1 CL_IBS CL_IsthmusBS
CLJRSGC1 CL_JRS CL_JosonRS
CLKRSGC1 CL_KRS CL_KoreanSouth
CLLRGC1 CL_LR CL_LloydsUK
CLNKKGC1 CL_NKK CL_NKKJapan
CLNCLGC1 CL_NCL CL_NoClass
CLOCLGC1 CL_OCL CL_OtherClass
CLPBSGC1 CL_PBS CL_PanamaBureauS
CLPMDGC1 CL_PMD CL_PanamaMDS
CLPMSCG1 CL_PMS CL_PanamaMSurveyorB
CLPRCGC1 CL_PRC CL_PanamaRegCorp
CLPSRCG1 CL_PSR CL_PanamaShipReg
CLPRSGC1 CL_PRS CL_PolskiReSt
CLRSAGC1 CL_RSA CL_RegisterAlbania
CLRCCGC1 CL_RCC CL_RegistroCubano
CLRSCGC1 CL_RSC CL_RegShipChina
CLRSKGC4 CL_RSK CL_RegShipDRKorea
CLRSGGC1 CL_RSG CL_RegShipGhana
CLRINGC1 CL_RIN CL_RINA
CLRIPGC1 CL_RIP CL_RINAVE
CLRNRGC1 CL_RNR CL_RomanianNaval
CLRMSGC1 CL_RMS CL_RussianMS
CLRRRGC1 CL_RRR CL_RussianRiver
CLSRUGC1 CL_SRU CL_SRUkraine
CLTLLGC1 CL_TLL CL_TurkishLloyd
CLVRSGC1 CL_VRS CL_VietnamRS
C0100GC1 Code_0100 Ship’s certificates and documents
C0200GC1 Code_0200 Crew certificates
C0300GC1 Code_0300 Accommodation
C0400GC1 Code_0400 Food and catering
C0500GC1 Code_0500 Working spaces and accident prev.
C0600GC1 Code_0600 Life saving appliances
C0700GC1 Code_0700 Fire Safety measures
C0800GC1 Code_0800 Accident prevention (ILO147)
C0900GC1 Code_0900 Structural Safety
C1000GC1 Code_1000 Alarm signals
214
Type of Models – Codes Used
Detention*) Casualty Explanation of Variable
C1100GC1 Code_1100 Cargoes
C1200GC1 Code_1200 Load lines
C1300GC1 Code_1300 Mooring arrangements (ILO 147)
C1400GC1 Code_1400 Propulsion & auxiliary engine
C1500GC1 Code_1500 Safety of navigation
C1600GC1 Code_1600 Radio communications
C1700GC1 Code_1700 MARPOL Annex I
C1800GC1 Code_1800 Gas and chemical carriers
C1900GC1 Code_1900 MARPOL Annex II
C2000GC1 Code_2000 SOLAS Operational deficiencies
C2100GC1 Code_2100 MARPOL related oper. deficiencies
C2200GC1 Code_2200 MARPOL Annex III
C2300GC1 Code_2300 MARPOL Annex V
C2500GC1 Code_2500 ISM related deficiencies
C2600GC1 Code_2600 Bulk carriers
C2700GC1 Code_2700 Security
C2900GC1 Code_2900 MARPOL Annex IV
C9800GC1 Code_9800 Other def. clearly hazardous safety
C9900GC1 Code_9900 Other def. not clearly hazardous
PS1_BE PS1_BE PS1_Belgium
PS1_CA PS1_CA PS1_Canada
PS1_HR PS1_HR PS1_Croatia
PS1_DK PS1_DK PS1_Denmark
PS1_FI PS1_FI PS1_Finland
PS1_FR PS1_FR PS1_France
PS1_DE PS1_DE PS1_Germany
PS1_GR PS1_GR PS1_Greece
PS1_IS PS1_IS PS1_Iceland
PS1_IE PS1_IE PS1_Ireland
PS1_IT PS1_IT PS1_Italy
PS1_NL PS1_NL PS1_Netherlands
PS1_NO PS1_NO PS1_Norway
PS1_PL PS1_PL PS1_Poland
PS1_PT PS1_PT PS1_Portugal
PS1_RU PS1_RU PS1_Russia
PS1_SI PS1_SI PS1_Slovenia
PS1_ES PS1_ES PS1_Spain
PS1_SE PS1_SE PS1_Sweden
PS1_UK PS1_UK PS1_UK
PS2_AG PS2_AG PS2_Antigua
PS2_AN PS2_AN PS2_AntillesNetherlands
PS2_BS PS2_BS PS2_Bahamas
PS2_BB PS2_BB PS2_Barbados
PS2_BVI PS2_BVI PS2_BVI
PS2_KY PS2_KY PS2_Cayman
PS2_CU PS2_CU PS2_Cuba
PS2_JM PS2_JM PS2_Jamaica
PS2_VC PS2_VC PS2_StVincentGren
PS2_SR PS2_SR PS2_Suriname
PS2_TT PS2_TT PS2_Trinidad
PS3_AR PS3_AR PS3_Argentina
PS3_BR PS3_BR PS3_Brasil
215
Type of Models – Codes Used
Detention*) Casualty Explanation of Variable
PS3_CHI PS3_CHI PS3_Chile
PS3_COL PS3_COL PS3_Colombia
PS3_CUB PS3_CUB PS3_Cuba
PS3_ECU PS3_ECU PS3_Ecuador
PS3_HN PS3_HN PS3_Honduras
PS3_MX PS3_MX PS3_Mexico
PS3_PA PS3_PA PS3_Panama
PS3_PE PS3_PE PS3_Peru
PS3_UY PS3_UY PS3_Uruguay
PS3_VE PS3_VE PS3_Venezuela
PS4_ER PS4_ER PS4_Eritrea
PS4_IN PS4_IN PS4_India
PS4_IR PS4_IR PS4_Iran
PS4_MU PS4_MU PS4_Mauritius
PS4_ZA PS4_ZA PS4_SouthAfrica
PS4_LK PS4_LK PS4_SriLanka
PS4_SD PS4_SD PS4_Sudan
PS4_TZ PS4_TZ PS4_Tanzania
PS5_ANCO PS5_ANCO PS5_Anchorage_AK
PS5_BALTI PS5_BALTI PS5_Baltimore_MD
PS5_BATR PS5_BATR PS5_BatonRouge_LA
PS5_BOST PS5_BOST PS5_Boston_MA
PS5_BROW PS5_BROW PS5_Brownsville_TX
PS5_BUFF PS5_BUFF PS5_Buffalo_NY
PS5_CHAR PS5_CHAR PS5_Charleston_SC
PS5_CHIC PS5_CHIC PS5_Chicago_IL
PS5_CLEV PS5_CLEV PS5_Cleveland_OH
PS5_CORP PS5_CORP PS5_CorpusChristi_TX
PS5_DETR PS5_DETR PS5_Detroit_MI
PS5_DULU PS5_DULU PS5_Duluth_MN
PS5_GUAM PS5_GUAM PS5_Guam
PS5_HAMP PS5_HAMP PS5_HamptonRoads_VA
PS5_HONO PS5_HONO PS5_Honolulu_HI
PS5_HOUS PS5_HOUS PS5_HoustonGalv_TX
PS5_JACK PS5_JACK PS5_Jacksonville_FL
PS5_JUNE PS5_JUNE PS5_Juneau_AK
PS5_LCHA PS5_LCHA PS5_LakeCharles_LA
PS5_LONG PS5_LONG PS5_LongIsland_NY
PS5_LANG PS5_LANG PS5_LosAngeles_CA
PS5_MASS PS5_MASS PS5_Massena_NY
PS5_MIAM PS5_MIAM PS5_Miami_FL
PS5_MOBI PS5_MOBI PS5_Mobile_AL
PS5_MORG PS5_MORG PS5_MorganCity_LA
PS5_NORL PS5_NORL PS5_NewOrleans_LA
PS5_NEWY PS5_NEWY PS5_NewYork_NY
PS5_PHIL PS5_PHIL PS5_Philadelphia_PA
PS5_POAR PS5_POAR PS5_PortArthur_TX
PS5_POCA PS5_POCA PS5_PortCarnaveral_FL
PS5_POLA PS5_POLA PS5_PortLavaca_TX
PS5_PORM PS5_PORM PS5_Portland_ME
PS5_PORO PS5_PORO PS5_Portland_OR
PS5_PORT PS5_PORT PS5_Portsmouth_NH
216
Type of Models – Codes Used
Detention*) Casualty Explanation of Variable
PS5_PROV PS5_PROV PS5_Providence_RI
PS5_PUGE PS5_PUGE PS5_PugetSound_WA
PS5_SAMO PS5_SAMO PS5_AmericanSamoa
PS5_SAND PS5_SAND PS5_SanDiego_CL
PS5_SANF PS5_SANF PS5_SanFrancisco_CL
PS5_SANJ PS5_SANJ PS5_SanJuan_PR
PS5_SANB PS5_SANB PS5_SantaBarbara_CL
PS5_SAUL PS5_SAUL PS5_SaultMarie_MI
PS5_SAVA PS5_SAVA PS5_Savannah_GA
PS5_TAMP PS5_TAMP PS5_Tampa_FL
PS5_TOLE PS5_TOLE PS5_Toledo_OH
PS5_VALD PS5_VALD PS5_Valdez_AK
PS5_WILM PS5_WILM PS5_Wilmington_NC
PS5_USVI PS5_USVI PS5_USVirginIslands
PS5_OTH PS5_OTH PS5_Other
PS6_BELL PS6_BELL PS6_BellBay_TAS
PS6_BRIS PS6_BRIS PS6_Brisbane_QLD
PS6_BUNB PS6_BUNB PS6_Bunbury_WA
PS6_BURN PS6_BURN PS6_Burnie_TAS
PS6_CAIR PS6_CAIR PS6_Cairns_QLD
PS6_DAMP PS6_DAMP PS6_Dampier_WA
PS6_DARW PS6_DARW PS6_Darwin_NT
PS6_DEVO PS6_DEVO PS6_Devonport_TAS
PS6_ESPE PS6_ESPE PS6_Esperance_WA
PS6_FREM PS6_FREM PS6_Fremantle_WA
PS6_GEEL PS6_GEEL PS6_Geelong_VIC
PS6_GERA PS6_GERA PS6_Geraldton_WA
PS6_GLAD PS6_GLAD PS6_Gladstone_QLD
PS6_GOVE PS6_GOVE PS6_Gove_NT
PS6_HAYP PS6_HAYP PS6_HayPoint_QLD
PS6_KURN PS6_KURN PS6_Kurnell_NSW
PS6_KWIN PS6_KWIN PS6_Kwinana_WA
PS6_MACK PS6_MACK PS6_Mackay_QLD
PS6_MELB PS6_MELB PS6_Melborne_VIC
PS6_NEWC PS6_NEWC PS6_Newcastle_NSW
PS6_OTH PS6_OTH PS6_Other
PS6_POAD PS6_POAD PS6_PortAdelaide_SA
PS6_POBO PS6_POBO PS6_PortBotany_NSW
PS6_POHE PS6_POHE PS6_PortHedland_WA
PS6_POKE PS6_POKE PS6_PortKembla_NSW
PS6_POWA PS6_POWA PS6_PortWalcott_WA
PS6_PORT PS6_PORT PS6_Portland_VIC
PS6_SYDN PS6_SYDN PS6_Sydney_NSW
PS6_TOWN PS6_TOWN PS6_Townsville_QLD
PS6_WALL PS6_WALL PS6_Wallaroo_SA
n/a SY_AU Australia
n/a SY_BE Belgium
n/a SY_BR Brazil
n/a SY_BG Bulgaria
n/a SY_CA Canada
n/a SY_CL Chile
n/a SY_CN China
217
Type of Models – Codes Used
Detention*) Casualty Explanation of Variable
n/a SY_HR Croatia
n/a SY_DK Denmark
n/a SY_FI Finland
n/a SY_FR France
n/a SY_DE Germany
n/a SY_GR Greece
n/a SY_IN India
n/a SY_ID Indonesia
n/a SY_IT Italy
n/a SY_JP Japan
n/a SY_MY Malaysia
n/a SY_NL Netherlands
n/a SY_NO Norway
n/a SY_OT Other
n/a SY_PH Philippines
n/a SY_PL Poland
n/a SY_PT Portugal
n/a SY_RO Romania
n/a SY_RU Russian Federation
n/a SY_SG Singapore
n/a SY_KR South Korea
n/a SY_ES Spain
n/a SY_SE Sweden
n/a SY_TW Taiwan
n/a SY_TR Turkey
n/a SY_UA Ukraine
n/a SY_UK United Kingdom
n/a SY_US United States of America
n/a SY_UN Unknown
n/a SY_VN Viet Nam
218
Appendix 5: Grouping of Countries of Ownership
The grouping of ownership of a vessel was made according to Alderton and Winchester
(1999) and is as follows:
1. Old Open Registries: Antigua and Barbuda, Bahamas, Bermuda, Cyprus, Honduras,
Liberia, Malta, Marshall Islands, Panama, St. Vincent & the Grenadines
2. New Open Registries: Barbados, Belize, Bolivia, Cambodia, Canary Islands, Cayman
Islands, Cook Islands, Equatorial Guinea, Gibraltar, Lebanon, Luxembourg,
Mauritius, Myanmar, Sri Lanka, Tuvalu and Vanuatu
3. International Registries: Anguila, British Virgin Islands, Channel Islands, DIS,
Falklands, Faeroes, Hong Kong, Isle of Man, Kerguelen Islands, Macao, Madeira,
NIS, Philippines, Sao Tome and Principe, Singapore, Turks and Caicos, Ukraine,
Wallis and Fortuna, Netherlands Antilles
4. Traditional Maritime Nations: Argentina, Australia, Austria, Belgium, Brazil,
Canada, Chile, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy,
Japan, Mexico, Netherlands, New Zealand, Norway, Portugal, Russia, South Africa,
Spain, Sweden, Switzerland, UK, Uruguay, USA, Venezuela.
5. Emerging Maritime Nations: Albania, Algeria, Angola, Azerbaijan, Bahrain,
Bangladesh, Benin, Brunei, Bulgaria, Cameroon, Cape Verde, China, Colombia,
Comoro, Congo, Costa Rica, Croatia, Cuba, Djibouti, Dominica, Dominican Republic,
Egypt, El Salvador, Ecuador, Eritrea, Estonia, Ethiopia, Fiji, Gabon, Gambia,
Georgia, Ghana, Grenada, Guatemala, Guinea, Guyana, Haiti, Hungary, India,
Indonesia, Iran, Iraq, Israel, Jamaica, Jordan, Kazakhstan, Kenya, Kiribati, North
Korea, South Korea, Kuwait, Laos, Latvia, Libya, Lithuania, Madagascar, Malaysia,
Maldives, Mauritania, Micronesia, Morocco, Mozambique, Namibia, Nicaragua,
Nigeria, Oman, Pakistan, Papua New Guinea, Paraguay, Peru, Poland, Qatar,
Romania, St. Helena, St. Kitts & Nevis, Samoa, Saudi Arabia, Senegal, Seychelles,
Sierra Leone, Slovakia, Slovenia, Solomon Islands, Somalia Republic, Sudan,
Surinam, Syria, Taiwan, Tanzania, Thailand, Togo, Trinidad, Tunisia, Turkey,
Turkmenistan, UAE, Vietnam, Yemen
6. Other/Unknown: Undefined by dataset, Unknown (Fairplay), Azores, Cameroon,
Greenland, Monaco, Puerto Rico, Serbia & Montenegro, St. Pierre & Miquel
219
Appendix 6: Results of Correspondence Analysis for Ship Type Selection
Explained Inertia chi2 = 1.7371e+004
Dim1 0.5679
Dim2 0.1878
Dim3 0.1166
Dim4 0.0453
Xpk = x coordinates Ypk = y coordinates
General’ 0.0740 -0.0587 ‘C0100′ -0.1099 -0.0162
Bulk’ 0.0506 0.1011 ‘C0200′ -0.0276 -0.1356
Oil’ -0.3927 -0.0029 ‘C0300′ 0.0897 -0.0690
Tanker’ -0.5035 -0.0768 ‘C0400’ 0.0730 0.0101
‘RoRoCargo’ 0.0180 0.0549 ‘C0500′ 0.0047 -0.0354
Other’ -0.0224 -0.0881 ‘C0600′ 0.0049 -0.0001
Chemical’ -0.6786 -0.0707 ‘C0700′ -0.0579 0.0966
Reefer’ 0.0598 -0.0430 ‘C0800’ -0.0392 0.0527
‘Passenger’ -0.0321 0.2328 ‘C0900′ 0.0172 0.1157
Gas’ -0.7187 -0.0144 ‘C1000′ -0.0698 0.2267
OBO’ -0.3603 0.1123 ‘C1100′ 0.1528 -0.0307
RoRoPax’ -0.0566 0.4441 ‘C1200′ 0.1032 -0.0243
Offshore’ 0.0127 -0.1609 ‘C1300′ 0.1057 -0.0456
Mobile’ -0.1659 -0.2607 ‘C1400′ -0.0078 0.0402
Factory’ 0.1349 -0.1266 ‘C1500′ 0.1108 -0.1742
Special’ 0.0873 -0.1850 ‘C1600’ 0.0723 -0.0883
‘HeavyLoad’ 0.0431 0.0662 ‘C1700′ -0.0342 -0.0681
HSPax’ -0.0877 0.2967 ‘C1800’ -2.9875 -0.3640
‘Container’ 0.0325 0.0296 ‘C1900’ -1.8985 -0.4299
‘C2000’ 0.0143 0.1520
‘C2100’ -0.0006 -0.0399
‘C2200’ 0.1443 0.0423
‘C2300’ 0.0657 -0.0758
‘C2500′ -0.1028 0.1714
Absolute Row Contribution Relative Row Contribution
General’ 0.0922 0.1753 General’ 0.5659 0.3559
Bulk’ 0.0164 0.1983 Bulk’ 0.0927 0.3703
Oil’ 0.2037 0.0000 Oil’ 0.9501 0.0001
Tanker’ 0.1782 0.0125 Tanker’ 0.9157 0.0213
‘RoRoCargo’ 0.0004 0.0112 ‘RoRoCargo’ 0.0097 0.0902
Other’ 0.0002 0.0095 Other’ 0.0050 0.0778
Chemical’ 0.3635 0.0119 Chemical’ 0.9622 0.0104
Reefer’ 0.0025 0.0040 Reefer’ 0.0973 0.0503
‘Passenger’ 0.0004 0.0706 ‘Passenger’ 0.0084 0.4386
Gas’ 0.1219 0.0001 Gas’ 0.8989 0.0004
OBO’ 0.0152 0.0045 OBO’ 0.6709 0.0652
RoRoPax’ 0.0025 0.4681 RoRoPax’ 0.0115 0.7086
Offshore’ 0.0000 0.0152 Offshore’ 0.0013 0.2069
Mobile’ 0.0002 0.0012 Mobile’ 0.0208 0.0514
Factory’ 0.0008 0.0020 Factory’ 0.1605 0.1413
Special’ 0.0005 0.0070 Special’ 0.0501 0.2250
‘HeavyLoad’ 0.0000 0.0003 ‘HeavyLoad’ 0.0069 0.0164
HSPax’ 0.0001 0.0050 HSPax’ 0.0244 0.2796
220
‘Container’ 0.0013 0.0033 ‘Container’ 0.0337 0.0280
Absolute Column Contribution Relative Column Contribution
‘C0100’ 0.0190 0.0012 ‘C0100’ 0.3349 0.0072
‘C0200’ 0.0011 0.0781 ‘C0200’ 0.0159 0.3833
‘C0300’ 0.0076 0.0136 ‘C0300’ 0.1959 0.1160
‘C0400’ 0.0028 0.0002 ‘C0400’ 0.1558 0.0030
‘C0500’ 0.0000 0.0031 ‘C0500’ 0.0013 0.0751
‘C0600’ 0.0001 0.0000 ‘C0600’ 0.0157 0.0000
‘C0700’ 0.0141 0.1192 ‘C0700’ 0.2146 0.5976
‘C0800’ 0.0007 0.0039 ‘C0800’ 0.0780 0.1406
‘C0900’ 0.0011 0.1466 ‘C0900’ 0.0164 0.7407
‘C1000’ 0.0009 0.0273 ‘C1000’ 0.0458 0.4831
‘C1100’ 0.0099 0.0012 ‘C1100’ 0.3434 0.0139
‘C1200’ 0.0182 0.0031 ‘C1200’ 0.2879 0.0160
‘C1300’ 0.0055 0.0031 ‘C1300’ 0.1797 0.0334
‘C1400’ 0.0001 0.0090 ‘C1400’ 0.0052 0.1377
‘C1500’ 0.0423 0.3164 ‘C1500’ 0.2760 0.6820
‘C1600’ 0.0057 0.0257 ‘C1600’ 0.2056 0.3066
‘C1700’ 0.0024 0.0292 ‘C1700’ 0.0718 0.2851
‘C1800’ 0.7437 0.0334 ‘C1800’ 0.9723 0.0144
‘C1900’ 0.1107 0.0172 ‘C1900’ 0.7992 0.0410
‘C2000’ 0.0002 0.0597 ‘C2000’ 0.0026 0.2896
‘C2100’ 0.0000 0.0006 ‘C2100’ 0.0000 0.0388
‘C2200’ 0.0002 0.0000 ‘C2200’ 0.0922 0.0079
‘C2300’ 0.0014 0.0056 ‘C2300’ 0.1593 0.2120
‘C2500’ 0.0122 0.1027 ‘C2500′ 0.2260 0.6280
QLTx = quality rows QLTy = quality columns
General’ 0.9218 ‘C0100′ 0.3422
Bulk’ 0.4631 ‘C0200′ 0.3991
Oil’ 0.9501 ‘C0300′ 0.3119
Tanker’ 0.9370 ‘C0400’ 0.1588
‘RoRoCargo’ 0.0999 ‘C0500′ 0.0764
Other’ 0.0828 ‘C0600′ 0.0157
Chemical’ 0.9726 ‘C0700′ 0.8122
Reefer’ 0.1476 ‘C0800’ 0.2187
‘Passenger’ 0.4470 ‘C0900′ 0.7570
Gas’ 0.8993 ‘C1000′ 0.5289
OBO’ 0.7362 ‘C1100′ 0.3573
RoRoPax’ 0.7201 ‘C1200′ 0.3039
Offshore’ 0.2081 ‘C1300′ 0.2131
Mobile’ 0.0722 ‘C1400′ 0.1430
Factory’ 0.3018 ‘C1500′ 0.9580
Special’ 0.2750 ‘C1600’ 0.5122
‘HeavyLoad’ 0.0233 ‘C1700′ 0.3570
HSPax’ 0.3041 ‘C1800’ 0.9867
‘Container’ 0.0617 ‘C1900’ 0.8402
‘C2000’ 0.2921
‘C2100’ 0.0388
‘C2200’ 0.1001
‘C2300’ 0.3713
‘C2500’ 0.8540
221
Appendix 7: Step 3: Final Models: General Cargo
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 13:21
Sample: 1 66921 IF ST_CARIB=0 AND OUTLIER=0
Included observations: 66473
Convergence achieved after 8 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
ST_PMOU -2.396743 0.258563 -9.269484 0.0000
ST_VINA -2.776742 0.574273 -4.835225 0.0000
ST_INDIA -5.423454 0.219420 -24.71725 0.0000
ST_USCG -3.560540 0.292474 -12.17386 0.0000
ST_AMSA -4.323396 0.169246 -25.54511 0.0000
AGEBYGC1+AGEBYGC3+AGEBYGC4+AGEBY
GC5+AGEBYGC6 0.303474 0.041264 7.354373 0.0000
SIZEGC1+SIZEGC3+SIZEGC5 -0.236769 0.023654 -10.00971 0.0000
CLDNVGC1+CLDNVGC4 -0.262804 0.106304 -2.472195 0.0134
CLIRSGC4 1.068258 0.393362 2.715712 0.0066
CLNKKGC3+CLNKKGC4 -0.389718 0.197598 -1.972275 0.0486
CLPRSGC5 1.556894 0.442767 3.516284 0.0004
CLRINGC4 1.364312 0.423487 3.221615 0.0013
FL_AGGC1 0.159694 0.085702 1.863360 0.0624
FL_CNGC4 -1.926337 0.709392 -2.715477 0.0066
FL_CYGC1+FL_CYGC3+FL_CYGC6 0.241435 0.088109 2.740176 0.0061
FL_EGGC1 0.520382 0.240305 2.165504 0.0303
FL_GEGC1+FL_GEGC4 0.383888 0.151440 2.534912 0.0112
FL_HKGC5 1.782175 0.658215 2.707589 0.0068
FL_KHGC1+FL_KHGC4+FL_KHGC5 0.487254 0.109845 4.435822 0.0000
FL_KPGC1+FL_KPGC4 0.742358 0.196086 3.785872 0.0002
FL_PAGC1+FL_PAGC3+FL_PAGC4+FL_PAGC
5 0.444409 0.071590 6.207729 0.0000
FL_PHGC5 1.850181 0.708661 2.610811 0.0090
FL_RUCG3+FL_RUGC1 0.479156 0.089758 5.338282 0.0000
FL_TOGC1 0.658288 0.271983 2.420325 0.0155
FL_TRGC1+FL_TRGC3 0.461756 0.080259 5.753360 0.0000
FL_TVGC1 1.143535 0.443701 2.577262 0.0100
FL_UAGC1+FL_UAGC3 0.375346 0.136738 2.744992 0.0061
FL_VCGC1+FL_VCGC3 0.408166 0.075652 5.395286 0.0000
C0100GC1 0.608328 0.028775 21.14102 0.0000
C0100GC3 0.289633 0.071321 4.060949 0.0000
C0100GC4 0.694804 0.085867 8.091633 0.0000
C0200GC1 0.323731 0.030793 10.51301 0.0000
C0200GC3 0.426773 0.101380 4.209658 0.0000
C0200GC4 1.095077 0.147104 7.444244 0.0000
C0200GC5 0.642589 0.302999 2.120764 0.0339
C0200GC6 1.262413 0.220674 5.720711 0.0000
C0300GC1 0.112677 0.031673 3.557575 0.0004
222
C0500GC1+C0500GC3+C0500GC4+C0500GC6 -0.134908 0.040555 -3.326539 0.0009
C0600GC1 0.273745 0.016976 16.12508 0.0000
C0600GC3 0.285547 0.045780 6.237320 0.0000
C0600GC4 0.551724 0.080190 6.880177 0.0000
C0700GC1 0.286724 0.018553 15.45433 0.0000
C0700GC3 0.376301 0.067764 5.553115 0.0000
C0700GC4 0.595082 0.083349 7.139609 0.0000
C0700GC5 0.532211 0.124600 4.271356 0.0000
C0900GC1 0.264649 0.019802 13.36500 0.0000
C0900GC4 0.293164 0.083018 3.531341 0.0004
C0900GC5 -0.355003 0.122205 -2.904985 0.0037
C0900GC6 0.381997 0.074775 5.108619 0.0000
C1000GC1+C1000GC3+C1000GC4+C1000GC5
+C1000GC6 0.593258 0.096774 6.130379 0.0000
C1100GC1 0.159904 0.059198 2.701185 0.0069
C1100GC3 0.957218 0.234715 4.078215 0.0000
C1200GC1+C1200GC4+C1200GC5+C1200GC6 0.260164 0.022788 11.41663 0.0000
C1300GC3+C1300GC4 0.445124 0.126868 3.508561 0.0005
C1400GC1+C1400GC3+C1400GC5 0.265603 0.022441 11.83583 0.0000
C1500GC1+C1500GC3+C1500GC4 0.229082 0.017127 13.37514 0.0000
C1600GC1 0.388279 0.032125 12.08635 0.0000
C1600GC3 0.207432 0.095971 2.161413 0.0307
C1600GC4 0.971456 0.204065 4.760533 0.0000
C1600GC6 0.476381 0.092952 5.125027 0.0000
C1700GC1+C1700GC4+C1700GC5+C1700GC6 0.528471 0.024906 21.21886 0.0000
C2500GC1 0.469264 0.032979 14.22920 0.0000
C2500GC3 0.809943 0.172625 4.691917 0.0000
C2500GC5 2.245713 0.324175 6.927470 0.0000
C2500GC6 1.098274 0.213912 5.134242 0.0000
OWOORGC3 -1.406355 0.529887 -2.654068 0.0080
OWIORGC3+OWIORGC5 -1.283822 0.525952 -2.440949 0.0146
OWTMNGC1 -0.253373 0.053410 -4.743950 0.0000
OWTMNGC3 -1.601115 0.504540 -3.173413 0.0015
OWEMNGC3 -1.308722 0.510937 -2.561418 0.0104
OWUNKGC1 0.404380 0.078866 5.127405 0.0000
OWUNKGC3 -1.278068 0.550262 -2.322655 0.0202
PS1_BE -1.128189 0.102522 -11.00440 0.0000
PS1_HR -0.642173 0.123584 -5.196264 0.0000
PS1_DK -0.576984 0.136279 -4.233855 0.0000
PS1_FI -0.604053 0.178547 -3.383153 0.0007
PS1_FR -0.388013 0.093980 -4.128663 0.0000
PS1_DE -0.873199 0.098076 -8.903262 0.0000
PS1_GR -1.067856 0.097729 -10.92667 0.0000
