Trackmouse Trackball in Pie Menu Use: Data on Accuracy Benoît MARTIN

Poika ISOKOSKI

University Paul Verlaine - Metz Ile du Saulcy 57045 Metz Cedex 1, France [email protected]

University of Tampere Department of Computer Sciences 33014 University of Tampere, Finland [email protected]

RESUME

Nous avons mené deux expérimentations afin d’aider à la conception de menus en anneau utilisables avec la partie trackball du TrackMouse qui combine un trackball et une souris. Dans la première expérimentation, nous avons mesuré la précision de sélection d’une cible ponctuelle dans un menu en anneau (valeur angulaire). Dans la seconde expérimentation, nous avons mesuré la performance de sélection d’un item (secteur angulaire) dans un menu en anneau conçu à partir des résultats de la première expérimentation. Les résultats montrent qu’un menu en anneau à six items peut être utilisé avec un taux d’erreur proche de 3%. MOTS-CLEFS : Trackball, Trackmouse, menu en

anneau. ABSTRACT

We conducted two experiments to guide the design of a pie menu to be used with the trackball component of a device that combines a trackball and a mouse. In the first experiment we measured the accuracy of pie menu specific target selection (angular error). In the second experiment we measured the error rate in menu item selection in a pie menu that was designed based on the findings of the first experiment. The results show that a six item menu can be used at about 3% error rate. CATEGORIES AND SUBJECT DESCRIPTORS: H5.2.

[Information interfaces and presentation]: User Interfaces --- Input devices and strategies, Evaluation, Interaction styles. GENERAL TERMS: Experimentation, Human Factors,

stems from our earlier work on a device called Trackmouse that combines a trackball and a mouse [1,3]. In this paper we report ongoing work on the use of Trackmouse in pie menu selection. We focus on the accuracy of the menu selection movements. There are other aspects of the interaction that need similar study. However, these are beyond the scope of this paper. Traditional trackball casings are stationary whereas a Trackmouse must slide easily in order to be usable as a mouse. This was one of the reasons why we considered it necessary to do low-level experiments on accuracy. Previous work on pie menus has shown that many designs are operable, but error rate tends to increase as the number of menu items increases [2]. In addition symmetric menu designs seem to be easier to learn. For example 8-item menu is preferable to a 7-item menu because of symmetry [2]. We conducted two experiments. The first measured the angular accuracy of the movements needed for menu selection. Based on these results, a menu was built and used in the second experiment to validate the results in a more realistic usage situation. EXPERIMENT 1 Participants

Ten participants (9 male and 1 female) were recruited from the students of our University. The average age of the participants was 24 years (varying from 23 to 25). Only right-handed participants were recruited because our Trackmouse prototype was not suitable for lefthanded use.

Performance. Apparatus KEYWORDS : Trackball, Trackmouse, pie menu. INTRODUCTION

Our interest in the use of trackballs in menu selection

The Trackmouse prototype used in the experiment was built using a Logitech TrackMan trackball product. The ball was operated with the thumb. The user’s palm and fingers rested on the body of the device allowing it to be moved like a mouse. The bottom of the device had been modified by removing the rubber pads to make it slide better, and by carving a hole for the mouse optics. Inside it had the electronics of a Mitsumi FreeStyle mouse in addition to the original trackball electronics. To be able to dissociate the trackball from the Microsoft Windows system cursor and still receive data from it, we

used a modified version of the CPN mouse driver originally developed by Westergaard [5]. The task of experiment 1 is shown in Figure 2. The user’s task was to move a pointer associated with the ball as close to the indicated direction as possible as fast as possible. The direction to aim for was chosen with a random number generator from angles between 0 and 360 degrees. It was indicated with the short black mark that appeared on the circumference of a circle drawn around the mouse cursor. The movement of the ball was shown as a green increasing radius that connected the current position of the ball to the center of the circle. When this radius crossed the circle, the angle of intersection was recorded and the task ended. The next task was shown after a button press. The bar above the circle in Figure 1 showed session statistics; the average angle between the target and the hits (green bar), the angle between the previous target and the previous hit (red line) and the largest angle deviation between a target and hit (black line). The task reflected the menu design that we had in mind. To be able to use the ball efficiently, the user should rest his thumb on the ball while using the mouse functionality of the Trackmouse. Therefore, the system must allow for small movements of the trackball without causing menu selections or other events. On the other hand, a menu selection should be as easy as possible. Because the ball has no other use, a button press is an unnecessary complication. Therefore, it may be desirable to activate the menu item instantly when it is touched by the “cursor” controlled by the ball.

