Review of Window Shopping: A Study of Desktop Window Switching.

ORIGINAL ARTICLE:

Warr, A., Chi, Ed H. , Harris, H., Kuscher, A., Chen, J., Flack, R. and Jitkoff. N. (2016) Window Shopping: A Study of Desktop Window Switching. In Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems (CHI ’16). ACM, New York, NY, USA. pp 3335-3338.

ABSTRACT

This is a review of a research article by Warr et al (2016), of Google Inc, Window shopping: A Study of Desktop Window Switching, which is an investigation into the efficiency of window systems. Having multiple windows open is commonplace but users often have problems switching between windows. This report presents an experiment used to evaluate window switching interfaces that are being considered for a new operating system. Its focus is an understanding of the effects of switching on user performances in terms of time to switch and errors in selection. The experiment compares three different styles of window switching interfaces and their effects on user performances. The interfaces chosen were the Cards interface, the Exposé interface and the Mosaic interface.

The results from the experiment suggest that the Mosaic interface allows faster selection than the Cards interface and is less erroneous than the Exposé interface.

SUMMARY OF ARTICLE

The report outlines some windows research since the 1980’s to the present time. Early research (Bly el al. 1986) showed that even though tiled window systems were faster than overlapping windows, users still preferred overlapping windows. Their conclusion being that it is imperative to create window systems that are both efficient and meet their user preferences. Research by Card and Henderson (1987) argued for designs that reduce the time and cognitive load of switching tasks. Hutchings et al’s (2004) study showed that users often have 8 or more windows open at any time with a median activation time of 3.8 seconds. Other research looked at arranging and grouping windows. These research works inspired Warr et al to investigate three window switching interfaces. The chosen interfaces were

The Cards Interface: These resemble playing cards with small conceptual representation of the window and small amounts of information. They are spread like a stack of cards with the most recent window showing at the front.

The Exposé interface: This arranges the windows in a map, based on their size and original position.

The Mosaic Interface: In this interface the windows are paced in order of most recent left to right in a grid form with the windows scaled to size of grid cell.

The report details the design and procedures used in the experiment. They used a 3X3 mixed factorial design for the experiment.  This being the three user interface types by three different number of windows open at a time namely 3, 6 and 9 windows.

The experiment was carried out to prove two hypothesises about window switching.

Hypothesis 1: That the Mosaic interface allows windows to be selected faster than the Cards and Exposé interfaces

Hypothesis 2: The Mosaic interface would have less errors in selection than the Cards and Exposé interfaces.

The experiment was carried out in a lab using 42 participants (27 female and 15 male) ranging in age from 18-60. The participants were volunteers and were selected on the basis that they used laptops as their main computing device, hadn’t used any of the selected interfaces previously and on average had more than one window open at any one time. They developed an application which prototyped the interfaces and allowed them to control and gather the data.

Trials were carried out using each of the interfaces and selected number of windows. The participant were instructed to switch to predefined windows and the time taken to select the correct window and any errors in selections were recorded. The details of the conditions used are defined.

The article displays the results for selection time and errors and shows them in table and graphical format for each interface and number of windows. Using a number of statistical tests the results allowed for partial acceptance of both hypothesis. They suggested that the Mosaic interface allows quicker window selection than the Cards interface that the Mosaic interface has fewer selection errors than the Exposé interface.

In the discussion they imply that the Mosaic interface was faster because the need to scroll in the card selection made it inefficient especially as the number of windows increases. Also in the Exposé interface as the number of open windows increases the size of the thumbnails in the Exposé interface decreases making them more difficult to distinguish. There wasn’t much difference in number of erroneous selections between the interfaces except the Exposé interface at the greater number of open windows. They surmise that was due to the fact the thumbnails were small and hard to distinguish. The limitations applied in conducting the experiment makes this a worst case scenario.

They conclude that the Mosaic interface enables fast, less error prone window selection.

DISCUSSION

The article is well laid out. It summarises some similar research previously carried out in window switching although they say that this is the first comparative study of these specific window overview schemes (p 3336). Their reference material is fairly old. The most recent being a study of their own presented at the CHI conference in 2011. The aim of the experiment, investigating the efficiency of three window interfaces, and the reason for it are well defined. It presents the experiment in distinct sections outlining the design of the experiment, how the participants were chosen, the materials used, the procedures, their interpretation of the results and conclusion.

They conclude that the Mosaic interface is the faster and the least error prone of the three chosen interfaces. This is internally valid as it was in a controlled environment however how transferable this is in the real world where prior knowledge and familiarity may affect the results is not measured. It also ignores user preferences that Bly et al considered important. There is no measure of the user experience in this experiment.

Purposive sampling of participants appears to have been used. They did not include a group with prior knowledge of the interfaces and were careful throughout the experiment to alleviate any learning effects. This may have been for quick and economic reasons even though participants received a gift card. The sample size may be small because of the criteria used to select the participants and there is no explanation as to why there are almost twice as many women. The data collected from this group can’t really be used to generalise about the interfaces. The results may be biased towards non experienced users. They could however be used as a control reference for any future research to externally validate these interfaces with other users.

The authors of the article are affiliated to Google inc and Warr, Chi and Harris have published many articles in areas relating to computer interface design. Their previous works has been cited many times. This paper was presented to the CHI 16 Conference on Human Factors in Computing Systems. This is a top conference for HCI where the future of how people interact with technology is discussed. It is a conference for top researchers, designers and corporations. This was an ideal place to present this research paper and may prompt further research in this area.

A good knowledge of statistics is required to understand the results as presented in the article. The researchers appear to be confident in their methods and choice of statistical tests but without a knowledge of statistics it is difficult to know whether their choice of statistical tests are biased.

In their discussion they explore the reasons why the Mosaic interface may have been faster than the Cards interface. The put forward three credible explanations for this. These factors could be explored in any future studies.

They acknowledge the fact that the experiment may not reflect real life scenarios.

CONCLUSION

Warr, Chi, Harris, Kuscher, Chen, Flack, and Jitkoff presented a detailed experiment in window switching interfaces that were being considered for a new operating system. They suggest that the Mosaic interface allows faster selection than the Cards interface and less erroneous than the Exposé interface from the results they obtained. The results may not transfer to the general population and does not study the ergonomics of the interfaces and ignores user preferences. However they may inspire further research in the area of window switching.

BIBLIOGRAPHY

Bly, S. A. and Rosenberg, J. K. 1986. A comparison of tiled and overlapping windows. Proc. CHI’86, ACM Press, 101-106.

Card S. K. and Henderson, A. 1987. A Multiple, Virtual-Workspace Interface to Support User Task Switching. Proc CHI’87, ACM Press, 53-59.

Hutchings, D. R., Smith, G., Meyers, B., Czerwinski, M., Robertson, G. 2004. Display space usage and window management operation comparisons between single monitor and multiple monitor users. Proc. AVI’04, ACM Press, 32-39.