PS1_IE -0.962211 0.168072 -5.725002 0.0000
PS1_NO -0.340505 0.147174 -2.313621 0.0207
PS1_NL -1.305595 0.121707 -10.72735 0.0000
PS1_PL -1.319757 0.133728 -9.868933 0.0000
PS1_PT -0.633462 0.099375 -6.374443 0.0000
PS1_RU -1.677765 0.098799 -16.98167 0.0000
PS1_ES -0.575641 0.075580 -7.616356 0.0000
PS1_SE -1.364781 0.181103 -7.535948 0.0000
223
PS1_UK -1.325170 0.101397 -13.06907 0.0000
PS3_AR -7.784620 1.188260 -6.551276 0.0000
PS3_CHI 2.135719 0.208660 10.23540 0.0000
PS3_CUB 2.103693 0.216106 9.734561 0.0000
PS3_HN 1.585447 0.365438 4.338486 0.0000
PS4_IR 0.406926 0.176013 2.311906 0.0208
PS4_SD 1.057724 0.346769 3.050222 0.0023
PS5_NORL 0.891379 0.328429 2.714067 0.0066
PS6_BRIS -0.650901 0.284299 -2.289497 0.0221
PS6_CAIR -1.872774 0.792762 -2.362341 0.0182
PS6_DARW -3.263450 0.720632 -4.528594 0.0000
PS6_TOWN -1.197512 0.422817 -2.832225 0.0046
Mean dependent var 0.083944 S.D. dependent var 0.277306
S.E. of regression 0.212382 Akaike info criterion 0.323725
Sum squared resid 2993.870 Schwarz criterion 0.337285
Log likelihood -10660.48 Hannan-Quinn criter. 0.327917
Avg. log likelihood -0.160373
Obs with Dep=0 60893 Total obs 66473
Obs with Dep=1 5580
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 13:21
Sample: 1 66921 IF ST_CARIB=0 AND OUTLIER=0
Included observations: 66473
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 5.E-07 0.0049 6642 6625.49 5 21.5071 6647 12.7107
2 0.0049 0.0076 6633 6605.44 14 41.5581 6647 18.3894
3 0.0076 0.0107 6612 6586.71 35 60.2907 6647 10.7060
4 0.0107 0.0147 6598 6564.38 50 83.6197 6648 13.6891
5 0.0147 0.0199 6561 6532.33 86 114.666 6647 7.29239
6 0.0199 0.0275 6536 6491.18 111 155.815 6647 13.1991
7 0.0275 0.0409 6447 6426.30 201 221.702 6648 1.99980
8 0.0409 0.0735 6228 6286.09 419 360.908 6647 9.88740
9 0.0735 0.2029 5673 5845.46 974 801.542 6647 42.1939
10 0.2029 1.0000 2963 2929.61 3685 3718.39 6648 0.68043
Total 60893 60893.0 5580 5580.00 66473 130.748
H-L Statistic: 130.7482 Prob. Chi-Sq(8) 0.0000
Andrews Statistic: 361.0461 Prob. Chi-Sq(10) 0.0000
224
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 13:21
Sample: 1 66921 IF ST_CARIB=0 AND OUTLIER=0
Included observations: 66473
Prediction Assessment (success cutoff C = 0.0842)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 53334 990 54324 60893 5580 66473
P(Dep=1)>C 7559 4590 12149 0 0 0
Total 60893 5580 66473 60893 5580 66473
Correct 53334 4590 57924 60893 0 60893
% Correct 87.59 82.26 87.14 100.00 0.00 91.61
% Incorrect 12.41 17.74 12.86 0.00 100.00 8.39
Total Gain* -12.41 82.26 -4.47
Percent Gain** NA 82.26 -53.21
225
Appendix 8: Step 3: Final Models: Dry Bulk
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/13/06 Time: 10:41
Sample: 1 48103 IF ST_DB2=0 AND OUTLIER=0
Included observations: 47777
Convergence achieved after 11 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
ST_DB1 -4.863602 0.460705 -10.55687 0.0000
ST_DB3 -10.25991 0.951265 -10.78554 0.0000
ST_DB4 -4.915276 0.494817 -9.933529 0.0000
ST_DB5 -9.713906 0.773077 -12.56526 0.0000
ST_DB6 -4.846630 0.467823 -10.35997 0.0000
AGEBYDB1+AGEBYDB3+AGEBYDB4+AGEBY
DB5+AGEBYDB6 0.605733 0.064172 9.439227 0.0000
SIZEDB1+SIZEDB3+SIZEDB4+SIZEDB5+SIZED
B6 -0.098601 0.039425 -2.501013 0.0124
CLCRRDB5 1.273347 0.525719 2.422106 0.0154
CLLRDB5 -0.964077 0.333624 -2.889710 0.0039
FL_BBDB5 -4.043765 0.831707 -4.862006 0.0000
FL_BRDB3 -2.921903 1.185654 -2.464380 0.0137
FL_BRDB5 1.823917 0.663879 2.747363 0.0060
FL_CYDB6 0.528363 0.174202 3.033043 0.0024
FL_GEDB1 1.231493 0.520730 2.364934 0.0180
FL_HKDB4 -2.071076 0.719395 -2.878915 0.0040
FL_KHDB1 1.236079 0.343956 3.593712 0.0003
FL_MTDB6 0.474747 0.208260 2.279593 0.0226
FL_MYDB3 1.692650 0.658565 2.570209 0.0102
FL_PHDB3 1.165037 0.506899 2.298360 0.0215
FL_PLDB3 1.786038 0.611841 2.919119 0.0035
FL_TRDB1+FL_TRDB3+FL_TRDB4+FL_TRDB5
+FL_TRDB6 0.272667 0.121374 2.246500 0.0247
C0100DB1+C0100DB3+C0100DB4+C0100DB5 0.492153 0.055502 8.867353 0.0000
C0200DB1 0.166678 0.069387 2.402165 0.0163
C0200DB3 0.437528 0.149470 2.927196 0.0034
C0200DB5 2.269534 0.292007 7.772186 0.0000
C0200DB6 0.943160 0.143645 6.565890 0.0000
C0300DB4+C0300DB5 0.414519 0.121660 3.407189 0.0007
C0400DB6 -0.484642 0.169869 -2.853030 0.0043
C0600DB1 0.272873 0.027685 9.856377 0.0000
C0600DB3 0.235242 0.073153 3.215756 0.0013
C0600DB4 0.376844 0.089535 4.208904 0.0000
C0600DB5 0.768639 0.131290 5.854525 0.0000
C0600DB6 0.101006 0.040829 2.473895 0.0134
C0700DB1 0.241506 0.028153 8.578302 0.0000
C0700DB3 0.267494 0.085915 3.113475 0.0018
C0700DB4 0.737742 0.115981 6.360903 0.0000
C0700DB5 0.726521 0.174089 4.173261 0.0000
226
C0700DB6 0.368198 0.040618 9.064800 0.0000
C0800DB4 -0.759015 0.228804 -3.317317 0.0009
C0900DB1+C0900DB3+C0900DB4+C0900DB5 0.211114 0.027856 7.578800 0.0000
C1100DB3 0.750629 0.322034 2.330904 0.0198
C1200DB1+C1200DB6 0.188567 0.030245 6.234700 0.0000
C1400DB1+C1400DB3+C1400DB4+C1400DB6 0.232820 0.030053 7.746960 0.0000
C1500DB1+C1500DB4 0.168953 0.037561 4.498104 0.0000
C1600DB1 0.201671 0.061314 3.289140 0.0010
C1600DB3 0.566916 0.202119 2.804860 0.0050
C1600DB4 1.058992 0.248197 4.266733 0.0000
C1600DB6 0.450457 0.062530 7.203794 0.0000
C1700DB1 0.524241 0.043123 12.15684 0.0000
C1700DB3 0.834377 0.123995 6.729146 0.0000
C1700DB4 0.893838 0.164103 5.446821 0.0000
C1700DB5 0.655515 0.147198 4.453278 0.0000
C1700DB6 0.724802 0.082148 8.823176 0.0000
C1800DB5 2.666533 0.845398 3.154175 0.0016
C2000DB5 2.289860 0.294759 7.768588 0.0000
C2300DB1 0.299613 0.123344 2.429076 0.0151
C2500DB1 0.599240 0.045846 13.07077 0.0000
C2500DB3 0.552931 0.159583 3.464858 0.0005
C2500DB5 2.256296 0.401158 5.624456 0.0000
C2500DB6 1.058245 0.083928 12.60896 0.0000
C2600DB1 0.506175 0.126805 3.991748 0.0001
C9900DB5 1.135346 0.424252 2.676111 0.0074
OWUNKDB4 -1.256157 0.606619 -2.070750 0.0384
PS1_BE -0.909947 0.162677 -5.593574 0.0000
PS1_GR -1.787809 0.433777 -4.121497 0.0000
PS1_IE -1.164652 0.470797 -2.473785 0.0134
PS1_NL -0.995044 0.175002 -5.685912 0.0000
PS1_RU -1.154220 0.184050 -6.271239 0.0000
PS3_BR 3.512539 0.762958 4.603845 0.0000
PS3_CHI 5.940680 0.813183 7.305467 0.0000
PS3_COL 3.718919 0.928646 4.004668 0.0001
PS3_CUB 5.492208 0.837512 6.557769 0.0000
PS3_HN 5.199727 0.938440 5.540819 0.0000
PS5_BATR 3.158521 0.997264 3.167188 0.0015
PS5_BROW 3.793653 1.198999 3.164016 0.0016
PS5_CHAR 5.154124 0.690391 7.465518 0.0000
PS5_CORP 4.294424 0.802039 5.354382 0.0000
PS5_DULU 4.082264 0.930248 4.388360 0.0000
PS5_HAMP 3.947380 0.675171 5.846485 0.0000
PS5_HONO 3.711200 1.344611 2.760055 0.0058
PS5_HOUS 3.933235 0.741468 5.304662 0.0000
PS5_JACK 6.196477 0.750042 8.261508 0.0000
PS5_LANG 4.192381 0.727509 5.762648 0.0000
PS5_MOBI 4.007495 0.735921 5.445551 0.0000
PS5_NORL 4.545155 0.649973 6.992837 0.0000
PS5_POAR 3.910559 1.215048 3.218439 0.0013
PS5_POCA 2.895390 0.895584 3.232962 0.0012
PS5_POLA 4.544957 0.904334 5.025754 0.0000
227
PS5_PROV 5.430488 0.849466 6.392822 0.0000
PS5_PUGE 3.702121 1.171584 3.159928 0.0016
PS5_SANF 3.534287 0.828315 4.266842 0.0000
PS5_SANJ 4.000131 0.799937 5.000556 0.0000
PS5_SAVA 4.173261 0.782037 5.336395 0.0000
PS5_WILM 4.029798 0.985652 4.088461 0.0000
PS5_USVI 4.037977 1.060904 3.806167 0.0001
PS6_NEWC 0.512666 0.131240 3.906340 0.0001
Mean dependent var 0.046173 S.D. dependent var 0.209861
S.E. of regression 0.167850 Akaike info criterion 0.213693
Sum squared resid 1343.345 Schwarz criterion 0.231324
Log likelihood -5008.805 Hannan-Quinn criter. 0.219227
Avg. log likelihood -0.104837
Obs with Dep=0 45571 Total obs 47777
Obs with Dep=1 2206
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/13/06 Time: 10:41
Sample: 1 48103 IF ST_DB2=0 AND OUTLIER=0
Included observations: 47777
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 9.E-07 0.0001 4777 4776.67 0 0.33305 4777 0.33307
2 0.0001 0.0022 4775 4773.10 3 4.90243 4778 0.73902
3 0.0022 0.0042 4768 4762.62 10 15.3793 4778 1.88762
4 0.0042 0.0069 4760 4750.34 17 26.6646 4777 3.52259
5 0.0069 0.0101 4755 4737.89 23 40.1120 4778 7.36186
6 0.0101 0.0151 4742 4718.69 36 59.3077 4778 9.27496
7 0.0151 0.0208 4742 4692.63 35 84.3714 4777 29.4100
8 0.0208 0.0342 4646 4651.99 132 126.009 4778 0.29259
9 0.0342 0.0845 4473 4526.99 305 251.008 4778 12.2576
10 0.0845 1.0000 3133 3180.09 1645 1597.91 4778 2.08477
Total 45571 45571.0 2206 2206.00 47777 67.1641
H-L Statistic: 67.1641 Prob. Chi-Sq(8) 0.0000
Andrews Statistic: 4190.678 Prob. Chi-Sq(10) 0.0000
228
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/13/06 Time: 10:41
Sample: 1 48103 IF ST_DB2=0 AND OUTLIER=0
Included observations: 47777
Prediction Assessment (success cutoff C = 0.0462)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 39896 349 40245 45571 2206 47777
P(Dep=1)>C 5675 1857 7532 0 0 0
Total 45571 2206 47777 45571 2206 47777
Correct 39896 1857 41753 45571 0 45571
% Correct 87.55 84.18 87.39 100.00 0.00 95.38
% Incorrect 12.45 15.82 12.61 0.00 100.00 4.62
Total Gain* -12.45 84.18 -7.99
Percent Gain** NA 84.18 -173.07
229
Appendix 9: Step 3: Final Models: Tanker
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 17:28
Sample: 1 34232 IF ST_TA2=0 AND OUTLIER=0
Included observations: 34045
Convergence achieved after 9 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
ST_TA1 -5.535312 0.410159 -13.49554 0.0000
ST_TA3 -6.690566 0.505985 -13.22284 0.0000
ST_TA4 -5.188466 0.417693 -12.42171 0.0000
ST_TA5 -7.030510 0.466257 -15.07862 0.0000
ST_TA6 -4.828701 0.450807 -10.71123 0.0000
AGEBYTA1+AGEBYTA3+AGEBYTA4+AGEBYT
A5+AGEBYTA6 0.584318 0.079500 7.349933 0.0000
SIZETA1+SIZETA3+SIZETA4+SIZETA5+SIZET
A6 -0.077829 0.032430 -2.399903 0.0164
CLDNVTA1+CLDNVTA3+CLDNVTA4+CLDNVT
A5+CLDNVTA6 -0.433907 0.116753 -3.716448 0.0002
CLLRTA1+CLLRTA3+CLLRTA4+CLLRTA5+CLL
RTA6 -0.304074 0.113948 -2.668539 0.0076
FL_DKTA5 2.425072 0.651415 3.722774 0.0002
FL_DZTA1 1.415417 0.427059 3.314338 0.0009
FL_GRTA1+FL_GRTA6 0.769372 0.218595 3.519622 0.0004
FL_INTA1 1.269371 0.645176 1.967481 0.0491
FL_LRTA5+FL_LRTA6 0.569171 0.227455 2.502344 0.0123
FL_MTTA1+FL_MTTA3+FL_MTTA4+FL_MTTA5 0.554515 0.133630 4.149638 0.0000
FL_MYTA5 2.517000 0.683128 3.684524 0.0002
FL_PATA1+FL_PATA3 0.628293 0.147813 4.250579 0.0000
FL_RUTA1+FL_RUTA4 -1.275017 0.482062 -2.644923 0.0082
FL_SGTA1+FL_SGTA4+FL_SGTA5+FL_SGTA6 0.748859 0.194485 3.850473 0.0001
FL_SVTA1+FL_SVTA3 0.864004 0.316449 2.730310 0.0063
FL_TRTA1 0.928918 0.230715 4.026248 0.0001
C0100TA1+C0100TA3+C0100TA4+C0100TA5+
C0100TA6 0.418259 0.049272 8.488800 0.0000
C0200TA3 0.410550 0.153947 2.666827 0.0077
C0200TA4 1.424165 0.260327 5.470679 0.0000
C0200TA5 1.845868 0.380577 4.850178 0.0000
C0200TA6 0.922030 0.307034 3.003018 0.0027
C0300TA1 0.230829 0.086337 2.673575 0.0075
C0300TA4 0.992158 0.302599 3.278784 0.0010
C0600TA1 0.205752 0.045251 4.546896 0.0000
C0600TA4 0.407612 0.105758 3.854205 0.0001
C0600TA5 0.602255 0.102714 5.863428 0.0000
C0700TA1+C0700TA3+C0700TA4+C0700TA5+
C0700TA6 0.318834 0.030905 10.31643 0.0000
C0900TA1+C0900TA4+C0900TA5 0.364193 0.044010 8.275235 0.0000
C1000TA1+C1000TA3+C1000TA4 0.549623 0.196366 2.798973 0.0051
C1100TA1+C1100TA5+C1100TA6 0.160004 0.073994 2.162402 0.0306
230
C1200TA1+C1200TA3+C1200TA4+C1200TA5+
C1200TA6 0.196855 0.055233 3.564102 0.0004
C1300TA1 0.504547 0.130558 3.864534 0.0001
C1300TA3 0.543656 0.194147 2.800232 0.0051
C1400TA1+C1400TA3+C1400TA4+C1400TA5+
C1400TA6 0.140424 0.030735 4.568870 0.0000
C1500TA1+C1500TA3+C1500TA4+C1500TA5+
C1500TA6 0.168793 0.053770 3.139189 0.0017
C1600TA3 -0.457937 0.171973 -2.662841 0.0077
C1600TA4 1.123342 0.419671 2.676719 0.0074
C1600TA6 0.595624 0.156004 3.818017 0.0001
C1700TA1 0.620208 0.070545 8.791703 0.0000
C1700TA3 0.300404 0.108723 2.763027 0.0057
C1700TA4 1.258187 0.196260 6.410826 0.0000
C1800TA5+C1800TA6 0.795913 0.257512 3.090783 0.0020
C2000TA5 1.856354 0.531754 3.491004 0.0005
C2300TA1 -0.875584 0.301646 -2.902689 0.0037
C2300TA6 0.963999 0.444912 2.166718 0.0303
C2500TA1+C2500TA5+C2500TA6 0.958027 0.073315 13.06732 0.0000
PS1_BE -0.973391 0.314221 -3.097789 0.0019
PS1_NL -0.457643 0.189581 -2.413973 0.0158
PS1_RU -0.846889 0.324301 -2.611427 0.0090
PS1_IT 0.802182 0.135479 5.921062 0.0000
PS1_NO 0.908834 0.298974 3.039843 0.0024
PS3_CHI 2.319149 0.406718 5.702112 0.0000
PS3_CUB 2.201791 0.344635 6.388761 0.0000
PS5_LANG 1.304536 0.537759 2.425874 0.0153
PS5_POLA 2.179698 0.657943 3.312899 0.0009
PS5_SANF 1.577417 0.438778 3.595022 0.0003
PS5_SANJ 1.329240 0.473945 2.804628 0.0050
PS5_SAVA 1.638115 0.524612 3.122529 0.0018
PS5_USVI 1.301552 0.529656 2.457354 0.0140
Mean dependent var 0.031135 S.D. dependent var 0.173686
S.E. of regression 0.140591 Akaike info criterion 0.154295
Sum squared resid 671.6644 Schwarz criterion 0.170152
Log likelihood -2562.481 Hannan-Quinn criter. 0.159352
Avg. log likelihood -0.075267
Obs with Dep=0 32985 Total obs 34045
Obs with Dep=1 1060
231
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 17:28
Sample: 1 34232 IF ST_TA2=0 AND OUTLIER=0
Included observations: 34045
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 0.0002 0.0011 3404 3401.53 0 2.47066 3404 2.47245
2 0.0011 0.0017 3403 3400.11 2 4.88934 3405 1.70990
3 0.0017 0.0025 3404 3396.80 0 7.20099 3404 7.21626
4 0.0025 0.0036 3401 3394.69 4 10.3067 3405 3.87082
5 0.0036 0.0054 3395 3388.99 9 15.0106 3404 2.41746
6 0.0054 0.0080 3397 3382.62 8 22.3764 3405 9.29764
7 0.0080 0.0121 3373 3370.38 31 33.6224 3404 0.20658
8 0.0121 0.0196 3359 3352.87 46 52.1292 3405 0.73186
9 0.0196 0.0450 3298 3304.82 106 99.1808 3404 0.48293
10 0.0450 1.0000 2551 2592.19 854 812.813 3405 2.74147
Total 32985 32985.0 1060 1060.00 34045 31.1474
H-L Statistic: 31.1474 Prob. Chi-Sq(8) 0.0001
Andrews Statistic: 6562.7804 Prob. Chi-Sq(10) 0.0000
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 17:28
Sample: 1 34232 IF ST_TA2=0 AND OUTLIER=0
Included observations: 34045
Prediction Assessment (success cutoff C = 0.0312)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 29293 142 29435 32985 1060 34045
P(Dep=1)>C 3692 918 4610 0 0 0
Total 32985 1060 34045 32985 1060 34045
Correct 29293 918 30211 32985 0 32985
% Correct 88.81 86.60 88.74 100.00 0.00 96.89
% Incorrect 11.19 13.40 11.26 0.00 100.00 3.11
Total Gain* -11.19 86.60 -8.15
Percent Gain** NA 86.60 -261.70
232
233
Appendix 10: Step 3: Final Models: Container
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 13:52
Sample: 1 18273 IF ST_CO2=0 AND OUTLIER=0
Included observations: 18211
Convergence achieved after 9 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
ST_CO1 -3.314751 0.777403 -4.263877 0.0000
ST_CO3 -4.604093 0.812864 -5.664041 0.0000
ST_CO4 -4.003957 0.760576 -5.264378 0.0000
ST_CO5 -7.255512 0.944948 -7.678210 0.0000
ST_CO6 -3.855125 0.867522 -4.443838 0.0000
AGEBYCO1+AGEBYCO3+AGEBYCO4+AGEBY
CO5+AGEBYCO6 0.399090 0.100698 3.963224 0.0001
SIZECO1+SIZECO3+SIZECO4+SIZECO5+SIZE
CO6 -0.234550 0.075783 -3.095044 0.0020
CLDNVCO5 1.624808 0.604100 2.689635 0.0072
CLGLCO5 1.168670 0.544606 2.145900 0.0319
CLKRSCO1 1.518730 0.577507 2.629802 0.0085
CLLRCO3 -3.746199 1.447404 -2.588219 0.0096
CLNKKCO5 1.608657 0.633728 2.538402 0.0111
FL_AECO1 1.834057 0.601684 3.048205 0.0023
FL_GRCO1+FL_GRCO5 0.864207 0.403136 2.143710 0.0321
FL_PACO6 0.751744 0.334081 2.250185 0.0244
FL_SVCO5 3.110014 0.639569 4.862670 0.0000
C0100CO1+C0100CO3+C0100CO4+C0100CO5
+C0100CO6 0.430801 0.080231 5.369531 0.0000
C0200CO3 0.779165 0.192860 4.040057 0.0001
C0200CO4 2.088648 0.466835 4.474056 0.0000
C0200CO5 2.026444 0.369552 5.483518 0.0000
C0200CO6 1.280295 0.285648 4.482076 0.0000
C0300CO6 -2.040599 0.894815 -2.280470 0.0226
C0400CO1+C0400CO6 0.419840 0.143674 2.922170 0.0035
C0600CO1+C0600CO3+C0600CO4+C0600CO5
+C0600CO6 0.341900 0.056925 6.006123 0.0000
C0700CO4 0.556971 0.210317 2.648239 0.0081
C0700CO5 1.074654 0.191472 5.612597 0.0000
C0700CO6 0.513815 0.127318 4.035685 0.0001
C0900CO1 0.316889 0.074006 4.281924 0.0000
C0900CO3 0.607121 0.120382 5.043295 0.0000
C0900CO4 0.992812 0.252719 3.928523 0.0001
C0900CO6 0.682726 0.212296 3.215912 0.0013
C1200CO1+C1200CO3+C1200CO4+C1200CO5
+C1200CO6 0.527114 0.080427 6.553939 0.0000
C1400CO1+C1400CO3+C1400CO5 0.496323 0.063345 7.835286 0.0000
C1600CO1+C1600CO3+C1600CO5+C1600CO6 0.706465 0.120606 5.857616 0.0000
C1700CO1+C1700CO4+C1700CO5+C1700CO6 0.755629 0.092480 8.170691 0.0000
C2100CO1+C2100CO3 1.226981 0.340824 3.600047 0.0003
234
C2500CO1 1.062322 0.123474 8.603605 0.0000
C2500CO5 0.822388 0.289862 2.837167 0.0046
C2500CO6 2.899858 0.416603 6.960726 0.0000
OWOORCO1+OWOORCO5 1.855869 0.375055 4.948251 0.0000
PS1_BE -2.284398 0.572662 -3.989083 0.0001
PS1_DE -0.644273 0.309791 -2.079700 0.0376
PS1_ES -0.861510 0.402300 -2.141462 0.0322
PS1_NL -1.226030 0.459249 -2.669642 0.0076
PS1_PL -5.910341 1.765251 -3.348159 0.0008
PS1_PT -1.014073 0.380828 -2.662811 0.0077
PS1_UK -2.009466 0.501136 -4.009824 0.0001
PS3_CHI 1.640829 0.390182 4.205294 0.0000
PS5_CHAR 3.307899 0.622088 5.317414 0.0000
PS5_HOUS 1.991272 0.840023 2.370498 0.0178
PS5_JACK 2.503537 0.766680 3.265426 0.0011
PS5_LANG 2.132072 0.585276 3.642846 0.0003
PS5_PUGE 1.873799 0.730002 2.566843 0.0103
PS5_SANJ 2.271206 0.586286 3.873891 0.0001
PS5_SAVA 1.818636 0.718386 2.531560 0.0114
PS6_MELB 1.095316 0.422420 2.592956 0.0095
PS6_SYDN 1.408712 0.615070 2.290329 0.0220
PS6_POBO 1.018581 0.349593 2.913617 0.0036