Last angle deviation

many trials as possible. The software terminated the block automatically after 5 minutes. The participants were capable of operating the experimental software on their own after a short briefing. They were left to schedule the blocks according to their preference. Usually several blocks were completed in one session. Results and Discussion

The dependent variables in this experiment were task completion time and angular error. Task completion time was measured from the display of a task to the first crossing of the circle. The angular error was the angle between the target and the circle crossing point entered by the user. For a visual inspection of the results we computed the average angle between the crossing points and the associated targets. This was done by first dividing the data based on the target1 location into 10 degree sectors and then computing the mean deviation in the trials in each sector (i.e. the mean of the absolute values of the angles between the ideal vector and the vector entered by the user). The results for the first and tenth block are shown in the large plot of Figure 2. The deviations were smaller during the tenth block than during the first block. We can also see, that even during the last block, the accuracy was different at different angles. The same visualization on the task completion times shown in the small plot of Figure 2 suggests that the inaccuracy in angles near the diagonals was not due to increased speed emphasis in the speed-accuracy trade-off. Therefore, angles near horizontal and vertical may really be less error prone. 11

Largest angle deviation

9

block 1

7

block 10

5 3 1

Average angle deviation

-1

Black mark to indicate the direction to aim for

1200 800 400 0

Increasing radius Circle drawn around the mouse cursor

Figure 1: A screenshot of experiment 1.

Figure 2: Large plot: Average deviation (in degrees) from the target in different directions. Small plot: The average task completion time (in ms).

The experimental software was run on a laptop computer so that it was easy to transport to locations that were convenient for the participants. 1

Procedure

The experiment consisted of ten 5-minute blocks. Within a block the participants were instructed to complete as

Because the data was divided based on the target location, the vector that the user entered did not necessarily end within the same sector.

Moyle and Cockburn reported average angular error of 3.6 degrees for a mouse, and 6.5 degrees for a stylus [4]. We measured an average of 6.7 degrees for the first block and 4.3 degrees for the tenth block. We had the disadvantage of having fewer participants, but the advantage of having more trials per participant. One of our blocks produced on average over 200 trials, which was about the same number that Moyle and Cockburn recorded for each device per participant in the entire experiment. There were further differences between the experiments that make head-to-head comparisons difficult. We omit detailed discussion in to save space. The point of this limited comparison is to show that our results on accuracy are of the same order of magnitude as earlier results in similar experiments.

means 45-degree slices, which means that deviations less than 22.5 degrees are needed. In Figure 3 slices this small almost always incur an error rate of 5% or more. We considered a 5% error rate to be too high. Therefore 6 slices seems to be the maximum that we can recommend based on the data from experiment 1.

While interesting, the data in Figure 2 was not very useful regarding our main question. What we wanted to measure was the number of errors that we could expect when using menus with different number of menu items. Because the distribution of the angular errors is highly skewed the means do not inform us much. The high-end tail of the distribution is very long meaning that to achieve acceptable error rates we need larger menu items than the means would lead us to assume. Because of this, we produced the alternative visualization in Figure 3.

EXPERIMENT 2 Participants

Largest deviation (deg)

50

1% 3% 5%

40

2% 4%

30 20 10 0 1

2

3

4

5

6

7

8

9

10

The conclusion to use 6 menu items relies on a number of assumptions. Most importantly, we assumed that the users would be aiming at the center of the menu item and that they would exhibit the same speed-accuracy tradeoff in menu use and in experiment 1. In reality, user behavior may be different. This is why we needed to conduct experiment 2 that used an actual menu instead of the target-based method of experiment 1.

Eight participants (7 male, 1 female) from experiment 1 continued in experiment 2. Two participants did not continue due to scheduling conflicts. These participants ranked 1st and 3rd in accuracy and 10th and 6th in speed in experiment 1. Thus, one should expect slightly lower accuracy and higher speed in experiment 2 on average. Apparatus

The hardware used in experiment 2 was the same as in experiment 1. The software was different. The task of experiment 2 is shown in Figure 4. The outer circle shows the color to choose in the trial. The inner circular structure is the menu. It presents six colors to choose from. A menu selection happened when the ball was moved enough for the end of the line to reach the inner border of the menu. In Figure 4 the selection of the bottom item (red) is just about to happen. After a menu selection the software showed feedback indicating that a correct selection or an error had occurred and then paused. The participant had to press the left button of the Trackmouse to start the next trial.