Mean dependent var 0.023392 S.D. dependent var 0.151151
S.E. of regression 0.122979 Akaike info criterion 0.123128
Sum squared resid 274.5452 Schwarz criterion 0.148001
Log likelihood -1063.139 Hannan-Quinn criter. 0.131303
Avg. log likelihood -0.058379
Obs with Dep=0 17785 Total obs 18211
Obs with Dep=1 426
235
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 13:52
Sample: 1 18273 IF ST_CO2=0 AND OUTLIER=0
Included observations: 18211
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 1.E-05 0.0004 1821 1820.60 0 0.39931 1821 0.39939
2 0.0004 0.0009 1821 1819.79 0 1.21397 1821 1.21478
3 0.0009 0.0016 1821 1818.73 0 2.27098 1821 2.27382
4 0.0016 0.0025 1821 1817.18 0 3.81798 1821 3.82600
5 0.0025 0.0037 1819 1815.46 2 5.54358 1821 2.27205
6 0.0037 0.0056 1816 1812.75 5 8.24782 1821 1.28474
7 0.0056 0.0082 1812 1808.57 9 12.4298 1821 0.95289
8 0.0082 0.0125 1810 1802.45 11 18.5480 1821 3.10322
9 0.0125 0.0278 1782 1788.05 39 32.9458 1821 1.13303
10 0.0278 1.0000 1462 1481.42 360 340.583 1822 1.36151
Total 17785 17785.0 426 426.000 18211 17.8214
H-L Statistic: 17.8214 Prob. Chi-Sq(8) 0.0226
Andrews Statistic: 6807.503 Prob. Chi-Sq(10) 0.0000
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 13:52
Sample: 1 18273 IF ST_CO2=0 AND OUTLIER=0
Included observations: 18211
Prediction Assessment (success cutoff C = 0.024)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 16094 60 16154 17785 426 18211
P(Dep=1)>C 1691 366 2057 0 0 0
Total 17785 426 18211 17785 426 18211
Correct 16094 366 16460 17785 0 17785
% Correct 90.49 85.92 90.38 100.00 0.00 97.66
% Incorrect 9.51 14.08 9.62 0.00 100.00 2.34
Total Gain* -9.51 85.92 -7.28
Percent Gain** NA 85.92 -311.03
236
237
Appendix 11: Step 1: Results of Regressions: Passenger Vessels
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 10:52
Sample: 1 6118
Included observations: 6118
Convergence achieved after 9 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
ST_PA1 -3.890047 0.722440 -5.384597 0.0000
ST_PA3 -4.017614 0.932956 -4.306328 0.0000
ST_PA4 -2.581474 0.897105 -2.877560 0.0040
ST_PA5 -5.140974 0.705674 -7.285198 0.0000
ST_PA6 -3.070255 0.922434 -3.328429 0.0009
LNAGE 0.379143 0.130987 2.894505 0.0038
LNSIZE -0.266163 0.058338 -4.562441 0.0000
CL_IBS 2.528660 0.642884 3.933308 0.0001
FL_UA 1.333388 0.522104 2.553873 0.0107
C0100S 0.381253 0.104339 3.653982 0.0003
C0200S 0.520087 0.119328 4.358477 0.0000
C0600S 0.180514 0.053974 3.344458 0.0008
C0700S 0.143179 0.040379 3.545888 0.0004
C0800S 0.424934 0.154150 2.756634 0.0058
C0900S 0.131048 0.041988 3.121065 0.0018
C1700S 0.522460 0.106176 4.920714 0.0000
C2500S 0.341149 0.116281 2.933830 0.0033
PS1_DK 1.312464 0.483531 2.714330 0.0066
PS1_DE 1.136637 0.421763 2.694966 0.0070
PS1_GR 1.812856 0.371036 4.885927 0.0000
PS1_IT 1.466290 0.270606 5.418549 0.0000
PS1_NO 2.022123 0.568614 3.556233 0.0004
PS1_ES 1.437482 0.408665 3.517503 0.0004
PS5_CORP 2.995867 0.871895 3.436041 0.0006
PS5_JACK 2.927490 0.722331 4.052836 0.0001
PS5_PORM 3.029211 0.886602 3.416651 0.0006
Mean dependent var 0.034488 S.D. dependent var 0.182495
S.E. of regression 0.150989 Akaike info criterion 0.184638
Sum squared resid 138.8830 Schwarz criterion 0.213192
Log likelihood -538.8064 Hannan-Quinn criter. 0.194544
Avg. log likelihood -0.088069
Obs with Dep=0 5907 Total obs 6118
Obs with Dep=1 211
238
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 10:52
Sample: 1 6118
Included observations: 6118
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 0.0003 0.0007 611 610.704 0 0.29585 611 0.29599
2 0.0007 0.0010 611 611.493 1 0.50651 612 0.48122
3 0.0010 0.0019 612 611.111 0 0.88919 612 0.89048
4 0.0019 0.0040 609 610.206 3 1.79449 612 0.81222
5 0.0040 0.0063 609 608.845 3 3.15517 612 0.00767
6 0.0063 0.0098 610 606.184 1 4.81648 611 3.04813
7 0.0098 0.0167 605 604.079 7 7.92130 612 0.10856
8 0.0167 0.0307 602 598.008 10 13.9915 612 1.16535
9 0.0307 0.0624 582 585.651 30 26.3486 612 0.52877
10 0.0625 1.0000 456 460.719 156 151.281 612 0.19555
Total 5907 5907.00 211 211.000 6118 7.53394
H-L Statistic: 7.5339 Prob. Chi-Sq(8) 0.4803
Andrews Statistic: 1194.704 Prob. Chi-Sq(10) 0.0000
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 10:52
Sample: 1 6118
Included observations: 6118
Prediction Assessment (success cutoff C = 0.0345)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 4994 28 5022 5907 211 6118
P(Dep=1)>C 913 183 1096 0 0 0
Total 5907 211 6118 5907 211 6118
Correct 4994 183 5177 5907 0 5907
% Correct 84.54 86.73 84.62 100.00 0.00 96.55
% Incorrect 15.46 13.27 15.38 0.00 100.00 3.45
Total Gain* -15.46 86.73 -11.93
Percent Gain** NA 86.73 -345.97
239
240
Appendix 12: Step 1: Results of Regressions: Other Ship Types
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 11:11
Sample: 1 10077 IF OUTLIER=0
Included observations: 10073
Convergence achieved after 8 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
ST_OT1 -6.883984 0.422584 -16.29022 0.0000
ST_OT3 -8.466110 0.593436 -14.26626 0.0000
ST_OT4 -4.405082 0.465133 -9.470582 0.0000
ST_OT5 -7.184651 0.392313 -18.31356 0.0000
ST_OT6 -6.395696 0.550852 -11.61056 0.0000
LNAGE 0.609421 0.124318 4.902099 0.0000
CL_GL -0.765959 0.365430 -2.096048 0.0361
CL_NCL 1.231356 0.236801 5.199963 0.0000
FL_AG 1.177324 0.374456 3.144092 0.0017
FL_CY 0.684888 0.306411 2.235189 0.0254
FL_VC 0.940171 0.244227 3.849573 0.0001
FL_UA 1.709392 0.445812 3.834337 0.0001
C0100S 0.536713 0.073369 7.315290 0.0000
C0200S 0.748906 0.119644 6.259473 0.0000
C0400S 0.465803 0.234115 1.989633 0.0466
C0600S 0.189598 0.053331 3.555099 0.0004
C0700S 0.469129 0.074097 6.331291 0.0000
C1100S 0.461109 0.139448 3.306664 0.0009
C1400S 0.418814 0.058910 7.109414 0.0000
C1700S 0.427365 0.078831 5.421298 0.0000
C2500S 0.752588 0.131418 5.726669 0.0000
PS1_IT 1.739326 0.350807 4.958067 0.0000
PS3_CHI 2.947174 0.572770 5.145474 0.0000
PS3_MX 3.194675 0.842992 3.789684 0.0002
PS3_PA 3.503819 0.656842 5.334341 0.0000
PS5_HOUS 0.905284 0.349862 2.587546 0.0097
PS5_JACK 2.065930 0.326899 6.319788 0.0000
PS5_MOBI 0.837545 0.344516 2.431076 0.0151
PS5_PHIL 1.189509 0.397842 2.989907 0.0028
PS5_SANF 1.230329 0.430642 2.856968 0.0043
PS5_WILM 1.546897 0.581667 2.659419 0.0078
PS5_OTH 1.290303 0.374795 3.442691 0.0006
Mean dependent var 0.037129 S.D. dependent var 0.189087
S.E. of regression 0.156133 Akaike info criterion 0.190012
Sum squared resid 244.7751 Schwarz criterion 0.212941
Log likelihood -924.9935 Hannan-Quinn criter. 0.197770
Avg. log likelihood -0.091829
241
Obs with Dep=0 9699 Total obs 10073
Obs with Dep=1 374
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 11:11
Sample: 1 10077 IF OUTLIER=0
Included observations: 10073
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 0.0001 0.0022 1007 1005.61 0 1.38771 1007 1.38963
2 0.0022 0.0038 1005 1004.07 2 2.93105 1007 0.29661
3 0.0038 0.0047 1006 1002.67 1 4.32559 1007 2.56780
4 0.0047 0.0056 1006 1002.83 2 5.17263 1008 1.95597
5 0.0056 0.0075 1005 1000.59 2 6.40621 1007 3.05001
6 0.0075 0.0108 1001 997.848 6 9.15246 1007 1.09579
7 0.0108 0.0156 1000 994.937 8 13.0632 1008 1.98825
8 0.0156 0.0255 988 987.017 19 19.9826 1007 0.04929
9 0.0255 0.0601 958 969.041 49 37.9586 1007 3.33752
10 0.0601 1.0000 723 734.380 285 273.620 1008 0.64965
Total 9699 9699.00 374 374.000 10073 16.3805
H-L Statistic: 16.3805 Prob. Chi-Sq(8) 0.0372
Andrews Statistic: 948.3127 Prob. Chi-Sq(10) 0.0000
Dependent Variable: DETAINED
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/12/06 Time: 11:11
Sample: 1 10077 IF OUTLIER=0
Included observations: 10073
Prediction Assessment (success cutoff C = 0.0372)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 8555 61 8616 9699 374 10073
P(Dep=1)>C 1144 313 1457 0 0 0
Total 9699 374 10073 9699 374 10073
Correct 8555 313 8868 9699 0 9699
% Correct 88.20 83.69 88.04 100.00 0.00 96.29
% Incorrect 11.80 16.31 11.96 0.00 100.00 3.71
Total Gain* -11.80 83.69 -8.25
Percent Gain** NA 83.69 -222.19
242
243
Appendix 13: Step 1: Results of Regressions: Caribbean MoU
Classification Table(c)
Observed Predicted
Selected Cases(a) Unselected Cases(b)
detained_new
Percentag
e Correct detained_new
Percentage
Correct
0 1 0 1
Step 1 detained_new 0 450 23 95.1 191 8 96.0
1 2 20 90.9 6 8 57.1
Overall Percentage 94.9 93.4
a Selected cases validate EQ 1, b Unselected cases validate NE 1, c The cut value is .050
Contingency Table for Hosmer and Lemeshow Test
detained_new = 0 detained_new = 1
Observed Expected Observed Expected Total
1 206 205.957 0 .043 206
2 65 64.960 0 .040 65
3 68 67.832 0 .168 68
4 9 9.931 1 .069 10
5 96 95.582 1 1.418 97
Step 1
6 29 28.738 20 20.262 49
Model Summary
Step
-2 Log
likelihood
Cox & Snell
R Square
Nagelkerke R
Square
1 56.271(a) .221 .725
a Estimation terminated at iteration number 9 because parameter estimates changed by less than .001.
Omnibus Tests of Model Coefficients
Chi-square df Sig.
Step 123.731 8 .000
Block 123.731 8 .000
Step 1
Model 123.731 8 .000
Variables in the Equation
B S.E. Wald Sig. Exp(B)
Step CL_GermanischerLloyd 2.537 1.333 3.621 .057 12.647
Code_0100 1.131 .417 7.373 .007 3.100
Code_0200 2.176 .752 8.386 .004 8.815
Code_1200 1.497 .468 10.240 .001 4.467
Code_1400 1.477 .575 6.594 .010 4.381
Code_1500 3.215 .839 14.695 .000 24.910
OW_TraditionalMN -4.270 1.360 9.860 .002 .014
OW_EmergingMN -3.194 1.537 4.320 .038 .041
Constant -4.210 .690 37.246 .000 .015
244
Appendix 14: Step 3: Probability of Detention: General Cargo
General Cargo Ship – Paris MoU
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery(3) Navigation & Comm (7)
Ship & Cargo Operations (4) Management (8)
Age: 13 yrs
Tonnage: 5965 gt
Flag: Panama
Class: GL
Port State: Italy
Owner: Unknown
2
3
1 4
5
6
7
8
General Cargo Ship – Vina MoU
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 13 yrs
Tonnage: 5965 gt
Flag: Panama
Class: GL
Port State: Cuba
Owner: Unknown
2 1
3
4
5
6
7
8
245
General Cargo Ship – USCG
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 13 yrs
Tonnage: 5965 gt
Flag: Panama
Class: GL
Port State: New Orleans
Owner: Unknown
1
2
3
4
56
7
8
General Cargo Ship – AMSA
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 13 yrs
Tonnage: 5965 gt
Flag: Panama
Class: GL
Port State: Brisbane
Owner: Unknown
1
2
3
4
5
6
8 7
246
Appendix 15: Step 3: Probability of Detention: Dry Bulk
Dry Bulk – Paris MoU
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 13 yrs
Tonnage: 38995 gt
Flag: Malta
Class: GL
Port State: France
Owner: EMN
1
2
3
4
5
6
7
8
Dry Bulk – Vina MoU
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 13 yrs
Tonnage: 38995 gt
Flag: Malta
Class: GL
Port State: Brazil
Owner: EMN
8
1
2
3
4
7
247
Dry Bulk – Indian Ocean MoU
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 13 yrs
Tonnage: 38995 gt
Flag: Malta
Class: GL
Port State: South Africa
Owner: EMN
2 1
3
4
5
6
7
8
Dry Bulk – USCG
-0.20
0.00
0.20
0.40
0.60
0.80
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 13 yrs
Tonnage: 38995 gt
Flag: Malta
Class: GL
Port State: Miami
Owner: EMN
1
2
4
248
Appendix 16: Step 3: Probability of Detention: Tanker
Tanker – Vina MoU
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 10 yrs
Tonnage: 28909 gt
Flag: Panama
Class: GL
Port State: Chile
Owner: TMN
1
2
3
4
5
6
7
8
Tanker – Indian Ocean MoU
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention Certificates (
1) Working Conditions (
5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 10 yrs
Tonnage: 28909 gt
Flag: Panama
Class: GL
Port State: South Africa
Owner: TMN
1
2, 4, 7
5
6
3
8
249
Tanker – USCG
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 10 yrs
Tonnage: 28909 gt
Flag: Panama
Class: GL
Port State: Houston
Owner: TMN
1
2
3
4 8
7
6
Tanker – AMSA
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 10 yrs
Tonnage: 28909 gt
Flag: Panama
Class: GL
Port State: Brisbane
Owner: TMN
1
2
6,3,5
8
4
7
250
Appendix 17: Step 3: Probability of Detention: Container
Container – Paris MoU
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 8 yrs
Tonnage: 27322 gt
Flag: Panama
Class: GL
Port State: Germany
Owner: TMN
1
2
3
4
5
6
7
8
Container – Vina MoU
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 8 yrs
Tonnage: 27322 gt
Flag: Panama
Class: GL
Port State: Chile
Owner: TMN
1
2
3
4
6
7
8
251
Container – Indian Ocean MoU
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 8 yrs
Tonnage: 27322 gt
Flag: Panama
Class: GL
Port State: Iran
Owner: TMN
1
2
4
6
8
Container – AMSA
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
0 1 2 3 4 5 6 7 8 9 10
Number of Deficiencies
Probability of Detention
Certificates (1) Working Conditions (5)
Safety & Fire (2) Stability & Structure (6)
Equipment & Machinery (3) Navigation & Commun (7)
Ship & Cargo Operations (4) Management (8)
Age: 8 yrs
Tonnage: 27322 gt
Flag: Panama
Class: GL
Port State: Port Botany
Owner: TMN
1
2
3
4
5
8 6
252
Appendix 18: LM Test for Very Serious Casualties
Tonnage
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 05/11/06 Time: 23:15
Sample: 1 47169 IF EXCESSOB=0 AND ST7_MAX=0 AND
OUTLIER=0 AND CMOU_S=0
Included observations: 38076
Variable Coefficient Std. Error t-Statistic Prob.
FAC -1.361108 0.545723 -2.494137 0.0126
ST1_MAX*FAC 0.778387 0.194644 3.999026 0.0001
LNTON*FAC -0.449967 0.100681 -4.469252 0.0000
LNAGE*FAC 0.256059 0.089978 2.845795 0.0044
CLWD_D*FAC 0.472311 0.203908 2.316295 0.0205
CLCHGD*FAC -0.478080 0.241530 -1.979379 0.0478
OWCHGD*FAC 3.753590 0.761337 4.930262 0.0000
CL_RNR*FAC 1.967802 0.791299 2.486800 0.0129
FL_AG*FAC 1.028536 0.469510 2.190659 0.0285
FL_BG*FAC 1.290962 0.783771 1.647117 0.0995
FL_BZ*FAC 0.829271 0.353460 2.346155 0.0190
FL_TR*FAC 0.504822 0.305642 1.651681 0.0986
OW_EMN*FAC 0.806087 0.238956 3.373377 0.0007
OW_TMN*FAC 1.111524 0.273993 4.056759 0.0000
OW_IOR*FAC 0.931463 0.320292 2.908166 0.0036
OW_OOR*FAC 1.876627 0.446509 4.202888 0.0000
SY_KR*FAC 0.631551 0.420017 1.503630 0.1327
SY_NO*FAC 0.735646 0.375644 1.958358 0.0502
SY_UN*FAC 2.570107 0.530377 4.845812 0.0000
LI_FLRET*FAC -0.033462 0.013931 -2.401985 0.0163
RS_INS*FAC -0.571063 0.182827 -3.123517 0.0018
AMSA_S*FAC -1.050391 0.266198 -3.945904 0.0001
IMOU_S*FAC -1.305526 0.342302 -3.813958 0.0001
PMOU_S*FAC -0.382504 0.085432 -4.477263 0.0000
USCG_S*FAC -1.025581 0.251547 -4.077092 0.0000
VMOU_S*FAC -0.346312 0.117173 -2.955573 0.0031
LNTON*(-XB)*FAC 0.085733 0.016986 5.047259 0.0000
R-squared 0.000645 Mean dependent var 0.004893
Adjusted R-squared -0.000039 S.D. dependent var 0.950329
S.E. of regression 0.950348 Akaike info criterion 2.736731
Sum squared resid 34349.90 Schwarz criterion 2.742794
Log likelihood -52053.00 Durbin-Watson stat 1.871794
253
Age
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 05/11/06 Time: 23:21
Sample: 1 47169 IF EXCESSOB=0 AND ST7_MAX=0 AND
OUTLIER=0 AND CMOU_S=0
Included observations: 38076
Variable Coefficient Std. Error t-Statistic Prob.
FAC 0.809542 0.520551 1.555162 0.1199
ST1_MAX*FAC -0.496444 0.174625 -2.842918 0.0045
LNTON*FAC 0.147913 0.059624 2.480777 0.0131
LNAGE*FAC 0.165110 0.085700 1.926591 0.0540
CLWD_D*FAC -0.205618 0.189997 -1.082216 0.2792
CLCHGD*FAC 0.265909 0.234107 1.135847 0.2560
OWCHGD*FAC -2.129391 0.577153 -3.689475 0.0002
CL_RNR*FAC -1.113987 0.748194 -1.488901 0.1365
FL_AG*FAC -0.670327 0.452134 -1.482585 0.1382
FL_BG*FAC -0.668855 0.763096 -0.876502 0.3808
FL_BZ*FAC -0.539366 0.342280 -1.575802 0.1151
FL_TR*FAC -0.324981 0.301973 -1.076191 0.2818
OW_EMN*FAC -0.489567 0.220287 -2.222410 0.0263
OW_TMN*FAC -0.710166 0.249947 -2.841267 0.0045
OW_IOR*FAC -0.578825 0.303549 -1.906856 0.0565
OW_OOR*FAC -1.083118 0.382319 -2.833023 0.0046
SY_KR*FAC -0.614768 0.430522 -1.427961 0.1533
SY_NO*FAC -0.476987 0.365988 -1.303285 0.1925
SY_UN*FAC -1.630392 0.440520 -3.701058 0.0002
LI_FLRET*FAC 0.020322 0.013307 1.527239 0.1267
RS_INS*FAC 0.349586 0.172224 2.029839 0.0424
AMSA_S*FAC 0.619145 0.231486 2.674661 0.0075
IMOU_S*FAC 0.908718 0.312149 2.911165 0.0036
PMOU_S*FAC 0.222707 0.069441 3.207120 0.0013
USCG_S*FAC 0.586517 0.209840 2.795071 0.0052
VMOU_S*FAC 0.206786 0.107447 1.924538 0.0543
LNAGE*(-XB)*FAC -0.130285 0.033647 -3.872119 0.0001
R-squared 0.000369 Mean dependent var 0.004893
Adjusted R-squared -0.000314 S.D. dependent var 0.950329
S.E. of regression 0.950478 Akaike info criterion 2.737007
Sum squared resid 34359.37 Schwarz criterion 2.743070
Log likelihood -52058.24 Durbin-Watson stat 1.871407
254
Appendix 19: Estimation Results of Greene Model, Very Serious Casualties
Age and Tonnage
LogL: LL1
Method: Maximum Likelihood (Marquardt)
Date: 05/13/06 Time: 00:31
Sample: 1 47169 IF EXCESSOB=0 AND ST7_MAX=0 AND
OUTLIER=0 AND CMOU_S=0
Included observations: 38076
Assessment order: By observation
Estimation settings: tol= 1.0e-05, derivs=accurate numeric
Initial Values: BETA(1)=0.41130, BETA(2)=0.52333, BETA(3)=0.51187,
BETA(4)=0.69811, BETA(5)=0.07680, BETA(6)=0.74075,
BETA(7)=0.95502, BETA(8)=0.81404, BETA(9)=0.07958,
BETA(10)=0.85638, BETA(11)=0.62122, BETA(12)=0.41764,
BETA(13)=0.57133, BETA(14)=0.99269, BETA(15)=0.69978,
BETA(16)=0.10224, BETA(17)=0.85978, BETA(18)=0.55634,
BETA(19)=0.30227, BETA(20)=0.20892, BETA(21)=0.42523,
BETA(22)=0.59285, BETA(23)=0.29655, BETA(24)=0.68481,
BETA(25)=0.93063, BETA(26)=0.08356, GAM(1)=0.00349,
GAM(2)=0.28542
Convergence achieved after 189 iterations
Coefficient Std. Error z-Statistic Prob.
BETA(1) -5.014612 0.721928 -6.946135 0.0000
BETA(2) 1.955363 0.405091 4.826969 0.0000
BETA(3) -0.935205 0.230269 -4.061357 0.0000
BETA(4) 0.754552 0.164262 4.593579 0.0000
BETA(5) 1.146784 0.403194 2.844252 0.0045
BETA(6) -1.009336 0.425415 -2.372592 0.0177
BETA(7) 9.008728 1.609522 5.597144 0.0000
BETA(8) 4.700616 1.273900 3.689941 0.0002
BETA(9) 2.650216 0.778979 3.402168 0.0007
BETA(10) 2.955407 1.713680 1.724596 0.0846
BETA(11) 2.143534 0.619450 3.460385 0.0005
BETA(12) 1.086505 0.544248 1.996342 0.0459
BETA(13) 1.969195 0.456945 4.309481 0.0000
BETA(14) 2.651248 0.526253 5.037971 0.0000
BETA(15) 2.290799 0.612264 3.741522 0.0002
BETA(16) 4.190399 0.831645 5.038686 0.0000
BETA(17) 1.432996 0.765205 1.872697 0.0611
BETA(18) 1.908346 0.708711 2.692699 0.0071
BETA(19) 6.501460 1.107278 5.871568 0.0000
BETA(20) -0.084136 0.024775 -3.396006 0.0007
BETA(21) -1.424921 0.370641 -3.844483 0.0001
BETA(22) -2.607840 0.529732 -4.922940 0.0000
BETA(23) -3.914086 0.788980 -4.960946 0.0000
BETA(24) -0.949919 0.182783 -5.196977 0.0000
BETA(25) -2.456774 0.467197 -5.258536 0.0000
255
BETA(26) -0.942375 0.233599 -4.034150 0.0001
GAM(1) 0.080157 0.017731 4.520697 0.0000
GAM(2) -0.048332 0.035069 -1.378193 0.1681
Log likelihood -1293.243 Akaike info criterion 0.069400
Avg. log likelihood -0.033965 Schwarz criterion 0.075686
Number of Coefs. 28 Hannan-Quinn criter. 0.071394
Tonnage Only
LogL: LL1
Method: Maximum Likelihood (Marquardt)
Date: 05/12/06 Time: 00:59
Sample: 1 47169 IF EXCESSOB=0 AND ST7_MAX=0 AND
OUTLIER=0 AND CMOU_S=0
Included observations: 38076
Assessment order: By observation
Estimation settings: tol= 1.0e-05, derivs=accurate numeric
Initial Values: BETA(1)=0.66868, BETA(2)=0.64867, BETA(3)=0.64281,
BETA(4)=0.82909, BETA(5)=0.29897, BETA(6)=0.51628,
BETA(7)=0.78892, BETA(8)=0.28128, BETA(9)=0.86711,
BETA(10)=0.79424, BETA(11)=0.30990, BETA(12)=0.46415,
BETA(13)=0.33391, BETA(14)=0.02463, BETA(15)=0.79715,
BETA(16)=0.20055, BETA(17)=0.69548, BETA(18)=0.92699,
BETA(19)=0.97617, BETA(20)=0.55996, BETA(21)=0.92988,
BETA(22)=0.46306, BETA(23)=0.31914, BETA(24)=0.72321,
BETA(25)=0.27320, BETA(26)=0.75714, GAM(1)=0.31511
Convergence achieved after 112 iterations
Coefficient Std. Error z-Statistic Prob.
BETA(1) -4.968107 0.846124 -5.871607 0.0000
BETA(2) 2.449522 0.439826 5.569297 0.0000
BETA(3) -1.240318 0.267346 -4.639370 0.0000
BETA(4) 0.686511 0.165193 4.155816 0.0000
BETA(5) 1.470035 0.481397 3.053688 0.0023
BETA(6) -1.243165 0.516467 -2.407057 0.0161
BETA(7) 11.41535 1.629473 7.005550 0.0000
BETA(8) 5.983110 1.499014 3.991364 0.0001
BETA(9) 3.300294 0.893889 3.692063 0.0002
BETA(10) 3.704389 2.184891 1.695457 0.0900
BETA(11) 2.667721 0.735841 3.625403 0.0003
BETA(12) 1.364804 0.667685 2.044083 0.0409
BETA(13) 2.405654 0.505676 4.757300 0.0000
BETA(14) 3.243644 0.555757 5.836446 0.0000
BETA(15) 2.858932 0.676138 4.228328 0.0000
BETA(16) 5.240314 0.856515 6.118183 0.0000
BETA(17) 1.902673 0.898725 2.117080 0.0343
BETA(18) 2.373656 0.865118 2.743737 0.0061
BETA(19) 8.186691 1.094707 7.478430 0.0000
256
BETA(20) -0.104756 0.028347 -3.695439 0.0002
BETA(21) -1.780047 0.412698 -4.313193 0.0000
BETA(22) -3.344526 0.564716 -5.922494 0.0000
BETA(23) -5.086684 0.833368 -6.103765 0.0000
BETA(24) -1.222197 0.190369 -6.420162 0.0000
BETA(25) -3.122190 0.483228 -6.461111 0.0000
BETA(26) -1.235696 0.269779 -4.580395 0.0000
GAM(1) 0.091619 0.017580 5.211461 0.0000
Log likelihood -1294.314 Akaike info criterion 0.069404
Avg. log likelihood -0.033993 Schwarz criterion 0.075465
Number of Coefs. 27 Hannan-Quinn criter. 0.071327
Greene Program Used
‘ HPROBIT1.PRG (1.0 – 6/29/98)
‘ example program for EViews LogL object
‘ revised for version 5.0 (3/26/2004)
‘ Estimate probit specification with multiplicative heterogeneity.
” Example 19.7 (p. 890) of Greene, William H. (1997) Econometric Analysis, ‘ 3rd edition, Prentice-
Hall.
coef(26) beta
rnd (beta)
coef(1) gam
rnd (gam)
‘ specify the likelihood
logl ll1
ll1.append @logl logl1
ll1.append index =
(beta(1)+beta(2)*st1_max+beta(3)*lnton+beta(4)*lnage+beta(5)*clwd_d+beta(6)*clchgd+beta(7)*ow
chgd+beta(8)*cl_rnr+beta(9)*fl_ag+beta(10)*fl_bg+beta(11)*fl_bz+beta(12)*fl_tr+beta(13)*ow_emn
+beta(14)*ow_tmn+beta(15)*ow_ior+beta(16)*ow_oor+beta(17)*sy_kr+beta(18)*sy_no+beta(19)*sy
_un+beta(20)*li_flret+beta(21)*rs_ins+beta(22)*amsa_s+beta(23)*imou_s+beta(24)*pmou_s+beta(2
5)*uscg_s+beta(26)*vmou_s)/exp(gam(1)*lnton)
ll1.append logl1 = casualty*log(@clogistic(index))+(1-casualty)*log(1-@clogistic(index))
‘ carry out MLE and display results
ll1.ml(showopts, m=1000, c=1e-5)
show ll1.output
Note: for the model with age and tonnage, the last term of the index is changed to the
following term: exp(gam(1)*lnton) + gam(2)*lnage and coef (1) is changed to coef (2) for
gam
257
Appendix 20: Average Probabilities of Corrected versus Uncorrected Model
Tonnage Groups
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
34
36
38
40
42
Tonnage Groups
Probability of VS Casualty
Corrected
Uncorrected
For an explanation on the groups in detail, please refer to 7.1.2. Methodology Used to Match Ships
258
Age Groups
0 0.02 0.04 0.06 0.08 0.1
1
2
3
4
5
6
7
8
9
10
11
Age Groups
Probability of VS Casualty
Corrected
Uncorrected
For an explanation on the groups in detail, please refer to 7.1.2. Methodology Used to Match Ships
Flag States Groups
0.00
0.01
0.01
0.02
0.02
0.03
0.03
0.04
Black Grey White Undefined
Probability of VS Casualty
Corrected
Uncorrected
For an explanation on the groups in detail, please refer to 3.2.1. Basic Port State and Casualty
Variables
259
Classification Society Groups
0.00
0.01
0.01
0.02
0.02
0.03
0.03
0.04
NIACS IACS Unknown
Probability of VS Casualty
Corrected Uncorrected
For an explanation on the groups in detail, please refer to 3.2.1. Basic Port State and Casualty
Variables
Ownership Groups
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
EMN TMN OOR NOR IOR Unknown
Probability of VS Casualty
Corrected
Uncorrected
For an explanation on the groups in detail, please refer to 3.2.1. Basic Port State and Casualty
Variables
260
Appendix 21: Results of Regression: Very Serious Casualties
Dependent Variable: CASUALTY
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/08/06 Time: 22:24
Sample: 1 47169 IF EXCESSOB=0 AND ST7_MAX=0 AND
OUTLIER=0 AND CMOU_S=0
Included observations: 38076
Convergence achieved after 9 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
C -4.589385 0.625279 -7.339740 0.0000
ST1_MAX 1.250761 0.128888 9.704245 0.0000
LNTON -0.371789 0.054477 -6.824711 0.0000
LNAGE 0.405964 0.088256 4.599860 0.0000
CLWD_D 0.580286 0.198103 2.929219 0.0034
CLCHGD -0.696574 0.245789 -2.834030 0.0046
OWCHGD 5.368643 0.189853 28.27788 0.0000
CL_RNR 2.774375 0.881943 3.145753 0.0017
FL_AG 1.685907 0.531297 3.173193 0.0015
FL_BG 1.754525 0.700529 2.504570 0.0123
FL_BZ 1.353529 0.335335 4.036353 0.0001
FL_TR 0.753613 0.289022 2.607455 0.0091
OW_EMN 1.309416 0.197466 6.631090 0.0000
OW_TMN 1.796132 0.197710 9.084668 0.0000
OW_IOR 1.510983 0.286985 5.265027 0.0000
OW_OOR 2.794337 0.323119 8.648001 0.0000
SY_KR 1.227254 0.440791 2.784209 0.0054
SY_NO 1.215751 0.331591 3.666423 0.0002
SY_UN 4.070845 0.169600 24.00267 0.0000
LI_FLRET -0.054392 0.013386 -4.063246 0.0000
RS_INS -0.945467 0.140214 -6.743014 0.0000
AMSA_S -1.520106 0.263189 -5.775720 0.0000
IMOU_S -2.176095 0.343539 -6.334354 0.0000
PMOU_S -0.544375 0.061421 -8.863061 0.0000
USCG_S -1.468562 0.278380 -5.275379 0.0000
VMOU_S -0.493493 0.125032 -3.946943 0.0001
Mean dependent var 0.018962 S.D. dependent var 0.136393
S.E. of regression 0.090137 Akaike info criterion 0.070020
Sum squared resid 309.1437 Schwarz criterion 0.075856
Log likelihood -1307.036 Hannan-Quinn criter. 0.071871
Restr. log likelihood -3578.068 Avg. log likelihood -0.034327
LR statistic (25 df) 4542.064 McFadden R-squared 0.634709
Probability(LR stat) 0.000000
Obs with Dep=0 37354 Total obs 38076
Obs with Dep=1 722
261
Dependent Variable: CASUALTY
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/08/06 Time: 22:24
Sample: 1 47169 IF EXCESSOB=0 AND ST7_MAX=0 AND OUTLIER=0 AND
CMOU_S=0
Included observations: 38076
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 0.0000 6.E-08 3807 3807.00 0 3.7E-05 3807 3.7E-05
2 6.E-08 5.E-06 3808 3808.00 0 0.00477 3808 0.00477
3 5.E-06 7.E-05 3807 3806.90 0 0.09870 3807 0.09870
4 7.E-05 0.0004 3806 3807.21 2 0.79211 3808 1.84230
5 0.0004 0.0010 3801 3805.32 7 2.68493 3808 6.93985
6 0.0010 0.0018 3800 3801.68 7 5.32050 3807 0.53091
7 0.0018 0.0032 3793 3798.67 15 9.33058 3808 3.45329
8 0.0032 0.0060 3792 3790.28 15 16.7211 3807 0.17793
9 0.0060 0.0137 3782 3774.80 26 33.2042 3808 1.57682
10 0.0137 0.9998 3158 3154.16 650 653.843 3808 0.02727
Total 37354 37354.0 722 722.000 38076 14.6519
H-L Statistic: 14.6519 Prob. Chi-Sq(8) 0.0663
Andrews Statistic: 5583.790 Prob. Chi-Sq(10) 0.0000
Dependent Variable: CASUALTY
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/08/06 Time: 22:24
Sample: 1 47169 IF EXCESSOB=0 AND ST7_MAX=0 AND
OUTLIER=0 AND CMOU_S=0
Included observations: 38076
Prediction Assessment (success cutoff C = 0.0189)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 34936 82 35018 0 0 0
P(Dep=1)>C 2418 640 3058 37354 722 38076
Total 37354 722 38076 37354 722 38076
Correct 34936 640 35576 0 722 722
% Correct 93.53 88.64 93.43 0.00 100.00 1.90
% Incorrect 6.47 11.36 6.57 100.00 0.00 98.10
Total Gain* 93.53 -11.36 91.54
Percent Gain** 93.53 NA 93.31
262
-10
0
10
20
30
40
50
60
10000 20000 30000 40000
Standardized (Pearson) Residuals
263
Appendix 22: Results of Regression: Serious Casualties
Dependent Variable: CASUALTY
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/27/06 Time: 12:30
Sample: 1 49854 IF EXCESSOB=0 AND ST7_MAX=0 AND
OUTLIER=0 AND CAS_FERRIES=0
Included observations: 41009
Convergence achieved after 7 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
C -3.092465 0.210296 -14.70530 0.0000
ST1_MAX 0.668466 0.054811 12.19580 0.0000
ST2_MAX 0.407800 0.073937 5.515484 0.0000
ST5_MAX 0.591748 0.100596 5.882422 0.0000
CLWD_D 0.670376 0.060123 11.15006 0.0000
CLCHGD -2.156467 0.100328 -21.49417 0.0000
OWCHGD 2.426333 0.063340 38.30655 0.0000
CL_ABS 0.220091 0.096757 2.274670 0.0229
CL_BV 0.283156 0.090137 3.141377 0.0017
CL_CRR 1.745061 0.395830 4.408614 0.0000
CL_DNV 0.239336 0.095810 2.498030 0.0125
CL_GL 0.315691 0.099929 3.159160 0.0016
CL_HEL 1.126282 0.191428 5.883587 0.0000
CL_LR 0.238559 0.076532 3.117114 0.0018
CL_PRS 0.989989 0.246507 4.016066 0.0001
FL_AG 1.677071 0.172287 9.734189 0.0000
FL_AN 3.791955 0.499641 7.589366 0.0000
FL_AU 4.896389 0.496010 9.871545 0.0000
FL_BB 3.698497 0.500170 7.394481 0.0000
FL_BG 2.615955 0.445411 5.873127 0.0000
FL_BS 1.288386 0.168382 7.651552 0.0000
FL_BZ 1.282967 0.245786 5.219847 0.0000
FL_CA 5.667824 0.367547 15.42067 0.0000
FL_CH 4.017011 0.772602 5.199329 0.0000
FL_CI -0.528456 0.052641 -10.03888 0.0000
FL_CL 3.449410 0.452342 7.625661 0.0000
FL_CN 0.926625 0.216719 4.275700 0.0000
FL_CY 1.795461 0.209121 8.585771 0.0000
FL_DE 4.757727 0.357369 13.31322 0.0000
FL_DK 3.158602 0.358656 8.806777 0.0000
FL_EE 1.865066 0.615691 3.029223 0.0025
FL_EG 2.443472 0.312348 7.822928 0.0000
FL_ES 3.567010 0.319016 11.18128 0.0000
FL_FI 4.431572 0.490034 9.043392 0.0000
FL_FR 3.305033 0.434322 7.609643 0.0000
FL_FO 5.123701 0.739233 6.931106 0.0000
FL_GI 2.277671 0.488577 4.661850 0.0000
264
FL_GR 2.731950 0.191291 14.28165 0.0000
FL_HK 2.556422 0.385528 6.630958 0.0000
FL_IE 4.447324 0.450956 9.861982 0.0000
FL_IM 4.530595 0.429767 10.54198 0.0000
FL_IN 2.327092 0.250481 9.290498 0.0000
FL_IT 2.416283 0.237086 10.19159 0.0000
FL_JP 3.117043 0.282379 11.03850 0.0000
FL_KH 1.320022 0.204835 6.444315 0.0000
FL_KR 2.678093 0.222986 12.01015 0.0000
FL_KW 2.518723 0.538670 4.675820 0.0000
FL_LR 0.773762 0.164247 4.710973 0.0000
FL_MH 2.317475 0.404445 5.730006 0.0000
FL_MT 4.398416 0.319629 13.76100 0.0000
FL_MY 1.576575 0.320774 4.914909 0.0000
FL_NIS 4.540584 0.401066 11.32129 0.0000
FL_NL 4.063273 0.356062 11.41170 0.0000
FL_NO 5.137925 0.383828 13.38600 0.0000
FL_NZ 5.063607 0.692862 7.308243 0.0000
FL_PA 0.888314 0.114369 7.767070 0.0000
FL_PL 2.155938 0.429161 5.023608 0.0000
FL_PT 2.495289 0.373013 6.689548 0.0000
FL_SE 4.451411 0.424953 10.47507 0.0000
FL_SG 3.946599 0.406321 9.713013 0.0000
FL_TO 1.792702 0.596090 3.007434 0.0026
FL_UK 5.129702 0.356429 14.39192 0.0000
FL_US 4.360653 0.325468 13.39810 0.0000
FL_VC 1.008740 0.160391 6.289266 0.0000
OW_EMN 0.998902 0.127474 7.836136 0.0000
OW_IOR 1.707517 0.161042 10.60291 0.0000
OW_OOR 2.596296 0.162111 16.01559 0.0000
OW_TMN 1.968789 0.160918 12.23472 0.0000
LI_FLRET -0.124485 0.010146 -12.26910 0.0000
LI_OWRET -0.036024 0.008848 -4.071546 0.0000
SY_AR 1.511256 0.460342 3.282900 0.0010
SY_BG 1.387078 0.247395 5.606738 0.0000
SY_BR 1.829303 0.250725 7.296056 0.0000
SY_CA 0.911227 0.230245 3.957638 0.0001
SY_CN 0.701901 0.153577 4.570359 0.0000
SY_DE 0.948511 0.093630 10.13044 0.0000
SY_DK 1.023127 0.168166 6.084046 0.0000
SY_ES 0.859322 0.162852 5.276711 0.0000
SY_FI 1.025017 0.181672 5.642145 0.0000
SY_FR 1.454359 0.207273 7.016644 0.0000
SY_GR 1.793982 0.278736 6.436137 0.0000
SY_HR 1.470937 0.201326 7.306258 0.0000
SY_IT 1.081655 0.195505 5.532619 0.0000
SY_JP 0.574232 0.077664 7.393845 0.0000
SY_KR 0.993873 0.109156 9.105032 0.0000
SY_NL 0.927623 0.116488 7.963272 0.0000
SY_NO 1.488261 0.118315 12.57879 0.0000
SY_OT 1.838009 0.183975 9.990551 0.0000
265
SY_PL 0.764665 0.178705 4.278919 0.0000
SY_RO 1.277681 0.160963 7.937723 0.0000
SY_RU 0.796697 0.168806 4.719605 0.0000
SY_SE 1.380393 0.200647 6.879717 0.0000
SY_TR 1.610577 0.186590 8.631655 0.0000
SY_UK 1.195786 0.137620 8.689071 0.0000
SY_US 0.886520 0.196212 4.518162 0.0000
AMSA_S -0.172419 0.025038 -6.886212 0.0000
PMOU_S 0.045986 0.007260 6.334319 0.0000
USCG_S 0.029542 0.009051 3.264050 0.0011
Mean dependent var 0.077983 S.D. dependent var 0.268148
S.E. of regression 0.231704 Akaike info criterion 0.398456
Sum squared resid 2196.387 Schwarz criterion 0.419059
Log likelihood -8072.138 Hannan-Quinn criter. 0.404970
Restr. log likelihood -11228.88 Avg. log likelihood -0.196838
LR statistic (97 df) 6313.485 McFadden R-squared 0.281127
Probability(LR stat) 0.000000
Obs with Dep=0 37811 Total obs 41009
Obs with Dep=1 3198
Dependent Variable: CASUALTY
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/27/06 Time: 12:30
Sample: 1 49854 IF EXCESSOB=0 AND ST7_MAX=0 AND OUTLIER=0 AND
CAS_FERRIES=0
Included observations: 41009
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 7.E-05 0.0067 4079 4083.85 21 16.1519 4100 1.46097
2 0.0067 0.0109 4061 4063.68 40 37.3179 4101 0.19454
3 0.0109 0.0160 4031 4046.42 70 54.5770 4101 4.41720
4 0.0160 0.0215 4034 4024.63 67 76.3722 4101 1.17196
5 0.0215 0.0291 3974 3997.56 127 103.443 4101 5.50328
6 0.0291 0.0426 3949 3955.05 152 145.950 4101 0.26008
7 0.0426 0.0594 3915 3896.16 186 204.842 4101 1.82419
8 0.0594 0.0999 3777 3785.32 324 315.677 4101 0.23772
9 0.0999 0.1917 3560 3529.82 541 571.183 4101 1.85300
10 0.1918 0.9986 2431 2428.51 1670 1672.49 4101 0.00624
Total 37811 37811.0 3198 3198.00 41009 16.9292
H-L Statistic: 16.9292 Prob. Chi-Sq(8) 0.0309
Andrews Statistic: 33.2464 Prob. Chi-Sq(10) 0.0002
266
Dependent Variable: CASUALTY
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/27/06 Time: 12:30
Sample: 1 49854 IF EXCESSOB=0 AND ST7_MAX=0 AND
OUTLIER=0 AND CAS_FERRIES=0
Included observations: 41009
Prediction Assessment (success cutoff C = 0.078)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 30198 819 31017 37811 3198 41009
P(Dep=1)>C 7613 2379 9992 0 0 0
Total 37811 3198 41009 37811 3198 41009
Correct 30198 2379 32577 37811 0 37811
% Correct 79.87 74.39 79.44 100.00 0.00 92.20
% Incorrect 20.13 25.61 20.56 0.00 100.00 7.80
Total Gain* -20.13 74.39 -12.76
Percent Gain** NA 74.39 -163.66
-4
0
4
8
12
16
20
10001 20000 30000 40000
Standardized (Pearson) Residuals
267
Appendix 23: Results of Regressions: Less Serious Casualties
Dependent Variable: CASUALTY
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/27/06 Time: 10:38
Sample: 1 48470 IF EXCESSOB=0 AND ST7_MAX=0 AND
OUTLIER=0 AND CAS_FERRIES=0
Included observations: 39929
Convergence achieved after 8 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
C -5.662095 0.431350 -13.12645 0.0000
ST1_MAX 0.845100 0.072206 11.70397 0.0000
ST2_MAX 0.824041 0.089573 9.199675 0.0000
ST3_MAX 0.723681 0.099356 7.283716 0.0000
ST5_MAX 0.821732 0.117352 7.002274 0.0000
CLWD_D 0.430677 0.077533 5.554728 0.0000
CLCHGD -2.029244 0.122660 -16.54367 0.0000
OWCHGD 2.158379 0.076610 28.17358 0.0000
CL_ABS 1.204653 0.295476 4.076985 0.0000
CL_BV 1.457173 0.292212 4.986704 0.0000
CL_CCS 1.311247 0.329339 3.981447 0.0001
CL_CRR 3.706405 0.498327 7.437704 0.0000
CL_DNV 1.402495 0.295897 4.739813 0.0000
CL_GL 1.236372 0.295490 4.184141 0.0000
CL_HEL 2.506026 0.337051 7.435146 0.0000
CL_KRS 1.974537 0.346740 5.694581 0.0000
CL_LR 1.317811 0.287180 4.588793 0.0000
CL_NCL 1.102077 0.284640 3.871832 0.0001
CL_NKK 1.169648 0.293507 3.985080 0.0001
CL_PRS 2.362387 0.416834 5.667447 0.0000
CL_RIN 1.079326 0.380013 2.840238 0.0045
CL_RMS 1.532007 0.318685 4.807282 0.0000
CL_RRR 1.711612 0.561667 3.047382 0.0023
CL_TLL 1.342912 0.514616 2.609543 0.0091
CL_VRS 2.256384 0.495766 4.551311 0.0000
FL_AE 4.149804 0.433652 9.569437 0.0000
FL_AG 1.838566 0.278182 6.609221 0.0000
FL_AN 5.608286 0.725942 7.725529 0.0000
FL_AU 5.631484 0.723772 7.780747 0.0000
FL_BB 4.933665 0.626587 7.873869 0.0000
FL_BE 3.893763 0.766928 5.077092 0.0000
FL_BG 2.685595 0.764053 3.514931 0.0004
FL_BS 1.759789 0.275944 6.377345 0.0000
FL_BZ 2.119129 0.317121 6.682398 0.0000
FL_CA 6.898436 0.567459 12.15672 0.0000
FL_CH 5.370431 0.909441 5.905199 0.0000
FL_CI -0.620512 0.083064 -7.470320 0.0000
FL_CL 3.994457 0.798261 5.003948 0.0000
268
FL_CN 0.777765 0.318388 2.442818 0.0146
FL_CY 2.544963 0.333703 7.626440 0.0000
FL_DE 5.972199 0.570626 10.46605 0.0000
FL_DK 3.153448 0.386906 8.150430 0.0000
FL_DZ 1.267793 0.430365 2.945859 0.0032
FL_EE 2.484665 0.581987 4.269282 0.0000
FL_EG 2.697314 0.473018 5.702354 0.0000
FL_ES 3.974581 0.506615 7.845370 0.0000
FL_FI 6.080535 0.741642 8.198752 0.0000
FL_FR 4.397584 0.605152 7.266913 0.0000
FL_FO 7.237445 1.059427 6.831470 0.0000
FL_GI 2.484202 0.676733 3.670873 0.0002
FL_GR 3.742477 0.314934 11.88337 0.0000
FL_HK 3.350857 0.556536 6.020920 0.0000
FL_IE 5.284034 0.654636 8.071714 0.0000
FL_IM 5.420643 0.650410 8.334199 0.0000
FL_IN 3.044888 0.363502 8.376536 0.0000
FL_IR 1.197281 0.364148 3.287893 0.0010
FL_IT 3.305025 0.403428 8.192352 0.0000
FL_JP 5.178334 0.436476 11.86395 0.0000
FL_KH 2.007363 0.292492 6.862967 0.0000
FL_KR 2.740739 0.366186 7.484554 0.0000
FL_KW 3.304806 0.540381 6.115694 0.0000
FL_LR 1.044934 0.258157 4.047670 0.0001
FL_LU 4.551713 0.803656 5.663762 0.0000
FL_MA 1.587842 0.533334 2.977199 0.0029
FL_MH 3.340688 0.611796 5.460458 0.0000
FL_MT 5.942110 0.556128 10.68479 0.0000
FL_MY 2.382863 0.358829 6.640672 0.0000
FL_NIS 6.008787 0.626406 9.592484 0.0000
FL_NL 5.154772 0.581143 8.870055 0.0000
FL_NO 5.967297 0.626447 9.525626 0.0000
FL_NZ 5.648706 0.969468 5.826603 0.0000
FL_PA 1.200002 0.202225 5.933991 0.0000
FL_PT 2.273818 0.523192 4.346050 0.0000
FL_RO 1.583088 0.473523 3.343211 0.0008
FL_RU 0.975368 0.298556 3.266955 0.0011
FL_SE 5.650561 0.675426 8.365922 0.0000
FL_SG 4.701044 0.647508 7.260205 0.0000
FL_ST 2.780440 0.428527 6.488363 0.0000
FL_TR 0.756630 0.298769 2.532494 0.0113
FL_UK 6.162960 0.587306 10.49361 0.0000
FL_US 4.881812 0.484951 10.06662 0.0000
FL_VC 1.452105 0.243401 5.965899 0.0000
FL_VU 1.599494 0.479847 3.333343 0.0009
OW_EMN 1.929815 0.191901 10.05632 0.0000
OW_IOR 2.442122 0.214462 11.38722 0.0000
OW_OOR 2.777308 0.225334 12.32530 0.0000
OW_TMN 2.641438 0.181311 14.56851 0.0000
LI_FLRET -0.180436 0.016916 -10.66689 0.0000
SY_BE 2.213023 0.344335 6.426948 0.0000
269
SY_BG 1.707687 0.274525 6.220527 0.0000
SY_BR 1.964688 0.313022 6.276508 0.0000
SY_CA 1.475623 0.298032 4.951221 0.0000
SY_CN 1.262579 0.176012 7.173265 0.0000
SY_DE 1.230984 0.124598 9.879638 0.0000
SY_DK 1.377129 0.192021 7.171746 0.0000
SY_ES 1.301859 0.196971 6.609412 0.0000
SY_FI 1.207492 0.216578 5.575336 0.0000
SY_FR 1.248108 0.265603 4.699141 0.0000
SY_GR 2.469552 0.309820 7.970914 0.0000
SY_HR 1.527037 0.242211 6.304571 0.0000
SY_IT 1.522639 0.234637 6.489328 0.0000
SY_JP 0.919026 0.108796 8.447201 0.0000
SY_KR 1.283047 0.131561 9.752507 0.0000
SY_NL 1.105055 0.161741 6.832242 0.0000
SY_NO 1.664099 0.162979 10.21050 0.0000
SY_OT 2.544598 0.236796 10.74596 0.0000
SY_PL 1.156815 0.201514 5.740614 0.0000
SY_PT 1.609993 0.389076 4.137992 0.0000
SY_RO 1.314498 0.213359 6.160954 0.0000
SY_RU 0.654423 0.217621 3.007165 0.0026
SY_SE 0.937418 0.325276 2.881918 0.0040
SY_TR 1.640065 0.277260 5.915255 0.0000
SY_UK 1.407815 0.181451 7.758633 0.0000
SY_US 1.483511 0.231454 6.409538 0.0000
AMSA_S -0.159485 0.027889 -5.718667 0.0000
PMOU_S 0.053558 0.008707 6.151326 0.0000
USCG_S 0.046780 0.010379 4.507163 0.0000
Mean dependent var 0.053044 S.D. dependent var 0.224124
S.E. of regression 0.201377 Akaike info criterion 0.316690
Sum squared resid 1614.486 Schwarz criterion 0.341874
Log likelihood -6205.550 Hannan-Quinn criter. 0.324662
Restr. log likelihood -8280.589 Avg. log likelihood -0.155415
LR statistic (116 df) 4150.079 McFadden R-squared 0.250591
Probability(LR stat) 0.000000
Obs with Dep=0 37811 Total obs 39929
Obs with Dep=1 2118
270
Dependent Variable: CASUALTY
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/27/06 Time: 10:38
Sample: 1 48470 IF EXCESSOB=0 AND ST7_MAX=0 AND OUTLIER=0 AND
CAS_FERRIES=0
Included observations: 39929
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 1.E-05 0.0020 3991 3987.56 1 4.43652 3992 2.66488
2 0.0020 0.0047 3974 3979.50 19 13.4998 3993 2.24851
3 0.0047 0.0083 3962 3966.64 31 26.3609 3993 0.82183
4 0.0083 0.0135 3946 3950.09 47 42.9095 3993 0.39417
5 0.0135 0.0204 3912 3927.14 81 65.8568 3993 3.54046
6 0.0204 0.0319 3877 3889.60 116 103.400 3993 1.57617
7 0.0319 0.0479 3814 3834.32 179 158.676 3993 2.71099
8 0.0479 0.0741 3773 3758.92 220 234.076 3993 0.89910
9 0.0742 0.1210 3626 3607.65 367 385.351 3993 0.96730
10 0.1210 0.9941 2936 2909.57 1057 1083.43 3993 0.88507
Total 37811 37811.0 2118 2118.00 39929 16.7085
H-L Statistic: 16.7085 Prob. Chi-Sq(8) 0.0333
Andrews Statistic: 74.1985 Prob. Chi-Sq(10) 0.0000
Dependent Variable: CASUALTY
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 04/27/06 Time: 10:38
Sample: 1 48470 IF EXCESSOB=0 AND ST7_MAX=0 AND
OUTLIER=0 AND CAS_FERRIES=0
Included observations: 39929
Prediction Assessment (success cutoff C = 0.053)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 28769 542 29311 0 0 0
P(Dep=1)>C 9042 1576 10618 37811 2118 39929
Total 37811 2118 39929 37811 2118 39929
Correct 28769 1576 30345 0 2118 2118
% Correct 76.09 74.41 76.00 0.00 100.00 5.30
% Incorrect 23.91 25.59 24.00 100.00 0.00 94.70
Total Gain* 76.09 -25.59 70.69
Percent Gain** 76.09 NA 74.65
271
-4
0
4
8
12
16
20
24
28
32
10009 20000 30000 40000
Standardized (Pearson) Residuals
272
Appendix 24: Results of Regression: Fishing Fleet (above 400gt)
Dependent Variable: CASUALTY
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/04/06 Time: 18:28
Sample: 1 6666 IF FISH_CAS=0 AND OUTLIER2=0
Included observations: 6289
Convergence achieved after 7 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
C -4.453866 0.152841 -29.14046 0.0000
CLWD_D 10.25721 1.915184 5.355734 0.0000
OWCHGD 5.227738 0.602226 8.680697 0.0000
CL_BV 0.738830 0.354451 2.084435 0.0371
CL_CCO 3.299891 1.137701 2.900490 0.0037
FL_BZ 1.647911 0.563364 2.925124 0.0034
FL_CA 2.611569 0.458237 5.699162 0.0000
FL_DE 3.018409 0.792268 3.809832 0.0001
FL_FR 2.375173 0.451581 5.259680 0.0000
FL_GR 3.620932 0.807403 4.484664 0.0000
FL_NL 2.664820 0.358722 7.428656 0.0000
FL_NO 1.483681 0.337163 4.400490 0.0000
FL_US 1.304832 0.413060 3.158942 0.0016
OW_UNKN -2.714024 0.398907 -6.803655 0.0000
SY_UN 7.812672 0.993658 7.862533 0.0000
PSC_INSP -6.007325 1.416856 -4.239897 0.0000
Mean dependent var 0.025441 S.D. dependent var 0.157474
S.E. of regression 0.124985 Akaike info criterion 0.147310
Sum squared resid 97.99257 Schwarz criterion 0.164474
Log likelihood -447.2170 Hannan-Quinn criter. 0.153257
Restr. log likelihood -745.3686 Avg. log likelihood -0.071111
LR statistic (15 df) 596.3032 McFadden R-squared 0.400006
Probability(LR stat) 0.000000
Obs with Dep=0 6129 Total obs 6289
Obs with Dep=1 160
273
Dependent Variable: CASUALTY
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/04/06 Time: 18:28
Sample: 1 6666 IF FISH_CAS=0 AND OUTLIER2=0
Included observations: 6289
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total HL
Low High Actual Expect Actual Expect Obs Value
1 2.E-06 0.0008 626 627.588 2 0.41164 628 6.13297
2 0.0008 0.0008 628 628.515 1 0.48455 629 0.54873
3 0.0008 0.0034 629 628.161 0 0.83897 629 0.84009
4 0.0034 0.0115 623 623.076 6 5.92412 629 0.00098
5 0.0115 0.0115 625 621.767 4 7.23333 629 1.46212
6 0.0115 0.0115 617 621.767 12 7.23333 629 3.17772
7 0.0115 0.0115 625 621.767 4 7.23333 629 1.46212
8 0.0115 0.0115 624 621.767 5 7.23333 629 0.69757
9 0.0115 0.0411 612 615.415 17 13.5855 629 0.87713
10 0.0411 1.0000 520 519.178 109 109.822 629 0.00745
Total 6129 6129.00 160 160.000 6289 15.2069
H-L Statistic: 15.2069 Prob. Chi-Sq(8) 0.0552
Andrews Statistic: 456.6065 Prob. Chi-Sq(10) 0.0000
Dependent Variable: CASUALTY
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/04/06 Time: 18:28
Sample: 1 6666 IF FISH_CAS=0 AND OUTLIER2=0
Included observations: 6289
Prediction Assessment (success cutoff C = 0.025)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 5515 46 5561 0 0 0
P(Dep=1)>C 614 114 728 6129 160 6289
Total 6129 160 6289 6129 160 6289
Correct 5515 114 5629 0 160 160
% Correct 89.98 71.25 89.51 0.00 100.00 2.54
% Incorrect 10.02 28.75 10.49 100.00 0.00 97.46
Total Gain* 89.98 -28.75 86.96
Percent Gain** 89.98 NA 89.23
274
275
Appendix 25: Probability of Casualty per DoC Country of Residence
Probability of Casualty (Very Serious and Serious)
0.000 0.005 0.010 0.015 0.020 0.025 0.030
Cyprus
Azerbaijan
Isle of Man
Denmark
Canada
Sweden
Irish Republic
Monaco
United Kingdom
Norway
Unknown
Latvia
Australia
Germany
Malta
Netherlands
Lithuania
UAE
Thailand
USA
Hong Kong
Malaysia
Italy
Croatia
Brazil
Switzerland
Greece
India
Philippines
Spain
Singapore
Belgium
Bahamas
Chile
276
Appendix 26: LM-Test Type I (Very Serious) and Type II Models
Type I: Very Serious Casualties – Age
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 05/30/06 Time: 17:18
Sample: 1 6007 IF ST5_MAX=0 AND CMOU_AV=0
Included observations: 5826
Variable Coefficient Std. Error t-Statistic Prob.
FAC 3.816158 2.248451 1.697239 0.0897
ST3_MAX*FAC -0.688024 0.594013 -1.158263 0.2468
LNAGE*FAC 0.856866 0.472976 1.811650 0.0701
LNTON*FAC -0.342561 0.191535 -1.788503 0.0737
OWCHD*FAC -0.459591 0.345862 -1.328829 0.1840
RS_2S*FAC 0.691854 0.415730 1.664189 0.0961
RS_3S*FAC 0.883220 0.549765 1.606540 0.1082
RS_4S*FAC 1.328754 0.836747 1.588000 0.1123
RS_5S*FAC 1.275832 1.062454 1.200835 0.2299
VMOU_AV*FAC 1.042776 0.671201 1.553597 0.1203
CL_CRR*FAC -1.752671 1.456234 -1.203565 0.2288
CL_RIP*FAC -2.790060 1.865696 -1.495453 0.1349
CL_RNR*FAC -3.035841 1.982525 -1.531301 0.1257
FL_AG*FAC -1.002843 0.662349 -1.514070 0.1301
FL_BB*FAC -1.377179 1.395092 -0.987160 0.3236
FL_BR*FAC -2.577928 1.834607 -1.405166 0.1600
FL_DK*FAC -0.843540 0.706944 -1.193220 0.2328
FL_KR*FAC -1.396368 1.033025 -1.351728 0.1765
FL_LT*FAC -1.874441 1.325640 -1.413989 0.1574
FL_TR*FAC -0.956142 0.640660 -1.492433 0.1356
C_1800S*FAC -0.344962 0.286488 -1.204106 0.2286
LI_OWRET*FAC -0.067387 0.051240 -1.315138 0.1885
OW_OUK*FAC -2.249439 1.318023 -1.706677 0.0879
LNAGE*(-XB)*FAC -0.224472 0.116523 -1.926428 0.0541
R-squared 0.000619 Mean dependent var 0.004756
Adjusted R-squared -0.003342 S.D. dependent var 1.069765
S.E. of regression 1.071551 Akaike info criterion 2.980202
Sum squared resid 6661.980 Schwarz criterion 3.007679
Log likelihood -8657.329 Durbin-Watson stat 2.062138
277
Type I: Very Serious Casualties – Tonnage
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 05/30/06 Time: 17:21
Sample: 1 6007 IF ST5_MAX=0 AND CMOU_AV=0
Included observations: 5826
Variable Coefficient Std. Error t-Statistic Prob.
FAC -4.277321 2.510763 -1.703594 0.0885
ST3_MAX*FAC 1.262280 0.822057 1.535514 0.1247
LNAGE*FAC -0.302720 0.227535 -1.330430 0.1834
LNTON*FAC 0.107195 0.091181 1.175628 0.2398
OWCHD*FAC 0.615127 0.411925 1.493297 0.1354
RS_2S*FAC -0.875860 0.510625 -1.715272 0.0863
RS_3S*FAC -1.219137 0.715719 -1.703375 0.0886
RS_4S*FAC -2.330176 1.326461 -1.756687 0.0790
RS_5S*FAC -2.911103 1.756744 -1.657102 0.0976
VMOU_AV*FAC -1.207598 0.754851 -1.599782 0.1097
CL_CRR*FAC 1.966090 1.544596 1.272883 0.2031
CL_RIP*FAC 3.119344 2.033283 1.534142 0.1250
CL_RNR*FAC 3.425904 2.182950 1.569392 0.1166
FL_AG*FAC 1.187037 0.752388 1.577692 0.1147
FL_BB*FAC 1.578585 1.462801 1.079153 0.2806
FL_BR*FAC 3.345773 2.177136 1.536777 0.1244
FL_DK*FAC 1.070293 0.794926 1.346404 0.1782
FL_KR*FAC 1.899741 1.249839 1.519988 0.1286
FL_LT*FAC 2.178138 1.466937 1.484821 0.1376
FL_TR*FAC 1.070821 0.698779 1.532416 0.1255
C_1800S*FAC 0.445996 0.325958 1.368263 0.1713
LI_OWRET*FAC 0.085394 0.058852 1.450997 0.1468
OW_OUK*FAC 2.841596 1.629914 1.743403 0.0813
LNTON*(-XB)*FAC 0.099291 0.052795 1.880703 0.0601
R-squared 0.000589 Mean dependent var 0.004756
Adjusted R-squared -0.003372 S.D. dependent var 1.069765
S.E. of regression 1.071567 Akaike info criterion 2.980232
Sum squared resid 6662.180 Schwarz criterion 3.007709
Log likelihood -8657.416 Durbin-Watson stat 2.062354
278
Type II: Combined Model – Age
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 06/23/06 Time: 14:45
Sample: 1 52150
Included observations: 52150
Variable Coefficient Std. Error t-Statistic Prob.
FAC 0.802483 0.542158 1.480166 0.1388
ST1_MAX*FAC -0.055114 0.104687 -0.526465 0.5986
ST3_MAX*FAC -0.191784 0.160979 -1.191359 0.2335
ST4_MAX*FAC -0.065201 0.123207 -0.529198 0.5967
ST5_MAX*FAC -0.464292 0.276618 -1.678459 0.0933
LNAGE*FAC 0.282333 0.142524 1.980942 0.0476
LNTON*FAC -0.101391 0.059035 -1.717479 0.0859
ST_CHDGC*FAC -0.112992 0.085962 -1.314434 0.1887
RS_2S*FAC 0.231018 0.130246 1.773708 0.0761
RS_3S*FAC 0.260911 0.151103 1.726713 0.0842
RS_4S*FAC 0.495869 0.268861 1.844333 0.0651
RS_5S*FAC 0.618580 0.374182 1.653152 0.0983
DH*FAC -0.072933 0.138917 -0.525011 0.5996
LNTIMEBW*FAC -0.024097 0.025575 -0.942221 0.3461
CL_BV*FAC 0.076292 0.101528 0.751431 0.4524
CL_HEL*FAC 0.465924 0.628728 0.741058 0.4587
FL_AG*FAC -0.103477 0.130718 -0.791607 0.4286
FL_BH*FAC -0.488965 1.499745 -0.326032 0.7444
FL_BS*FAC -0.074702 0.126477 -0.590642 0.5548
FL_CA*FAC -0.660126 0.471895 -1.398884 0.1619
FL_DE*FAC -0.159733 0.254910 -0.626624 0.5309
FL_GI*FAC -0.219494 0.356653 -0.615428 0.5383
FL_IM*FAC -0.123927 0.280042 -0.442530 0.6581
FL_KR*FAC -0.212736 0.422223 -0.503848 0.6144
FL_RU*FAC 0.280691 0.304270 0.922507 0.3563
FL_ST*FAC -0.413538 0.622629 -0.664181 0.5066
OW_EMN*FAC 0.065538 0.085435 0.767110 0.4430
C_0400S3*FAC 0.135533 0.264393 0.512620 0.6082
C_0700S1*FAC -0.011850 0.012397 -0.955919 0.3391
C_0900S3*FAC -0.019664 0.028536 -0.689103 0.4908
C_1200S1*FAC 0.013908 0.018527 0.750688 0.4528
C_1200S4*FAC -0.032308 0.041020 -0.787596 0.4309
C_1400S2*FAC -0.010947 0.010645 -1.028362 0.3038
C_2000S1*FAC 0.026954 0.037100 0.726527 0.4675
C_2500S1*FAC -0.010139 0.020403 -0.496968 0.6192
C_2500S4*FAC -0.029720 0.045217 -0.657285 0.5110
AMSA_AV*FAC 0.093556 0.175380 0.533446 0.5937
IMOU_AV*FAC 0.322921 0.286854 1.125730 0.2603
USCG_AV*FAC 0.084551 0.132922 0.636097 0.5247
VMOU_AV*FAC 0.181581 0.181086 1.002733 0.3160
LNAGE*(-XB)*FAC -0.078366 0.037127 -2.110739 0.0348
279
R-squared 0.000082 Mean dependent var 0.001891
Adjusted R-squared -0.000685 S.D. dependent var 1.048421
S.E. of regression 1.048780 Akaike info criterion 2.933919
Sum squared resid 57316.80 Schwarz criterion 2.940886
Log likelihood -76460.94 Durbin-Watson stat 1.887829
Type II: Combined Model – Tonnage
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 06/23/06 Time: 14:58
Sample: 1 52150
Included observations: 52150
Variable Coefficient Std. Error t-Statistic Prob.
FAC 2.987090 0.785241 3.804041 0.0001
ST1_MAX*FAC -0.330938 0.126530 -2.615490 0.0089
ST3_MAX*FAC -1.077770 0.280147 -3.847152 0.0001
ST4_MAX*FAC -0.352821 0.144016 -2.449881 0.0143
ST5_MAX*FAC -2.316394 0.555976 -4.166355 0.0000
LNAGE*FAC 0.231532 0.072305 3.202158 0.0014
LNTON*FAC -0.102351 0.041544 -2.463670 0.0138
ST_CHDGC*FAC -0.386258 0.111065 -3.477756 0.0005
RS_2S*FAC 0.832613 0.203191 4.097681 0.0000
RS_3S*FAC 1.026036 0.250359 4.098260 0.0000
RS_4S*FAC 2.478482 0.582044 4.258239 0.0000
RS_5S*FAC 3.527098 0.839988 4.198987 0.0000
DH*FAC -0.405185 0.163390 -2.479868 0.0131
LNTIMEBW*FAC -0.092060 0.031103 -2.959834 0.0031
CL_BV*FAC 0.360533 0.125722 2.867708 0.0041
CL_HEL*FAC 1.538719 0.685903 2.243348 0.0249
FL_AG*FAC -0.391809 0.150727 -2.599454 0.0093
FL_BH*FAC -2.286643 1.571232 -1.455319 0.1456
FL_BS*FAC -0.355356 0.146125 -2.431865 0.0150
FL_CA*FAC -2.396406 0.652358 -3.673452 0.0002
FL_DE*FAC -0.865098 0.313730 -2.757458 0.0058
FL_GI*FAC -0.945998 0.404004 -2.341556 0.0192
FL_IM*FAC -0.605664 0.306863 -1.973725 0.0484
FL_KR*FAC -0.955816 0.464660 -2.057024 0.0397
FL_RU*FAC 0.829785 0.333065 2.491365 0.0127
FL_ST*FAC -1.339052 0.665643 -2.011667 0.0443
OW_EMN*FAC 0.250449 0.098071 2.553740 0.0107
C_0400S3*FAC 0.530983 0.283772 1.871159 0.0613
C_0700S1*FAC -0.038521 0.014136 -2.725026 0.0064
C_0900S3*FAC -0.047446 0.029072 -1.632007 0.1027
C_1200S1*FAC 0.042411 0.019848 2.136750 0.0326
C_1200S4*FAC -0.114117 0.046151 -2.472673 0.0134
C_1400S2*FAC -0.041536 0.013294 -3.124297 0.0018
C_2000S1*FAC 0.086310 0.040039 2.155647 0.0311
C_2500S1*FAC -0.031763 0.021117 -1.504164 0.1325
280
C_2500S4*FAC -0.094432 0.048091 -1.963593 0.0496
AMSA_AV*FAC 0.474735 0.201463 2.356430 0.0185
IMOU_AV*FAC 1.207106 0.367651 3.283291 0.0010
USCG_AV*FAC 0.245800 0.138647 1.772845 0.0763
VMOU_AV*FAC 0.755109 0.235039 3.212695 0.0013
LNTON*(-XB)*FAC -0.096752 0.022141 -4.369893 0.0000
R-squared 0.000363 Mean dependent var 0.001891
Adjusted R-squared -0.000404 S.D. dependent var 1.048421
S.E. of regression 1.048633 Akaike info criterion 2.933638
Sum squared resid 57300.70 Schwarz criterion 2.940605
Log likelihood -76453.61 Durbin-Watson stat 1.888229
281
Appendix 27: Matching Models: Type I Models
6 months- Very Serious
Dependent Variable: TOTALCAS
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/30/06 Time: 16:23
Sample: 1 6007 IF ST5_MAX=0 AND CMOU_AV=0
Included observations: 5826
Convergence achieved after 7 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
C -8.131409 0.966171 -8.416118 0.0000
ST3_MAX 1.315392 0.456566 2.881057 0.0040
LNAGE -0.366059 0.119815 -3.055207 0.0022
LNTON 0.475650 0.065115 7.304805 0.0000
OWCHD 0.702768 0.245771 2.859440 0.0042
RS_2S -1.001358 0.194661 -5.144104 0.0000
RS_3S -1.344727 0.287850 -4.671626 0.0000
RS_4S -2.499658 0.408670 -6.116572 0.0000
RS_5S -3.005631 0.765229 -3.927752 0.0001
VMOU_AV -1.381298 0.353008 -3.912941 0.0001
CL_CRR 2.738680 0.775992 3.529261 0.0004
CL_RIP 4.005732 0.909348 4.405061 0.0000
CL_RNR 4.173554 0.903595 4.618832 0.0000
FL_AG 1.488832 0.390656 3.811107 0.0001
FL_BB 2.060601 0.785528 2.623204 0.0087
FL_BR 3.683345 1.176175 3.131631 0.0017
FL_DK 1.325811 0.509765 2.600825 0.0093
FL_KR 2.127879 0.777016 2.738528 0.0062
FL_LT 2.762602 0.868769 3.179903 0.0015
FL_TR 1.381422 0.366013 3.774246 0.0002
C_1800S 0.503767 0.174895 2.880395 0.0040
LI_OWRET 0.102830 0.034406 2.988733 0.0028
OW_OUK 3.363223 0.570478 5.895451 0.0000
Mean dependent var 0.027635 S.D. dependent var 0.163938
S.E. of regression 0.157107 Akaike info criterion 0.218708
Sum squared resid 143.2335 Schwarz criterion 0.245040
Log likelihood -614.0955 Hannan-Quinn criter. 0.227866
Restr. log likelihood -736.5324 Avg. log likelihood -0.105406
LR statistic (22 df) 244.8738 McFadden R-squared 0.166234
Probability(LR stat) 0.000000
Obs with Dep=0 5665 Total obs 5826
Obs with Dep=1 161
282
Dependent Variable: TOTALCAS
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/30/06 Time: 16:23
Sample: 1 6007 IF ST5_MAX=0 AND CMOU_AV=0
Included observations: 5826
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 0.0003 0.0027 580 581.031 2 0.96937 582 1.09758
2 0.0027 0.0048 581 580.811 2 2.18867 583 0.01633
3 0.0048 0.0065 578 578.706 4 3.29357 582 0.15238
4 0.0065 0.0090 576 578.525 7 4.47538 583 1.43518
5 0.0090 0.0129 577 576.723 6 6.27668 583 0.01233
6 0.0129 0.0185 570 572.990 12 9.01004 582 1.00782
7 0.0185 0.0243 575 570.580 8 12.4197 583 1.60702
8 0.0243 0.0370 567 564.260 15 17.7401 582 0.43655
9 0.0370 0.0658 562 553.671 21 29.3294 583 2.49081
10 0.0662 0.8278 499 507.703 84 75.2971 583 1.15507
Total 5665 5665.00 161 161.000 5826 9.41107
H-L Statistic: 9.4111 Prob. Chi-Sq(8) 0.3088
Andrews Statistic: 11.4298 Prob. Chi-Sq(10) 0.3250
Dependent Variable: TOTALCAS
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/30/06 Time: 16:23
Sample: 1 6007 IF ST5_MAX=0 AND CMOU_AV=0
Included observations: 5826
Prediction Assessment (success cutoff C = 0.0276)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 4188 46 4234 0 0 0
P(Dep=1)>C 1477 115 1592 5665 161 5826
Total 5665 161 5826 5665 161 5826
Correct 4188 115 4303 0 161 161
% Correct 73.93 71.43 73.86 0.00 100.00 2.76
% Incorrect 26.07 28.57 26.14 100.00 0.00 97.24
Total Gain* 73.93 -28.57 71.10
Percent Gain** 73.93 NA 73.12
283
-5
0
5
10
15
20
25
30
1000 2000 3000 4000 5000 6000
Standardized (Pearson) Residuals
6m – Serious
Dependent Variable: TOTALCAS
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/24/06 Time: 12:58
Sample: 1 46522 IF CMOU_AV=0
Included observations: 45486
Convergence achieved after 10 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
C -6.623591 0.376920 -17.57294 0.0000
ST3_MAX 0.826772 0.118911 6.952848 0.0000
ST4_MAX 0.423335 0.108020 3.919047 0.0001
ST5_MAX 2.082840 0.169789 12.26720 0.0000
LNAGE -0.300454 0.049777 -6.035965 0.0000
LNTON 0.428150 0.034595 12.37602 0.0000
ST_CHDGC 0.474506 0.069366 6.840660 0.0000
CLWD_D -0.199273 0.069145 -2.881952 0.0040
RS_2S -0.961319 0.074250 -12.94713 0.0000
RS_3S -1.153201 0.083478 -13.81449 0.0000
RS_4S -2.749773 0.130410 -21.08558 0.0000
RS_5S -3.905507 0.239559 -16.30290 0.0000
DH 0.368804 0.136463 2.702595 0.0069
AMSA_AV -0.566170 0.169921 -3.331960 0.0009
IMOU_AV -1.560759 0.248083 -6.291277 0.0000
VMOU_AV -0.632282 0.160008 -3.951565 0.0001
LNTIMEBW 0.152592 0.021536 7.085344 0.0000
CL_DNV 0.360955 0.095647 3.773807 0.0002
CL_GL 0.849893 0.084105 10.10520 0.0000
FL_AL -5.142419 2.116853 -2.429275 0.0151
FL_BM 1.132910 0.407502 2.780133 0.0054
FL_BS 0.388054 0.125488 3.092348 0.0020
FL_CA 2.683677 0.304742 8.806386 0.0000
284
FL_CH 1.550844 0.623005 2.489295 0.0128
FL_GI 0.889458 0.381067 2.334125 0.0196
FL_IM 0.697365 0.264829 2.633263 0.0085
FL_KY 0.776826 0.318833 2.436467 0.0148
FL_PL 1.738386 0.529635 3.282236 0.0010
FL_ST 2.112332 0.567356 3.723114 0.0002
FL_UK 0.716200 0.220370 3.249991 0.0012
C_0700S 0.033536 0.007677 4.368208 0.0000
C_1400S 0.031815 0.007447 4.272185 0.0000
C_2000S -0.072425 0.024310 -2.979266 0.0029
C_2500S 0.039248 0.014616 2.685257 0.0072
Mean dependent var 0.029943 S.D. dependent var 0.170433
S.E. of regression 0.165060 Akaike info criterion 0.232922
Sum squared resid 1238.337 Schwarz criterion 0.239444
Log likelihood -5263.344 Hannan-Quinn criter. 0.234974
Restr. log likelihood -6119.911 Avg. log likelihood -0.115713
LR statistic (33 df) 1713.135 McFadden R-squared 0.139964
Probability(LR stat) 0.000000
Obs with Dep=0 44124 Total obs 45486
Obs with Dep=1 1362
Dependent Variable: TOTALCAS
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/24/06 Time: 12:58
Sample: 1 46522 IF CMOU_AV=0
Included observations: 45486
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 2.E-05 0.0030 4536 4539.72 12 8.27936 4548 1.67505
2 0.0030 0.0057 4532 4529.42 17 19.5801 4549 0.34145
3 0.0057 0.0090 4517 4514.78 31 33.2180 4548 0.14918
4 0.0090 0.0121 4503 4500.98 46 48.0207 4549 0.08593
5 0.0121 0.0171 4480 4483.19 69 65.8108 4549 0.15682
6 0.0171 0.0237 4462 4455.75 86 92.2514 4548 0.43239
7 0.0237 0.0311 4428 4425.24 121 123.760 4549 0.06327
8 0.0311 0.0433 4380 4381.88 168 166.119 4548 0.02210
9 0.0433 0.0670 4306 4307.20 243 241.798 4549 0.00631
10 0.0670 0.8432 3980 3985.84 569 563.163 4549 0.06905
Total 44124 44124.0 1362 1362.00 45486 3.00156
H-L Statistic: 3.0016 Prob. Chi-Sq(8) 0.9343
Andrews Statistic: 3.3017 Prob. Chi-Sq(10) 0.9734
285
Dependent Variable: TOTALCAS
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/24/06 Time: 12:58
Sample: 1 46522 IF CMOU_AV=0
Included observations: 45486
Prediction Assessment (success cutoff C = 0.029943)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 30886 363 31249 0 0 0
P(Dep=1)>C 13238 999 14237 44124 1362 45486
Total 44124 1362 45486 44124 1362 45486
Correct 30886 999 31885 0 1362 1362
% Correct 70.00 73.35 70.10 0.00 100.00 2.99
% Incorrect 30.00 26.65 29.90 100.00 0.00 97.01
Total Gain* 70.00 -26.65 67.10
Percent Gain** 70.00 NA 69.18
-10
0
10
20
30
40
50
60
70
10000 20000 30000 40000
Standardized (Pearson) Residuals
286
6 month – Less Serious
Dependent Variable: TOTALCAS
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/22/06 Time: 12:04
Sample: 1 28008 IF CMOU_AV=0
Included observations: 27411
Convergence achieved after 10 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
C -7.425953 0.355822 -20.86983 0.0000
ST3_MAX 0.462733 0.136248 3.396253 0.0007
ST4_MAX 0.443063 0.116058 3.817600 0.0001
ST5_MAX 1.733403 0.184055 9.417846 0.0000
DET_AMSA 0.764803 0.191423 3.995350 0.0001
LNTON 0.337702 0.037472 9.012093 0.0000
OWCHD 0.277172 0.102161 2.713082 0.0067
ST_CHDGC 0.579459 0.084296 6.874060 0.0000
AMSA_AV -0.973242 0.233935 -4.160317 0.0000
IMOU_AV -1.545097 0.355240 -4.349452 0.0000
VMOU_AV -0.494673 0.205342 -2.409021 0.0160
LNTIMEBW 0.114478 0.025498 4.489647 0.0000
RS_3S -0.306589 0.099377 -3.085114 0.0020
RS_4S -0.365418 0.100622 -3.631608 0.0003
RS_5S -0.493139 0.147355 -3.346597 0.0008
CL_INC 2.281758 0.920734 2.478194 0.0132
CL_LR 0.250316 0.099454 2.516892 0.0118
FL_BB 1.022071 0.380614 2.685325 0.0072
FL_BH 4.362188 1.385884 3.147585 0.0016
FL_CA 2.533080 0.443200 5.715434 0.0000
FL_EG 1.934990 0.435610 4.442022 0.0000
FL_GE -3.507990 1.084069 -3.235948 0.0012
FL_IR 1.136576 0.388430 2.926079 0.0034
FL_SY -3.049523 1.054591 -2.891665 0.0038
FL_UK 0.818191 0.240268 3.405333 0.0007
C_0900S 0.024972 0.007878 3.170072 0.0015
C_1400S 0.023392 0.010582 2.210591 0.0271
C_1700S 0.047355 0.015966 2.965966 0.0030
OW_TMN 0.285909 0.086792 3.294167 0.0010
Mean dependent var 0.031374 S.D. dependent var 0.174330
S.E. of regression 0.171646 Akaike info criterion 0.260000
Sum squared resid 806.7374 Schwarz criterion 0.268695
Log likelihood -3534.435 Hannan-Quinn criter. 0.262802
Restr. log likelihood -3823.486 Avg. log likelihood -0.128942
LR statistic (28 df) 578.1021 McFadden R-squared 0.075599
Probability(LR stat) 0.000000
Obs with Dep=0 26551 Total obs 27411
Obs with Dep=1 860
287
Dependent Variable: TOTALCAS
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/22/06 Time: 12:04
Sample: 1 28008 IF CMOU_AV=0
Included observations: 27411
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 2.E-05 0.0095 2735 2728.26 6 12.7367 2741 3.57983
2 0.0095 0.0140 2706 2707.51 35 33.4913 2741 0.06881
3 0.0140 0.0168 2710 2698.70 31 42.2964 2741 3.06429
4 0.0168 0.0199 2693 2690.90 48 50.0994 2741 0.08961
5 0.0199 0.0235 2679 2681.69 62 59.3053 2741 0.12514
6 0.0235 0.0282 2674 2670.29 67 70.7110 2741 0.19991
7 0.0282 0.0345 2630 2656.03 111 84.9708 2741 8.22864
8 0.0345 0.0426 2635 2636.12 106 104.883 2741 0.01236
9 0.0426 0.0580 2599 2606.74 142 134.265 2741 0.46857
10 0.0580 0.8180 2490 2474.76 252 267.241 2742 0.96305
Total 26551 26551.0 860 860.000 27411 16.8002
H-L Statistic: 16.8002 Prob. Chi-Sq(8) 0.0323
Andrews Statistic: 36.3934 Prob. Chi-Sq(10) 0.0001
Dependent Variable: TOTALCAS
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/22/06 Time: 12:04
Sample: 1 28008 IF CMOU_AV=0
Included observations: 27411
Prediction Assessment (success cutoff C = 0.03137)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 17708 313 18021 0 0 0
P(Dep=1)>C 8843 547 9390 26551 860 27411
Total 26551 860 27411 26551 860 27411
Correct 17708 547 18255 0 860 860
% Correct 66.69 63.60 66.60 0.00 100.00 3.14
% Incorrect 33.31 36.40 33.40 100.00 0.00 96.86
Total Gain* 66.69 -36.40 63.46
Percent Gain** 66.69 NA 65.52
288
-10
0
10
20
30
40
5000 10000 15000 20000 25001
Standardized (Pearson) Residuals
289
Appendix 28: Matching Models: Type II Model
Very Serious and Serious Combined
Dependent Variable: TOTALCAS
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 06/22/06 Time: 22:52
Sample: 1 52150
Included observations: 52150
Convergence achieved after 7 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
C -6.459764 0.417999 -15.45403 0.0000
ST1_MAX 0.310843 0.104659 2.970046 0.0030
ST3_MAX 1.152232 0.124555 9.250785 0.0000
ST4_MAX 0.348282 0.114469 3.042589 0.0023
ST5_MAX 2.446845 0.163026 15.00895 0.0000
LNAGE -0.274650 0.045661 -6.015006 0.0000
LNTON 0.453563 0.038683 11.72519 0.0000
ST_CHDGC 0.466394 0.066705 6.991895 0.0000
RS_2S -0.979696 0.068677 -14.26520 0.0000
RS_3S -1.183811 0.080034 -14.79134 0.0000
RS_4S -2.771417 0.122720 -22.58320 0.0000
RS_5S -3.879339 0.226207 -17.14952 0.0000
DH 0.439719 0.131212 3.351204 0.0008
LNTIMEBW 0.110732 0.019175 5.774709 0.0000
CL_BV -0.424477 0.090092 -4.711605 0.0000
CL_HEL -1.864871 0.571568 -3.262727 0.0011
FL_AG 0.472199 0.119882 3.938857 0.0001
FL_BH 2.534354 0.988805 2.563048 0.0104
FL_BS 0.373358 0.117892 3.166960 0.0015
FL_CA 2.664072 0.321357 8.290062 0.0000
FL_DE 0.962096 0.228742 4.206037 0.0000
FL_GI 1.086758 0.355024 3.061083 0.0022
FL_IM 0.699019 0.263999 2.647808 0.0081
FL_KR 1.039287 0.399637 2.600580 0.0093
FL_RU -1.114659 0.263470 -4.230681 0.0000
FL_ST 1.704192 0.538017 3.167544 0.0015
OW_EMN -0.282571 0.076981 -3.670639 0.0002
C_0400S3 -0.614959 0.222312 -2.766198 0.0057
C_0700S1 0.049588 0.010973 4.518925 0.0000
C_0900S3 0.065725 0.021861 3.006474 0.0026
C_1200S1 -0.052770 0.015964 -3.305644 0.0009
C_1200S4 0.126155 0.034439 3.663143 0.0002
C_1400S2 0.046748 0.007643 6.116191 0.0000
C_2000S1 -0.111775 0.030128 -3.709996 0.0002
C_2500S1 0.041474 0.016796 2.469206 0.0135
C_2500S4 0.119394 0.041185 2.898938 0.0037
AMSA_AV -0.518260 0.165448 -3.132458 0.0017
290
IMOU_AV -1.419812 0.232253 -6.113200 0.0000
USCG_AV -0.318479 0.122117 -2.607979 0.0091
VMOU_AV -0.826658 0.146462 -5.644186 0.0000
Mean dependent var 0.029530 S.D. dependent var 0.169289
S.E. of regression 0.164276 Akaike info criterion 0.232897
Sum squared resid 1406.275 Schwarz criterion 0.239694
Log likelihood -6032.783 Hannan-Quinn criter. 0.235022
Restr. log likelihood -6941.441 Avg. log likelihood -0.115681
LR statistic (39 df) 1817.316 McFadden R-squared 0.130903
Probability(LR stat) 0.000000
Obs with Dep=0 50610 Total obs 52150
Obs with Dep=1 1540
Dependent Variable: TOTALCAS
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 06/22/06 Time: 22:52
Sample: 1 52150
Included observations: 52150
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 6.E-05 0.0033 5199 5205.17 16 9.82815 5215 3.88310
2 0.0033 0.0060 5194 5190.64 21 24.3634 5215 0.46650
3 0.0060 0.0088 5183 5176.89 32 38.1117 5215 0.98731
4 0.0088 0.0129 5155 5159.00 60 56.0044 5215 0.28815
5 0.0129 0.0179 5122 5135.88 93 79.1166 5215 2.47380
6 0.0179 0.0241 5106 5105.95 109 109.048 5215 2.2E-05
7 0.0241 0.0314 5089 5071.09 126 143.914 5215 2.29310
8 0.0314 0.0428 5037 5023.94 178 191.059 5215 0.92656
9 0.0428 0.0657 4932 4941.12 283 273.880 5215 0.32054
10 0.0657 0.9732 4593 4600.33 622 614.675 5215 0.09896
Total 50610 50610.0 1540 1540.00 52150 11.7380
H-L Statistic: 11.7380 Prob. Chi-Sq(8) 0.1633
Andrews Statistic: 10.9579 Prob. Chi-Sq(10) 0.3608
291
Dependent Variable: TOTALCAS
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 06/22/06 Time: 22:52
Sample: 1 52150
Included observations: 52150
Prediction Assessment (success cutoff C = 0.0295)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 34858 423 35281 0 0 0
P(Dep=1)>C 15752 1117 16869 50610 1540 52150
Total 50610 1540 52150 50610 1540 52150
Correct 34858 1117 35975 0 1540 1540
% Correct 68.88 72.53 68.98 0.00 100.00 2.95
% Incorrect 31.12 27.47 31.02 100.00 0.00 97.05
Total Gain* 68.88 -27.47 66.03
Percent Gain** 68.88 NA 68.04
-10
0
10
20
30
40
50
60
10000 20000 30000 40000 50000
Standardized (Pearson) Residuals
292
Appendix 29: LM Test: Type III Models
Fire/Explosion
Tonnage
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 05/31/06 Time: 11:15
Sample: 1 6218 IF CMOU_AV=0 AND ST3_MAX=0
Included observations: 5675
Variable Coefficient Std. Error t-Statistic Prob.
FAC 3.096052 2.851226 1.085867 0.2776
ST1_MAX*FAC 0.926331 0.853471 1.085369 0.2778
ST2_MAX*FAC 0.960204 0.895525 1.072225 0.2837
ST4_MAX*FAC 0.876841 0.827890 1.059127 0.2896
LNTON*FAC -0.155338 0.158336 -0.981067 0.3266
OW_EMN*FAC 0.844088 0.798600 1.056959 0.2906
OW_TMN*FAC 0.437090 0.435238 1.004255 0.3153
OW_IOR*FAC 1.154002 1.129190 1.021973 0.3068
ST_CHDGC*FAC -0.320626 0.331008 -0.968636 0.3328
IMOU_AV*FAC 1.161316 1.198474 0.968995 0.3326
VMOU_AV*FAC 0.899939 0.908426 0.990657 0.3219
LNTIMEBW*FAC -0.111826 0.115127 -0.971331 0.3314
FL_BG*FAC -1.617296 1.595029 -1.013960 0.3106
C_1400S*FAC -0.035910 0.036208 -0.991770 0.3214
C_1800S*FAC -0.234127 0.241923 -0.967776 0.3332
LNTON*(-XB)*FAC -0.083227 0.073513 -1.132148 0.2576
R-squared 0.000226 Mean dependent var -0.000103
Adjusted R-squared -0.002424 S.D. dependent var 0.988561
S.E. of regression 0.989758 Akaike info criterion 2.820103
Sum squared resid 5543.674 Schwarz criterion 2.838834
Log likelihood -7986.041 Durbin-Watson stat 1.953133
Wrecked/Stranded/Grounded
Age
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 05/30/06 Time: 22:13
Sample: 1 19131 IF CMOU_AV=0 AND ST5_MAX=0 AND
ST5_MAX=0
Included observations: 18600
Variable Coefficient Std. Error t-Statistic Prob.
FAC 0.156277 0.192608 0.811375 0.4172
ST1_MAX*FAC 0.071050 0.031129 2.282418 0.0225
ST2_MAX*FAC 0.067233 0.029557 2.274659 0.0229
293
DET_AMSA*FAC -0.040433 0.049176 -0.822201 0.4110
LNTON*FAC -0.018231 0.018131 -1.005513 0.3147
LNAGE*FAC 0.085997 0.039433 2.180819 0.0292
OWCHD*FAC -0.028728 0.023295 -1.233213 0.2175
ST_CHDGC*FAC -0.041397 0.030272 -1.367487 0.1715
AMSA_AV*FAC 0.083611 0.072560 1.152301 0.2492
IMOU_AV*FAC 0.125406 0.081172 1.544942 0.1224
CL_LR*FAC -0.045160 0.025192 -1.792631 0.0730
FL_BM*FAC -0.259467 0.177691 -1.460216 0.1442
FL_IM*FAC -0.040411 0.075706 -0.533781 0.5935
FL_RU*FAC 0.097281 0.083196 1.169301 0.2423
FL_UK*FAC -0.121256 0.068763 -1.763386 0.0779
C_0700S*FAC -0.001538 0.002164 -0.711027 0.4771
C_1000S*FAC -0.019906 0.017924 -1.110590 0.2668
C_1400S*FAC -0.003465 0.002365 -1.465588 0.1428
LNAGE*(-XB)*FAC -0.027083 0.013742 -1.970839 0.0488
R-squared 0.000587 Mean dependent var -1.48E-13
Adjusted R-squared -0.000381 S.D. dependent var 0.160131
S.E. of regression 0.160161 Akaike info criterion -0.824248
Sum squared resid 476.6338 Schwarz criterion -0.816249
Log likelihood 7684.509 Durbin-Watson stat 1.906424
Tonnage
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 05/30/06 Time: 22:18
Sample: 1 19131 IF CMOU_AV=0 AND ST5_MAX=0 AND
ST5_MAX=0
Included observations: 18600
Variable Coefficient Std. Error t-Statistic Prob.
FAC 0.785217 0.341382 2.300111 0.0215
ST1_MAX*FAC 0.137705 0.043500 3.165632 0.0015
ST2_MAX*FAC 0.164673 0.051731 3.183255 0.0015
DET_AMSA*FAC -0.147034 0.068045 -2.160819 0.0307
LNTON*FAC -0.000483 0.010015 -0.048235 0.9615
LNAGE*FAC -0.050594 0.024485 -2.066275 0.0388
OWCHD*FAC -0.074916 0.031244 -2.397808 0.0165
ST_CHDGC*FAC -0.112035 0.044446 -2.520706 0.0117
AMSA_AV*FAC 0.293536 0.117397 2.500373 0.0124
IMOU_AV*FAC 0.289128 0.110300 2.621290 0.0088
CL_LR*FAC -0.099497 0.035234 -2.823880 0.0047
FL_BM*FAC -0.558821 0.224304 -2.491351 0.0127
FL_IM*FAC -0.202903 0.103530 -1.959851 0.0500
FL_RU*FAC 0.206305 0.097293 2.120447 0.0340
FL_UK*FAC -0.331594 0.114365 -2.899429 0.0037
C_0700S*FAC -0.004267 0.002503 -1.705048 0.0882
C_1000S*FAC -0.048190 0.022207 -2.170057 0.0300
294
C_1400S*FAC -0.005912 0.002623 -2.254248 0.0242
LNTON*(-XB)*FAC -0.023044 0.007890 -2.920527 0.0035
R-squared 0.000837 Mean dependent var -1.48E-13
Adjusted R-squared -0.000131 S.D. dependent var 0.160131
S.E. of regression 0.160141 Akaike info criterion -0.824498
Sum squared resid 476.5147 Schwarz criterion -0.816499
Log likelihood 7686.833 Durbin-Watson stat 1.906463
Collision/Contact
Age
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 05/30/06 Time: 21:48
Sample: 1 23254 IF CMOU_AV=0 AND ST6_MAX=0
Included observations: 22329
Variable Coefficient Std. Error t-Statistic Prob.
FAC -0.607202 0.854825 -0.710323 0.4775
ST1_MAX*FAC -0.318448 0.384232 -0.828789 0.4072
ST2_MAX*FAC -0.320912 0.386051 -0.831268 0.4058
ST3_MAX*FAC -0.269751 0.360442 -0.748389 0.4542
ST4_MAX*FAC -0.263790 0.343996 -0.766841 0.4432
LNTON*FAC 0.073409 0.089489 0.820313 0.4120
LNAGE*FAC -0.168408 0.177503 -0.948758 0.3428
C_0900S*FAC 0.006767 0.011950 0.566276 0.5712
C_2100S*FAC -0.062268 0.144537 -0.430810 0.6666
C_2500S*FAC 0.013557 0.024830 0.545995 0.5851
ST_CHDGC*FAC 0.127889 0.158941 0.804632 0.4210
AMSA_AV*FAC -0.142339 0.278291 -0.511476 0.6090
IMOU_AV*FAC -0.212939 0.387633 -0.549331 0.5828
VMOU_AV*FAC -0.155418 0.290754 -0.534533 0.5930
CL_BV*FAC 0.082962 0.176251 0.470703 0.6379
CL_DNV*FAC 0.085382 0.179435 0.475835 0.6342
CL_GL*FAC 0.107528 0.178628 0.601969 0.5472
CL_LR*FAC 0.153876 0.201849 0.762333 0.4459
CL_NKK*FAC 0.082130 0.172115 0.477181 0.6332
CL_TLL*FAC 0.324735 0.526211 0.617119 0.5372
FL_GI*FAC 0.174485 0.469608 0.371555 0.7102
FL_KY*FAC 0.172906 0.441715 0.391441 0.6955
LNAGE*(-XB)*FAC 0.066550 0.066316 1.003527 0.3156
R-squared 0.000045 Mean dependent var -8.31E-05
Adjusted R-squared -0.000941 S.D. dependent var 1.005635
S.E. of regression 1.006108 Akaike info criterion 2.851085
Sum squared resid 22579.31 Schwarz criterion 2.859339
Log likelihood -31807.94 Durbin-Watson stat 1.908377
295
Tonnage
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 05/30/06 Time: 21:53
Sample: 1 23254 IF CMOU_AV=0 AND ST6_MAX=0
Included observations: 22329
Variable Coefficient Std. Error t-Statistic Prob.
FAC 2.398870 1.656344 1.448292 0.1475
ST1_MAX*FAC 1.094541 0.736331 1.486480 0.1372
ST2_MAX*FAC 1.181894 0.790078 1.495921 0.1347
ST3_MAX*FAC 0.836628 0.589076 1.420239 0.1556
ST4_MAX*FAC 0.920799 0.632245 1.456396 0.1453
LNTON*FAC -0.038124 0.057078 -0.667932 0.5042
LNAGE*FAC -0.250617 0.171200 -1.463880 0.1432
C_0900S*FAC -0.019789 0.016100 -1.229150 0.2190
C_2100S*FAC 0.197589 0.182151 1.084754 0.2780
C_2500S*FAC -0.036984 0.031614 -1.169868 0.2421
ST_CHDGC*FAC -0.427042 0.290532 -1.469865 0.1416
AMSA_AV*FAC 0.595048 0.451329 1.318435 0.1874
IMOU_AV*FAC 0.730147 0.570624 1.279559 0.2007
VMOU_AV*FAC 0.632690 0.475423 1.330793 0.1833
CL_BV*FAC -0.259189 0.228034 -1.136622 0.2557
CL_DNV*FAC -0.312453 0.255567 -1.222587 0.2215
CL_GL*FAC -0.368885 0.276907 -1.332160 0.1828
CL_LR*FAC -0.551158 0.377899 -1.458480 0.1447
CL_NKK*FAC -0.260178 0.225627 -1.153133 0.2489
CL_TLL*FAC -0.966813 0.747382 -1.293599 0.1958
FL_GI*FAC -0.649875 0.604051 -1.075861 0.2820
FL_KY*FAC -0.650373 0.583329 -1.114933 0.2649
LNTON*(-XB)*FAC -0.070640 0.045418 -1.555322 0.1199
R-squared 0.000108 Mean dependent var -8.31E-05
Adjusted R-squared -0.000878 S.D. dependent var 1.005635
S.E. of regression 1.006076 Akaike info criterion 2.851021
Sum squared resid 22577.88 Schwarz criterion 2.859276
Log likelihood -31807.23 Durbin-Watson stat 1.908578
296
Deck Related First Events
Age
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 05/31/06 Time: 11:56
Sample: 1 8357 IF CMOU_AV=0 AND ST5_MAX=0 AND ST6_MAX=0
Included observations: 7771
Variable Coefficient Std. Error t-Statistic Prob.
FAC -0.290573 0.216955 -1.339323 0.1805
LNTON*FAC 0.025859 0.022878 1.130328 0.2584
LNAGE*FAC -0.009030 0.039642 -0.227785 0.8198
C_1200S*FAC 0.004655 0.004407 1.056280 0.2909
C_1900S*FAC -0.014794 0.040770 -0.362858 0.7167
ST_CHDGC*FAC 0.016218 0.034417 0.471229 0.6375
CLWD_D*FAC -0.005494 0.032465 -0.169214 0.8656
IMOU_AV*FAC 0.048569 0.158642 0.306157 0.7595
VMOU_AV*FAC -0.067803 0.085471 -0.793284 0.4276
RS_INSP*FAC 0.007430 0.081195 0.091504 0.9271
CL_BV*FAC -0.032141 0.074734 -0.430070 0.6672
FL_AG*FAC 0.030087 0.064500 0.466470 0.6409
FL_BB*FAC 0.050082 0.142663 0.351048 0.7256
FL_NL*FAC 0.021603 0.077449 0.278936 0.7803
FL_NS*FAC 0.086400 0.084777 1.019147 0.3082
FL_TR*FAC 0.036180 0.066581 0.543392 0.5869
LNAGE*(-XB)*FAC 0.006771 0.013590 0.498216 0.6183
R-squared 0.000889 Mean dependent var -5.87E-15
Adjusted R-squared -0.001173 S.D. dependent var 0.167372
S.E. of regression 0.167470 Akaike info criterion -0.733835
Sum squared resid 217.4713 Schwarz criterion -0.718613
Log likelihood 2868.315 Durbin-Watson stat 1.876014
Tonnage
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 05/31/06 Time: 11:59
Sample: 1 8357 IF CMOU_AV=0 AND ST5_MAX=0 AND ST6_MAX=0
Included observations: 7771
Variable Coefficient Std. Error t-Statistic Prob.
FAC 0.243312 0.259581 0.937325 0.3486
LNTON*FAC -0.000256 0.013579 -0.018828 0.9850
LNAGE*FAC -0.026940 0.025254 -1.066760 0.2861
C_1200S*FAC -0.002612 0.004222 -0.618543 0.5362
C_1900S*FAC -0.065274 0.039404 -1.656540 0.0977
ST_CHDGC*FAC -0.041784 0.035867 -1.164981 0.2441
CLWD_D*FAC -0.060170 0.033416 -1.800643 0.0718
297
IMOU_AV*FAC 0.297172 0.163836 1.813835 0.0697
VMOU_AV*FAC 0.066930 0.091390 0.732356 0.4640
RS_INSP*FAC 0.153095 0.081322 1.882578 0.0598
CL_BV*FAC 0.087826 0.075419 1.164509 0.2443
FL_AG*FAC -0.071357 0.065929 -1.082338 0.2791
FL_BB*FAC -0.170930 0.148684 -1.149616 0.2503
FL_NL*FAC -0.093856 0.079392 -1.182190 0.2372
FL_NS*FAC -0.047023 0.085780 -0.548183 0.5836
FL_TR*FAC -0.072301 0.067188 -1.076098 0.2819
LNTON*(-XB)*FAC -0.009672 0.004756 -2.033650 0.0420
R-squared 0.001389 Mean dependent var -5.87E-15
Adjusted R-squared -0.000671 S.D. dependent var 0.167372
S.E. of regression 0.167428 Akaike info criterion -0.734336
Sum squared resid 217.3624 Schwarz criterion -0.719114
Log likelihood 2870.262 Durbin-Watson stat 1.875529
Engine Related First Events
Age
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 05/30/06 Time: 22:37
Sample: 1 27079
Included observations: 27079
Variable Coefficient Std. Error t-Statistic Prob.
FAC -0.344573 0.960055 -0.358910 0.7197
ST1_MAX*FAC -0.032034 0.122330 -0.261862 0.7934
ST5_MAX*FAC 0.101521 0.297301 0.341476 0.7327
DET_AMSA*FAC 0.037739 0.235424 0.160302 0.8726
DET_USCG*FAC -0.024448 0.153848 -0.158911 0.8737
LNTON*FAC 0.023245 0.069450 0.334698 0.7379
LNAGE*FAC -0.074277 0.182448 -0.407113 0.6839
ST_CHDGC*FAC 0.037491 0.118671 0.315921 0.7521
PMOU_AV*FAC 0.060002 0.210048 0.285660 0.7751
USCG_AV*FAC 0.052137 0.239443 0.217745 0.8276
LNTIMEBW*FAC 0.006542 0.036453 0.179454 0.8576
CL_DNV*FAC 0.022422 0.130787 0.171442 0.8639
CL_GL*FAC 0.017996 0.110230 0.163257 0.8703
CL_LR*FAC 0.025808 0.123779 0.208497 0.8348
FL_BB*FAC 0.075488 0.432535 0.174524 0.8615
FL_PA*FAC -0.025606 0.151126 -0.169438 0.8655
FL_RU*FAC -0.117518 0.471316 -0.249341 0.8031
FL_UK*FAC 0.049978 0.252050 0.198284 0.8428
C_0400S*FAC -0.010052 0.046209 -0.217534 0.8278
C_0700S*FAC 0.003550 0.012477 0.284545 0.7760
C_1400S*FAC 0.005211 0.015244 0.341821 0.7325
C_1700S*FAC 0.003106 0.019009 0.163398 0.8702
298
C_2000S*FAC -0.005667 0.034924 -0.162260 0.8711
LNAGE*(-XB)*FAC 0.026194 0.061081 0.428831 0.6680
R-squared 0.000003 Mean dependent var -0.001900
Adjusted R-squared -0.000847 S.D. dependent var 0.978978
S.E. of regression 0.979392 Akaike info criterion 2.797117
Sum squared resid 25951.41 Schwarz criterion 2.804391
Log likelihood -37847.57 Durbin-Watson stat 1.883404
Tonnage
Dependent Variable: BRMR_Y
Method: Least Squares
Date: 05/30/06 Time: 22:40
Sample: 1 27079
Included observations: 27079
Variable Coefficient Std. Error t-Statistic Prob.
FAC 0.853480 1.717190 0.497021 0.6192
ST1_MAX*FAC 0.066313 0.159747 0.415112 0.6781
ST5_MAX*FAC -0.305586 0.612350 -0.499038 0.6178
DET_AMSA*FAC -0.094761 0.283948 -0.333725 0.7386
DET_USCG*FAC 0.053150 0.175453 0.302928 0.7619
LNTON*FAC 0.005418 0.044641 0.121376 0.9034
LNAGE*FAC -0.035126 0.088393 -0.397381 0.6911
ST_CHDGC*FAC -0.075092 0.164709 -0.455907 0.6485
PMOU_AV*FAC -0.143957 0.317143 -0.453919 0.6499
USCG_AV*FAC -0.163142 0.374440 -0.435696 0.6631
LNTIMEBW*FAC -0.015314 0.044233 -0.346205 0.7292
CL_DNV*FAC -0.056625 0.161666 -0.350260 0.7261
CL_GL*FAC -0.044178 0.132478 -0.333471 0.7388
CL_LR*FAC -0.067887 0.169143 -0.401362 0.6882
FL_BB*FAC -0.156483 0.495927 -0.315536 0.7524
FL_PA*FAC 0.074478 0.199063 0.374145 0.7083
FL_RU*FAC 0.209400 0.554915 0.377355 0.7059
FL_UK*FAC -0.162426 0.383074 -0.424007 0.6716
C_0400S*FAC 0.021103 0.056743 0.371903 0.7100
C_0700S*FAC -0.007265 0.016757 -0.433557 0.6646
C_1400S*FAC -0.010685 0.022442 -0.476127 0.6340
C_1700S*FAC -0.007054 0.022168 -0.318189 0.7503
C_2000S*FAC 0.013391 0.041267 0.324493 0.7456
LNTON*(-XB)*FAC -0.019114 0.036613 -0.522060 0.6016
R-squared 0.000006 Mean dependent var -0.001900
Adjusted R-squared -0.000844 S.D. dependent var 0.978978
S.E. of regression 0.979391 Akaike info criterion 2.797114
Sum squared resid 25951.33 Schwarz criterion 2.804387
Log likelihood -37847.52 Durbin-Watson stat 1.883431
299
Appendix 30: Matching Models: Type III Models
Fire/Explosion
Dependent Variable: DFIRE
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/31/06 Time: 11:07
Sample: 1 6218 IF CMOU_AV=0 AND ST3_MAX=0
Included observations: 5675
Convergence achieved after 8 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
C -7.229885 0.921698 -7.844097 0.0000
ST1_MAX -1.223850 0.242689 -5.042883 0.0000
ST2_MAX -1.228998 0.289421 -4.246398 0.0000
ST4_MAX -1.160351 0.316853 -3.662111 0.0003
LNTON 0.539948 0.092407 5.843167 0.0000
OW_EMN -1.124903 0.285198 -3.944286 0.0001
OW_TMN -0.589246 0.205808 -2.863080 0.0042
OW_IOR -1.492678 0.468662 -3.184981 0.0014
ST_CHDGC 0.439970 0.175683 2.504344 0.0123
IMOU_AV -1.510359 0.607167 -2.487552 0.0129
VMOU_AV -1.095074 0.453933 -2.412411 0.0158
LNTIMEBW 0.150324 0.048607 3.092675 0.0020
FL_BG 1.994803 0.701750 2.842612 0.0045
C_1400S 0.048602 0.016427 2.958722 0.0031
C_1800S 0.284939 0.119094 2.392545 0.0167
Mean dependent var 0.033656 S.D. dependent var 0.180359
S.E. of regression 0.176983 Akaike info criterion 0.273701
Sum squared resid 177.2886 Schwarz criterion 0.291262
Log likelihood -761.6271 Hannan-Quinn criter. 0.279817
Restr. log likelihood -835.5356 Avg. log likelihood -0.134207
LR statistic (14 df) 147.8170 McFadden R-squared 0.088456
Probability(LR stat) 0.000000
Obs with Dep=0 5484 Total obs 5675
Obs with Dep=1 191
300
Dependent Variable: DFIRE
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/31/06 Time: 11:07
Sample: 1 6218 IF CMOU_AV=0 AND ST3_MAX=0
Included observations: 5675
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 0.0002 0.0082 566 564.446 1 2.55361 567 0.94949
2 0.0082 0.0123 560 562.022 8 5.97841 568 0.69087
3 0.0123 0.0157 555 559.019 12 7.98093 567 2.05283
4 0.0157 0.0183 563 558.303 5 9.69703 568 2.31465
5 0.0184 0.0238 554 555.318 13 11.6820 567 0.15182
6 0.0238 0.0293 553 552.916 15 15.0838 568 0.00048
7 0.0293 0.0371 546 548.234 21 18.7657 567 0.27512
8 0.0372 0.0466 550 544.349 18 23.6514 568 1.40903
9 0.0467 0.0679 530 535.587 37 31.4132 567 1.05190
10 0.0679 0.5048 507 503.806 61 64.1940 568 0.17916
Total 5484 5484.00 191 191.000 5675 9.07536
H-L Statistic: 9.0754 Prob. Chi-Sq(8) 0.3360
Andrews Statistic: 13.4956 Prob. Chi-Sq(10) 0.1973
Dependent Variable: DFIRE
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/31/06 Time: 11:07
Sample: 1 6218 IF CMOU_AV=0 AND ST3_MAX=0
Included observations: 5675
Prediction Assessment (success cutoff C = 0.03365)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 3654 64 3718 0 0 0
P(Dep=1)>C 1830 127 1957 5484 191 5675
Total 5484 191 5675 5484 191 5675
Correct 3654 127 3781 0 191 191
% Correct 66.63 66.49 66.63 0.00 100.00 3.37
% Incorrect 33.37 33.51 33.37 100.00 0.00 96.63
Total Gain* 66.63 -33.51 63.26
Percent Gain** 66.63 NA 65.46
301
-4
0
4
8
12
16
1000 2002 3000 4000 5001 6002
Standardized (Pearson) Residuals
Wrecked/Stranded/Grounded
Dependent Variable: DWSG
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/26/06 Time: 18:59
Sample: 1 19131 IF CMOU_AV=0 AND ST5_MAX=0 AND
ST5_MAX=0
Included observations: 18600
Convergence achieved after 8 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
C -7.733199 0.717113 -10.78379 0.0000
ST1_MAX -0.531549 0.164562 -3.230080 0.0012
ST2_MAX -0.600987 0.137824 -4.360537 0.0000
DET_AMSA 0.684428 0.246508 2.776491 0.0055
LNTON 0.414156 0.063435 6.528846 0.0000
LNAGE 0.264708 0.076264 3.470938 0.0005
OWCHD 0.339034 0.119542 2.836110 0.0046
ST_CHDGC 0.604921 0.112651 5.369877 0.0000
AMSA_AV -1.337353 0.352355 -3.795474 0.0001
IMOU_AV -1.296433 0.370949 -3.494909 0.0005
CL_LR 0.396235 0.125791 3.149952 0.0016
FL_BM 2.314809 0.984986 2.350095 0.0188
FL_IM 1.095883 0.437259 2.506258 0.0122
FL_RU -1.188837 0.409279 -2.904710 0.0037
FL_UK 1.386813 0.357793 3.876016 0.0001
C_0700S 0.023768 0.011421 2.081060 0.0374
C_1000S 0.263622 0.086601 3.044095 0.0023
302
C_1400S 0.022311 0.011088 2.012256 0.0442
Mean dependent var 0.026989 S.D. dependent var 0.162056
S.E. of regression 0.160204 Akaike info criterion 0.232718
Sum squared resid 476.9137 Schwarz criterion 0.240296
Log likelihood -2146.278 Hannan-Quinn criter. 0.235206
Restr. log likelihood -2308.547 Avg. log likelihood -0.115391
LR statistic (17 df) 324.5370 McFadden R-squared 0.070290
Probability(LR stat) 0.000000
Obs with Dep=0 18098 Total obs 18600
Obs with Dep=1 502
Dependent Variable: DWSG
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/26/06 Time: 18:59
Sample: 1 19131 IF CMOU_AV=0 AND ST5_MAX=0 AND ST5_MAX=0
Included observations: 18600
Prediction Assessment (success cutoff C = 0.02698)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 12135 179 12314 0 0 0
P(Dep=1)>C 5963 323 6286 18098 502 18600
Total 18098 502 18600 18098 502 18600
Correct 12135 323 12458 0 502 502
% Correct 67.05 64.34 66.98 0.00 100.00 2.70
% Incorrect 32.95 35.66 33.02 100.00 0.00 97.30
Total Gain* 67.05 -35.66 64.28
Percent Gain** 67.05 NA 66.06
303
Dependent Variable: DWSG
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/26/06 Time: 18:59
Sample: 1 19131 IF CMOU_AV=0 AND ST5_MAX=0 AND
ST5_MAX=0
Included observations: 18600
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 0.0003 0.0076 1849 1852.20 11 7.79651 1860 1.32182
2 0.0076 0.0119 1833 1841.11 27 18.8918 1860 3.51564
3 0.0119 0.0139 1846 1835.88 14 24.1208 1860 4.30236
4 0.0139 0.0163 1830 1832.06 30 27.9425 1860 0.15381
5 0.0163 0.0193 1828 1827.05 32 32.9470 1860 0.02771
6 0.0193 0.0236 1825 1820.28 35 39.7250 1860 0.57426
7 0.0236 0.0293 1809 1811.04 51 48.9649 1860 0.08687
8 0.0293 0.0374 1808 1798.52 52 61.4755 1860 1.51041
9 0.0374 0.0535 1783 1776.79 77 83.2099 1860 0.48514
10 0.0535 0.4445 1687 1703.07 173 156.926 1860 1.79813
Total 18098 18098.0 502 502.000 18600 13.7762
H-L Statistic: 13.7762 Prob. Chi-Sq(8) 0.0878
Andrews Statistic: 19.5736 Prob. Chi-Sq(10) 0.0336
-5
0
5
10
15
20
25
30
5000 10000 15000
Standardized (Pearson) Residuals
304
Collision/Contact
Dependent Variable: DCOCO
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/26/06 Time: 17:44
Sample: 1 23254 IF CMOU_AV=0 AND ST6_MAX=0
Included observations: 22329
Convergence achieved after 6 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
C -7.067282 0.664045 -10.64278 0.0000
ST1_MAX -1.621938 0.225563 -7.190606 0.0000
ST2_MAX -1.711995 0.220717 -7.756526 0.0000
ST3_MAX -1.241333 0.245229 -5.061941 0.0000
ST4_MAX -1.341532 0.225556 -5.947671 0.0000
LNTON 0.416932 0.058917 7.076635 0.0000
LNAGE 0.362874 0.055431 6.546457 0.0000
C_0900S 0.031497 0.008968 3.512042 0.0004
C_2100S -0.309480 0.121664 -2.543730 0.0110
C_2500S 0.062421 0.023477 2.658779 0.0078
ST_CHDGC 0.680648 0.097255 6.998552 0.0000
AMSA_AV -0.853452 0.230808 -3.697669 0.0002
IMOU_AV -1.170421 0.334437 -3.499672 0.0005
VMOU_AV -0.915033 0.229946 -3.979340 0.0001
CL_BV 0.394910 0.157638 2.505171 0.0122
CL_DNV 0.470664 0.160140 2.939085 0.0033
CL_GL 0.583936 0.142099 4.109364 0.0000
CL_LR 0.829407 0.133258 6.224067 0.0000
CL_NKK 0.411940 0.154476 2.666698 0.0077
CL_TLL 1.652974 0.428944 3.853593 0.0001
FL_GI 1.052753 0.438834 2.398978 0.0164
FL_KY 1.019645 0.400478 2.546071 0.0109
Mean dependent var 0.030812 S.D. dependent var 0.172812
S.E. of regression 0.170507 Akaike info criterion 0.258502
Sum squared resid 648.5264 Schwarz criterion 0.266397
Log likelihood -2864.040 Hannan-Quinn criter. 0.261071
Restr. log likelihood -3071.429 Avg. log likelihood -0.128265
LR statistic (21 df) 414.7769 McFadden R-squared 0.067522
Probability(LR stat) 0.000000
Obs with Dep=0 21641 Total obs 22329
Obs with Dep=1 688
305
Dependent Variable: DCOCO
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/26/06 Time: 17:44
Sample: 1 23254 IF CMOU_AV=0 AND ST6_MAX=0
Included observations: 22329
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 0.0009 0.0109 2219 2215.81 13 16.1947 2232 0.63482
2 0.0109 0.0138 2198 2205.22 35 27.7845 2233 1.89744
3 0.0138 0.0166 2208 2199.10 25 33.8971 2233 2.37127
4 0.0166 0.0195 2197 2192.80 36 40.1987 2233 0.44659
5 0.0195 0.0228 2181 2185.96 52 47.0371 2233 0.53490
6 0.0228 0.0268 2177 2177.76 56 55.2437 2233 0.01062
7 0.0268 0.0328 2160 2166.71 73 66.2922 2233 0.69950
8 0.0328 0.0420 2152 2150.31 81 82.6904 2233 0.03588
9 0.0420 0.0592 2125 2122.02 108 110.984 2233 0.08444
10 0.0592 0.7821 2024 2025.32 209 207.677 2233 0.00929
Total 21641 21641.0 688 688.000 22329 6.72475
H-L Statistic: 6.7247 Prob. Chi-Sq(8) 0.5666
Andrews Statistic: 8.3372 Prob. Chi-Sq(10) 0.5959
Dependent Variable: DCOCO
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/26/06 Time: 17:44
Sample: 1 23254 IF CMOU_AV=0 AND ST6_MAX=0
Included observations: 22329
Prediction Assessment (success cutoff C = 0.030811)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 14693 265 14958 0 0 0
P(Dep=1)>C 6948 423 7371 21641 688 22329
Total 21641 688 22329 21641 688 22329
Correct 14693 423 15116 0 688 688
% Correct 67.89 61.48 67.70 0.00 100.00 3.08
% Incorrect 32.11 38.52 32.30 100.00 0.00 96.92
Total Gain* 67.89 -38.52 64.62
Percent Gain** 67.89 NA 66.67
306
-5
0
5
10
15
20
25
5000 10000 15000 20000
Standardized (Pearson) Residuals
Deck Related First Events
Dependent Variable: DDECK
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/31/06 Time: 11:33
Sample: 1 8357 IF CMOU_AV=0 AND ST5_MAX=0 AND ST6_MAX=0
Included observations: 7771
Convergence achieved after 8 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
C -8.198026 0.948578 -8.642441 0.0000
LNTON 0.536441 0.073247 7.323709 0.0000
LNAGE 0.360412 0.111891 3.221106 0.0013
C_1200S 0.070160 0.017665 3.971639 0.0001
C_1900S 0.514635 0.174084 2.956239 0.0031
ST_CHDGC 0.540449 0.164759 3.280230 0.0010
CLWD_D 0.532854 0.151507 3.517021 0.0004
IMOU_AV -2.336155 0.581131 -4.020016 0.0001
VMOU_AV -1.206711 0.455764 -2.647665 0.0081
RS_INSP -1.434488 0.353287 -4.060399 0.0000
CL_BV -1.182662 0.355922 -3.322807 0.0009
FL_AG 0.972321 0.329541 2.950534 0.0032
FL_BB 2.191105 0.693409 3.159903 0.0016
FL_NL 1.125001 0.384359 2.926955 0.0034
FL_NS 1.296348 0.393428 3.295003 0.0010
FL_TR 1.040046 0.308477 3.371548 0.0007
Mean dependent var 0.029983 S.D. dependent var 0.170552
S.E. of regression 0.167993 Akaike info criterion 0.251942
Sum squared resid 218.8597 Schwarz criterion 0.266268
307
Log likelihood -962.9202 Hannan-Quinn criter. 0.256853
Restr. log likelihood -1046.629 Avg. log likelihood -0.123912
LR statistic (15 df) 167.4185 McFadden R-squared 0.079980
Probability(LR stat) 0.000000
Obs with Dep=0 7538 Total obs 7771
Obs with Dep=1 233
Dependent Variable: DDECK
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/31/06 Time: 11:33
Sample: 1 8357 IF CMOU_AV=0 AND ST5_MAX=0 AND
ST6_MAX=0
Included observations: 7771
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 0.0002 0.0061 775 774.733 2 2.26670 777 0.03147
2 0.0061 0.0107 768 770.295 9 6.70473 777 0.79259
3 0.0107 0.0140 764 767.428 13 9.57184 777 1.24311
4 0.0140 0.0179 766 764.724 11 12.2764 777 0.13484
5 0.0179 0.0217 767 761.587 10 15.4128 777 1.93941
6 0.0217 0.0267 756 758.216 21 18.7836 777 0.26801
7 0.0267 0.0347 753 753.325 24 23.6751 777 0.00460
8 0.0347 0.0442 744 746.604 33 30.3958 777 0.23220
9 0.0443 0.0618 738 736.172 39 40.8275 777 0.08634
10 0.0619 0.3995 707 704.915 71 73.0854 778 0.06568
Total 7538 7538.00 233 233.000 7771 4.79825
H-L Statistic: 4.7982 Prob. Chi-Sq(8) 0.7789
Andrews Statistic: 9.2049 Prob. Chi-Sq(10) 0.5128
308
Dependent Variable: DDECK
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/31/06 Time: 11:33
Sample: 1 8357 IF CMOU_AV=0 AND ST5_MAX=0 AND ST6_MAX=0
Included observations: 7771
Prediction Assessment (success cutoff C = 0.02998)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 4934 75 5009 0 0 0
P(Dep=1)>C 2604 158 2762 7538 233 7771
Total 7538 233 7771 7538 233 7771
Correct 4934 158 5092 0 233 233
% Correct 65.46 67.81 65.53 0.00 100.00 3.00
% Incorrect 34.54 32.19 34.47 100.00 0.00 97.00
Total Gain* 65.46 -32.19 62.53
Percent Gain** 65.46 NA 64.46
-4
0
4
8
12
16
20
1000 2000 3000 4000 5000 6002 7000 8000
Standardized (Pearson) Residuals
309
Engine Related First Events
Dependent Variable: DENG
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/26/06 Time: 17:03
Sample: 1 27079
Included observations: 27079
Convergence achieved after 8 iterations
QML (Huber/White) standard errors & covariance
Variable Coefficient Std. Error z-Statistic Prob.
C -8.147009 0.576580 -14.12989 0.0000
ST1_MAX -0.414850 0.115153 -3.602603 0.0003
ST5_MAX 1.583319 0.189842 8.340188 0.0000
DET_AMSA 0.520215 0.212531 2.447720 0.0144
DET_USCG -0.316929 0.116563 -2.718949 0.0065
LNTON 0.322429 0.050952 6.328103 0.0000
LNAGE 0.183819 0.058603 3.136709 0.0017
ST_CHDGC 0.466723 0.084339 5.533884 0.0000
PMOU_AV 0.822478 0.157007 5.238488 0.0000
USCG_AV 0.854522 0.207726 4.113693 0.0000
LNTIMEBW 0.093861 0.027306 3.437371 0.0006
CL_DNV 0.327837 0.126642 2.588692 0.0096
CL_GL 0.264476 0.106413 2.485374 0.0129
CL_LR 0.362008 0.110516 3.275622 0.0011
FL_BB 1.033412 0.403574 2.560653 0.0104
FL_PA -0.393960 0.145391 -2.709660 0.0067
FL_RU -1.432597 0.386218 -3.709297 0.0002
FL_UK 0.921571 0.236530 3.896206 0.0001
C_0400S -0.124567 0.039619 -3.144153 0.0017
C_0700S 0.043168 0.009199 4.692785 0.0000
C_1400S 0.064510 0.010054 6.416617 0.0000
C_1700S 0.044865 0.017112 2.621925 0.0087
C_2000S -0.080135 0.031397 -2.552318 0.0107
Mean dependent var 0.030245 S.D. dependent var 0.171263
S.E. of regression 0.168400 Akaike info criterion 0.250030
Sum squared resid 767.2715 Schwarz criterion 0.257000
Log likelihood -3362.283 Hannan-Quinn criter. 0.252278
Restr. log likelihood -3671.702 Avg. log likelihood -0.124166
LR statistic (22 df) 618.8387 McFadden R-squared 0.084271
Probability(LR stat) 0.000000
Obs with Dep=0 26260 Total obs 27079
Obs with Dep=1 819
310
Dependent Variable: DENG
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/26/06 Time: 17:03
Sample: 1 27079
Included observations: 27079
Andrews and Hosmer-Lemeshow Goodness-of-Fit Tests
Grouping based upon predicted risk (randomize ties)
Quantile of Risk Dep=0 Dep=1 Total H-L
Low High Actual Expect Actual Expect Obs Value
1 0.0003 0.0071 2700 2696.19 7 10.8127 2707 1.34978
2 0.0071 0.0127 2682 2680.16 26 27.8400 2708 0.12287
3 0.0127 0.0161 2673 2668.63 35 39.3694 2708 0.49210
4 0.0161 0.0190 2661 2660.75 47 47.2456 2708 0.00130
5 0.0190 0.0223 2651 2652.45 57 55.5454 2708 0.03889
6 0.0223 0.0266 2650 2642.24 58 65.7631 2708 0.93923
7 0.0266 0.0319 2622 2629.35 86 78.6508 2708 0.70726
8 0.0319 0.0405 2616 2611.29 92 96.7052 2708 0.23741
9 0.0405 0.0571 2562 2579.13 146 128.867 2708 2.39163
10 0.0571 0.9919 2443 2439.80 265 268.201 2708 0.04239
Total 26260 26260.0 819 819.000 27079 6.32286
H-L Statistic: 6.3229 Prob. Chi-Sq(8) 0.6111
Andrews Statistic: 22.1248 Prob. Chi-Sq(10) 0.0145
Dependent Variable: DENG
Method: ML – Binary Logit (Quadratic hill climbing)
Date: 05/26/06 Time: 17:03
Sample: 1 27079
Included observations: 27079
Prediction Assessment (success cutoff C = 0.031)
Estimated Equation Constant Probability
Dep=0 Dep=1 Total Dep=0 Dep=1 Total
P(Dep=1)<=C 18308 303 18611 26260 819 27079
P(Dep=1)>C 7952 516 8468 0 0 0
Total 26260 819 27079 26260 819 27079
Correct 18308 516 18824 26260 0 26260
% Correct 69.72 63.00 69.52 100.00 0.00 96.98
% Incorrect 30.28 37.00 30.48 0.00 100.00 3.02
Total Gain* -30.28 63.00 -27.46
Percent Gain** NA 63.00 -907.94
311
-4
0
4
8
12
16
20
24
5000 10000 15000 20000 25000
Standardized (Pearson) Residuals
312
ERASMUS RESEARCH INSTITUTE OF MANAGEMENT (ERIM)
ERIM PH.D. SERIES
RESEARCH IN MANAGEMENT
ERIM Electronic Series Portal: http://hdl.handle.net/1765/1
Appelman, J.H., Governance of Global Interorganizational Tourism Networks:
Changing Forms of Co-ordination between the Travel Agency and Aviation
Sector, Promotors: Prof. dr. F.M. Go & Prof. dr. B. Nooteboom, EPS-2004-
036-MKT, ISBN 90-5892-060-7, http://hdl.handle.net/1765/1199
Assen, M.F. van, Empirical Studies in Discrete Parts Manufacturing
Management, Promotors: Prof. dr. S.L. van de Velde & Prof. dr. W.H.M. Zijm,
EPS-2005-056-LIS, ISBN 90-5892-085-2, http://hdl.handle.net/1765/6767
Berens, G., Corporate Branding: The Development of Corporate Associations
and their Influence on Stakeholder Reactions, Promotor: Prof. dr. C. B. M. van
Riel, EPS-2004-039-ORG, ISBN 90 –5892–065–8,
http://hdl.handle.net/1765/1273
Berghe, D.A.F., Working Across Borders: Multinational Enterprises and the
Internationalization of Employment, Promotors: Prof. dr. R.J.M. van Tulder &
Prof. dr. E.J.J. Schenk, EPS-2003-029-ORG, ISBN 90-5892-05-34,
http://hdl.handle.net/1765/1041
Bijman, W.J.J., Essays on Agricultural Co-operatives: Governance Structure
in Fruit and Vegetable Chains, Promotor: Prof. dr. G.W.J. Hendrikse, EPS-
2002-015-ORG, ISBN: 90-5892-024-0, http://hdl.handle.net/1765/867
Boer, N.I., Knowledge Sharing within Organizations: A situated and relational
Perspective, Promotor: Prof. dr. K. Kumar, EPS-2005-060-LIS, ISBN: 90-
5892-086-0, http://hdl.handle.net/1765/6770
Boer, C.A., Distributed Simulation in Industry, Promotors: Prof. dr. A. de
Bruin & Prof. dr. ir. A. Verbraeck, EPS-2005-065-LIS, ISBN: 90-5892-093-3,
http://hdl.handle.net/1765/6925
Brito, M.P. de, Managing Reverse Logistics or Reversing Logistics
Management? Promotors: Prof. dr. ir. R. Dekker & Prof. dr. M. B. M. de
Koster, EPS-2004-035-LIS, ISBN: 90-5892-058-5,
http://hdl.handle.net/1765/1132
Brohm, R., Polycentric Order in Organizations: a dialogue between Michael
Polanyi and IT-consultants on knowledge, morality, and organization,
Promotors: Prof. dr. G. W. J. Hendrikse & Prof. dr. H. K. Letiche, EPS-2005-
063-ORG, ISBN: 90-5892-095-X, http://hdl.handle.net/1765/6911
Campbell, R.A.J., Rethinking Risk in International Financial Markets,
Promotor: Prof. dr. C.G. Koedijk, EPS-2001-005-F&A, ISBN: 90-5892-008-9,
http://hdl.handle.net/1765/306
313
Chen, Y., Labour Flexibility in China’s Companies: An Empirical Study,
Promotors: Prof. dr. A. Buitendam & Prof. dr. B. Krug, EPS-2001-006-ORG,
ISBN: 90-5892-012-7, http://hdl.handle.net/1765/307
Daniševská, P., Empirical Studies on Financial Intermediation and Corporate
Policies, Promotor: Prof. dr. C.G. Koedijk, EPS-2004-044-F&A, ISBN 90–
5892–070–4, http://hdl.handle.net/1765/1518
Delporte-Vermeiren, D.J.E., Improving the Flexibility and Profitability of ICTenabled
Business Networks: An Assessment Method and Tool, Promotors:
Prof.mr. dr. P.H.M. Vervest & Prof. dr. ir. H.W.G.M. van Heck, EPS-2003-
020-LIS, ISBN: 90-5892-040-2, http://hdl.handle.net/1765/359
Dijksterhuis, M., Organizational Dynamics of Cognition and Action in the
Changing Dutch and US Banking Industries, Promotors: Prof. dr. ir. F.A.J. van
den Bosch & Prof. dr. H.W. Volberda, EPS-2003-026-STR, ISBN: 90-5892-
048-8, http://hdl.handle.net/1765/1037
Fenema, P.C. van, Coordination and Control of Globally Distributed Software
Projects, Promotor: Prof. dr. K. Kumar, EPS-2002-019-LIS, ISBN: 90-5892-
030-5, http://hdl.handle.net/1765/360
Fleischmann, M., Quantitative Models for Reverse Logistics, Promoters: Prof.
dr. ir. J.A.E.E. van Nunen & Prof. dr. ir. R. Dekker, EPS-2000-002-LIS, ISBN:
35-4041-711-7, http://hdl.handle.net/1765/1044
Flier, B., Strategic Renewal of European Financial Incumbents: Coevolution of
Environmental Selection, Institutional Effects, and Managerial Intentionality,
Promotors: Prof. dr. ir. F.A.J. van den Bosch & Prof. dr. H.W. Volberda,
EPS-2003-033-STR, ISBN: 90–5892–055–0, http://hdl.handle.net/1765/1071
Fok, D., Advanced Econometric Marketing Models, Promotor: Prof. dr.
P.H.B.F. Franses, EPS-2003-027-MKT, ISBN: 90-5892-049-6,
http://hdl.handle.net/1765/1035
Ganzaroli , A., Creating Trust between Local and Global Systems, Promotors:
Prof. dr. K. Kumar & Prof. dr. R.M. Lee, EPS-2002-018-LIS, ISBN: 90-5892-
031-3, http://hdl.handle.net/1765/361
Gilsing, V.A., Exploration, Exploitation and Co-evolution in Innovation
Networks, Promotors: Prof. dr. B. Nooteboom & Prof. dr. J.P.M.
Groenewegen, EPS-2003-032-ORG, ISBN 90-5892-054-2,
http://hdl.handle.net/1765/1040
Govers, R., Virtual Tourism Destination Image: Glocal identities constructed,
perceived and experienced, Promotors: Prof. dr. F.M. Go & Prof. dr. K.
Kumar, EPS-2005-069-MKT, ISBN 90-5892-107-7,
http://hdl.handle.net/1765/6981
Graaf, G. de, Tractable Morality: Customer Discourses of Bankers,
Veterinarians and Charity Workers, Promotors: Prof. dr. F. Leijnse & Prof. dr.
T. van Willigenburg, EPS-2003-031-ORG, ISBN: 90–5892–051–8,
http://hdl.handle.net/1765/1038
314
Hagemeijer, R.E., The Unmasking of the Other, Promotors: Prof. dr. S.J.
Magala, Prof. dr. H.K. Letiche, EPS-2005-068-ORG, ISBN: 90–5892–097–6,
http://hdl.handle.net/1765/6963
Hartigh, E. den, Increasing Returns and Firm Performance: An Empirical
Study, Promotor: Prof. dr. H.R. Commandeur, EPS-2005-067-STR, ISBN: 90–
5892–098–4, http://hdl.handle.net/1765/6939
Hermans. J.M., ICT in Information Services, Use and deployment of the Dutch
securities trade, 1860-1970. Promotor: Prof. dr. drs. F.H.A. Janszen, EPS-
2004-046-ORG, ISBN 90-5892-072-0, http://hdl.handle.net/1765/1793
Heugens, P.M.A.R., Strategic Issues Management: Implications for Corporate
Performance, Promotors: Prof. dr. ir. F.A.J. van den Bosch & Prof. dr. C.B.M.
van Riel, EPS-2001-007-STR, ISBN: 90-5892-009-7,
http://hdl.handle.net/1765/358
Hooghiemstra, R., The Construction of Reality, Promotors: Prof. dr. L.G. van
der Tas RA & Prof. dr. A.Th.H. Pruyn, EPS-2003-025-F&A, ISBN: 90-5892-
047-X, http://hdl.handle.net/1765/871
Iastrebova, K, Manager’s Information Overload: The Impact of Coping
Strategies on Decision-Making Performance, Promotor: Prof. dr. H.G. van
Dissel, EPS-2006-077-LIS, ISBN 90-5892-111-5, http://hdl.handle.net/1765/
Jansen, J.J.P., Ambidextrous Organizations, Promotors: Prof. dr. ir. F.A.J. Van
den Bosch & Prof. dr. H.W. Volberda, EPS-2005-055-STR, ISBN 90-5892-
081-X, http://hdl.handle.net/1765/
Jong, C. de, Dealing with Derivatives: Studies on the Role, Informational
Content and Pricing of Financial Derivatives, Promotor: Prof. dr. C.G.
Koedijk, EPS-2003-023-F&A, ISBN: 90-5892-043-7,
http://hdl.handle.net/1765/1043
Keizer, A.B., The Changing Logic of Japanese Employment Practices: A Firm-
Level Analysis of Four Industries. Promotors: Prof. dr. J.A. Stam & Prof. dr.
J.P.M. Groenewegen, EPS-2005-057-ORG, ISBN: 90-5892-087-9,
http://hdl.handle.net/1765/6667
Kippers, J., Empirical Studies on Cash Payments, Promotor: Prof. dr.
Ph.H.B.F. Franses, EPS-2004-043-F&A. ISBN 90-5892-069-0,
http://hdl.handle.net/1765/1520
Koppius, O.R., Information Architecture and Electronic Market Performance,
Promotors: Prof. dr. P.H.M. Vervest & Prof. dr. ir. H.W.G.M. van Heck, EPS-
2002-013-LIS, ISBN: 90-5892-023-2, http://hdl.handle.net/1765/921
Kotlarsky, J., Management of Globally Distributed Component-Based Software
Development Projects, Promotor: Prof. dr. K. Kumar, EPS-2005-059-LIS,
ISBN: 90-5892-088-7, http://hdl.handle.net/1765/6772
Kuilman, J., The re-emergence of foreign banks in Shanghai: An ecological
analysis, Promotor: Prof. dr. B. Krug, EPS-2005-066-ORG, ISBN: 90-5892-
096-8, http://hdl.handle.net/1765/6926
315
Langen, P.W. de, The Performance of Seaport Clusters: A Framework to
Analyze Cluster Performance and an Application to the Seaport Clusters of
Durban, Rotterdam and the Lower Mississippi, Promotors: Prof. dr. B.
Nooteboom & Prof. drs. H.W.H. Welters, EPS-2004-034-LIS, ISBN: 90-5892-
056-9, http://hdl.handle.net/1765/1133
Le Anh, T., Intelligent Control of Vehicle-Based Internal Transport Systems,
Promotors: Prof. dr. M.B.M. de Koster & Prof. dr. ir. R. Dekker, EPS-2005-
051-LIS, ISBN 90-5892-079-8, http://hdl.handle.net/1765/6554
Le-Duc, T., Design and control of efficient order picking processes, Promotor:
Prof. dr. M.B.M. de Koster, EPS-2005-064-LIS, ISBN 90-5892-094-1,
http://hdl.handle.net/1765/6910
Lentink, R.M., Algorithmic Decision Support for Shunt Planning, Promotors:
Prof. dr. L.G. Kroon & Prof. dr. ir. J.A.E.E. van Nunen, EPS-2006-073-LIS,
ISBN 90-5892-104-2, http://hdl.handle.net/1765/
Liang, G., New Competition: Foreign Direct Investment And Industrial
Development In China, Promotor: Prof. dr. R.J.M. van Tulder, EPS-2004-047-
ORG, ISBN 90–5892–073–9, http://hdl.handle.net/1765/1795
Loef, J., Incongruity between Ads and Consumer Expectations of Advertising,
Promotors: Prof. dr. W.F. van Raaij & Prof. dr. G. Antonides, EPS-2002-017-
MKT, ISBN: 90-5892-028-3, http://hdl.handle.net/1765/869
Londoño, M. del Pilar, Institutional Arrangements that Affect Free Trade
Agreements: Economic Rationality Versus Interest Groups, Promotors: Prof.
dr. H.E. Haralambides & Prof. dr. J.F. Francois, EPS-2006-078-LIS, ISBN: 90-
5892-108-5, http://hdl.handle.net/1765
Maeseneire, W., de, Essays on Firm Valuation and Value Appropriation,
Promotor: Prof. dr. J.T.J. Smit, EPS-2005-053-F&A, ISBN 90-5892-082-8,
Mandele, L.M., van der, Leadership and the Inflection Point: A Longitudinal
Perspective, Promotors: Prof. dr. H.W. Volberda, Prof. dr. H.R. Commandeur,
EPS-2004-042-STR, ISBN 90–5892–067–4, http://hdl.handle.net/1765/1302
Meer, J.R. van der, Operational Control of Internal Transport, Promotors:
Prof. dr. M.B.M. de Koster & Prof. dr. ir. R. Dekker, EPS-2000-001-LIS,
ISBN: 90-5892-004-6, http://hdl.handle.net/1765/859
Mentink, A., Essays on Corporate Bonds, Promotor: Prof. dr. A.C.F. Vorst,
EPS-2005-070-F&A, ISBN: 90-5892-100-X, http://hdl.handle.net/1765/7121
Miltenburg, P.R., Effects of Modular Sourcing on Manufacturing Flexibility in
the Automotive Industry: A Study among German OEMs, Promotors: Prof. dr.
J. Paauwe & Prof. dr. H.R. Commandeur, EPS-2003-030-ORG, ISBN: 90-
5892-052-6, http://hdl.handle.net/1765/1039
Moerman, G.A., EmpiricalStudies on Asset Pricing and Banking in the Euro
Area, Promotors: Prof. dr. C.G. Koedijk, EPS-2005-058-F&A, ISBN: 90-5892-
090-9, http://hdl.handle.net/1765/6666
316
Mol, M.M., Outsourcing, Supplier-relations and Internationalisation: Global
Source Strategy as a Chinese Puzzle, Promotor: Prof. dr. R.J.M. van Tulder,
EPS-2001-010-ORG, ISBN: 90-5892-014-3, http://hdl.handle.net/1765/355
Mulder, A., Government Dilemmas in the Private Provision of Public Goods,
Promotor: Prof. dr. R.J.M. van Tulder, EPS-2004-045-ORG, ISBN: 90-5892-
071-2, http://hdl.handle.net/1765
Muller, A.R., The Rise of Regionalism: Core Company Strategies Under The
Second Wave of Integration, Promotor: Prof. dr. R.J.M. van Tulder, EPS-2004-
038-ORG, ISBN 90–5892–062–3, http://hdl.handle.net/1765/1272
Oosterhout, J., van, The Quest for Legitimacy: On Authority and Responsibility
in Governance, Promotors: Prof. dr. T. van Willigenburg & Prof.mr. H.R. van
Gunsteren, EPS-2002-012-ORG, ISBN: 90-5892-022-4,
http://hdl.handle.net/1765/362
Pak, K., Revenue Management: New Features and Models, Promotor: Prof. dr.
ir. R. Dekker, EPS-2005-061-LIS, ISBN: 90-5892-092-5,
Peeters, L.W.P., Cyclic Railway Timetable Optimization, Promotors: Prof. dr.
L.G. Kroon & Prof. dr. ir. J.A.E.E. van Nunen, EPS-2003-022-LIS, ISBN: 90-
5892-042-9, http://hdl.handle.net/1765/429
Pietersz, R., Pricing Models for Bermudan-style Interest Rate Derivatives,
Promotors: Prof. dr. A.A.J. Pelsser & Prof. dr. A.C.F. Vorst, EPS-2005-071-
F&A, ISBN 90-5892-099-2, http://hdl.handle.net/1765/7122
Popova, V., Knowledge Discovery and Monotonicity, Promotor: Prof. dr. A. de
Bruin, EPS-2004-037-LIS, ISBN 90-5892-061-5,
http://hdl.handle.net/1765/1201
Pouchkarev, I., Performance Assessment of Constrained Portfolios, Promotors:
Prof. dr. J. Spronk & Dr. W.G.P.M. Hallerbach, EPS-2005-052-F&A, ISBN
90-5892-083-6, http://hdl.handle.net/1765/6731
Puvanasvari Ratnasingam, P., Interorganizational Trust in Business to
Business E-Commerce, Promotors: Prof. dr. K. Kumar & Prof. dr. H.G. van
Dissel, EPS-2001-009-LIS, ISBN: 90-5892-017-8,
http://hdl.handle.net/1765/356
Romero Morales, D., Optimization Problems in Supply Chain Management,
Promotors: Prof. dr. ir. J.A.E.E. van Nunen & Dr. H.E. Romeijn, EPS-2000-
003-LIS, ISBN: 90-9014078-6, http://hdl.handle.net/1765/865
Roodbergen , K.J., Layout and Routing Methods for Warehouses, Promotors:
Prof. dr. M.B.M. de Koster & Prof. dr. ir. J.A.E.E. van Nunen, EPS-2001-004-
LIS, ISBN: 90-5892-005-4, http://hdl.handle.net/1765/861
Schweizer, T.S., An Individual Psychology of Novelty-Seeking, Creativity and
Innovation, Promotor: Prof. dr. R.J.M. van Tulder. EPS-2004-048-ORG,
ISBN: 90-5892-07-71, http://hdl.handle.net/1765/1818
317
Six, F.E., Trust and Trouble: Building Interpersonal Trust Within
Organizations, Promotors: Prof. dr. B. Nooteboom & Prof. dr. A.M. Sorge,
EPS-2004-040-ORG, ISBN 90–5892–064–X, http://hdl.handle.net/1765/1271
Slager, A.M.H., Banking across Borders, Promotors: Prof. dr. D.M.N. van
Wensveen & Prof. dr. R.J.M. van Tulder, EPS-2004-041-ORG, ISBN 90-5892-
066–6, http://hdl.handle.net/1765/1301
Sloot, L., Understanding Consumer Reactions to Assortment Unavailability,
Promotors: Prof. dr. H.R. Commandeur , Prof. dr. E. Peelen & Prof. dr. P.C.
Verhoef, EPS-2006-074-MKT, ISBN 90-5892-102–6,
http://hdl.handle.net/1765/
Smit, W., Market Information Sharing in Channel Relationships: Its Nature,
Antecedents and Consequences, Promotors: Prof. dr. H.R. Commandeur , Prof.
dr. ir. G.H. van Bruggen & Prof. dr. ir. B. Wierenga, EPS-2006-076-MKT,
ISBN 90-5892-106-9, http://hdl.handle.net/1765/
Speklé, R.F., Beyond Generics: A closer look at Hybrid and Hierarchical
Governance, Promotor: Prof. dr. M.A. van Hoepen RA, EPS-2001-008-F&A,
ISBN: 90-5892-011-9, http://hdl.handle.net/1765/357
Teunter, L.H., Analysis of Sales Promotion Effects on Household Purchase
Behavior, Promotors: Prof. dr. ir. B. Wierenga & Prof. dr. T. Kloek, EPS-
2002-016-ORG, ISBN: 90-5892-029-1, http://hdl.handle.net/1765/868
Valck, K. de, Virtual Communities of Consumption: Networks of Consumer
Knowledge and Companionship, Promotors: Prof. dr. ir. G.H. van Bruggen, &
Prof. dr. ir. B. Wierenga, EPS-2005-050-MKT, ISBN 90-5892-078-X,
http://hdl.handle.net/1765/6663
Verheul, I., Is there a (fe)male approach? Understanding gender differences
in entrepreneurship, Prof. dr. A.R. Thurik, EPS-2005-054-ORG, ISBN 90-
5892-080-1, http://hdl.handle.net/1765/2005
Vis, I.F.A., Planning and Control Concepts for Material Handling Systems,
Promotors: Prof. dr. M.B.M. de Koster & Prof. dr. ir. R. Dekker, EPS-2002-
014-LIS, ISBN: 90-5892-021-6, http://hdl.handle.net/1765/866
Vlaar, P.W.L., Making Sense of Formalization in Interorganizational
Relationships: Beyond Coordination and Control, Promotors: Prof. dr. ir.
F.A.J. Van den Bosch & Prof. dr. H.W. Volberda, EPS-2006-075-STR, ISBN
90-5892-103-4, http://hdl.handle.net/1765
Vliet, P. van, Downside Risk and Empirical Asset Pricing, Promotor: Prof. dr.
G.T. Post, EPS-2004-049-F&A ISBN 90-5892-07-55,
http://hdl.handle.net/1765/1819
Vries-van Ketel E. de, How Assortment Variety Affects Assortment
Attractiveness:
A Consumer Perspective, Promotors: Prof. dr. G.H. van Bruggen, Prof.dr.ir.
A.Smidts, EPS-2006-072-MKT, ISBN 90-5892-101-8,
http://hdl.handle.net/1765/
318
Vromans, M.J.C.M., Reliability of Railway Systems, Promotors: Prof. dr. L.G.
Kroon, Prof. dr. ir. R. Dekker & Prof. dr. ir. J.A.E.E. van Nunen, EPS-2005-
062-LIS, ISBN: 90-5892-089-5, http://hdl.handle.net/1765/6773
Waal, T. de, Processing of Erroneous and Unsafe Data, Promotor: Prof. dr. ir.
R. Dekker, EPS-2003-024-LIS, ISBN: 90-5892-045-3,
http://hdl.handle.net/1765/870
Wielemaker, M.W., Managing Initiatives: A Synthesis of the Conditioning and
Knowledge-Creating View, Promotors: Prof. dr. H.W. Volberda & Prof. dr.
C.W.F. Baden-Fuller, EPS-2003-28-STR, ISBN 90-5892-050-X,
http://hdl.handle.net/1765/1036
Wijk, R.A.J.L. van, Organizing Knowledge in Internal Networks: A Multilevel
Study, Promotor: Prof. dr. ir. F.A.J. van den Bosch, EPS-2003-021-STR, ISBN:
90-5892-039-9, http://hdl.handle.net/1765/347
Wolters, M.J.J., The Business of Modularity and the Modularity of Business,
Promotors: Prof. mr. dr. P.H.M. Vervest & Prof. dr. ir. H.W.G.M. van Heck,
EPS-2002-011-LIS, ISBN: 90-5892-020-8, http://hdl.handle.net/1765/920