Block Figure 3: The largest angle between the target and the crossing point after removing 1, 2, 3, 4, and 5% of the largest angles.

Figure 3 shows the largest deviation left after removing 1%, 2%, 3%, 4% and 5% of the trials with the largest deviations. The data were removed separately for each participant and each block. Figure 3 shows the largest deviation left among all participants. We can see that the maximum deviations tend to shrink with training until about block 5, without significant changes thereafter. By block 5 all of the shown curves are below 30 degrees. Because ±30-degree deviation means 60-degree menu items, this tells us that we may be able to use 6 menu items without any of the participants experiencing more than a few percent of errors. The curve for 2% of the largest deviations removed is shown in bold in Figure 3 to illustrate how it seems to be the level at which 60-degree menu items become viable. The next symmetric menu design would be 8 slices. It

Figure 4: Screenshot of experiment 2.

The procedure was the same as in experiment 1. Participants were briefed on the task, and left to complete the ten 5-minute blocks at times that suited their schedule. Results and discussion

The dependent measures in experiment 2 were the time to do the selection after the trial was presented, and the error rate. Trials where the participant selected a color other than the one presented in the outer circle were considered errors. If the results of experiment 1 had been directly transferable to the task of experiment 2, we would have expected an error rate of about 2%. As seen in Figure 5, the error rate was higher. After block 2 the average was about 3%. In addition to the omission of the two participants there are many possible explanations for the difference. One possible explanation is that the small visual stimulus guiding selection in experiment 1 may have encouraged more accurate aiming. In experiment 2 the participants did not necessarily aim at the center of the 60-degree target sectors. 7 error rate

Error rate (%)

6 5 4

require different cognitive operations. One is concerned with planning a movement towards a visual stimulus whereas the other requires observing a color in addition to planning a movement. We expected that the additional color processing would have shown in task completion times. Of course, by the end of experiment 2 the participants were more experienced in Trackmouse use. 1200 time 1000 Time (ms)

Procedure

600 400 200 0 1

2

3

4

5 6 Block

7

8

9

10

Figure 6 : Task completion time in experiment 2. CONCLUSIONS AND FURTHER WORK

Our results suggest that six is the right number of menu items to use when a Trackmouse ball is used for pie menu selections. This does not mean that it makes sense to use pie menus with the Trackmouse. Further experiments are needed to find out whether using a Trackmouse in this manner improves user interfaces over the state of the art.

3 ACKNOWLEDGMENTS

2

We thank Vincent PETRY, Nicolas JANEL, Julien TRUCHOT, Carine BAPTISTE, Jocelyn AUBERT and Maxime WOJTCZAK for their help in this project.

1 0 1

2

3

4

5 6 Block

7

8

9

10

REFERENCES

1.

Isokoski, P., Raisamo, R., Martin, B., and Evreinov, G. User performance with trackball mice. Interacting with Computers, Vol. 19, Elsevier, 2007, pp. 407-427.

2.

Kurtenbach, G., Sellen, A. and Buxton, W. An empirical evaluation of some articulatory and cognitive aspects of "marking menus". Journal of Human Computer Interaction, Vol. 8, No. 1, 1993, pp. 1-23.

3.

Martin, B. and Raisamo, R. TrackMouse: a new solution for 2+2D interactions. Proceedings of NordiCHI 2004, ACM Press, 2004, pp. 89-92.

4.

Moyle, M., and Cockburn, A., A Flick in the Right Direction: A Case Study of Gestural Input. Behaviour and Information Technology. Vol. 24, No. 4 2005, pp.275-288.

5.

Westergaard, M., Supporting Multiple Pointing Devices in Microsoft Windows. Microsoft Summer Workshop for Faculty and PhDs, 2002 http://klafbang.dk/personlig/publications.php3

Figure 5: Error rate in experiment 2. The error bars show standard deviation computed over per participant averages.

The average task completion time is shown in Figure 6. Note that this time includes the time that the participants needed for perceiving the color in the outer ring, selecting the menu item to aim at, and finally moving the ball. The colors were always on the same places in the menu, so the participants were likely to learn the menu layout towards the end of the experiment. The average trial completion time for block 10 in experiment 1 was 865 (SD=253) ms. In experiment 2 the same average was 729 (SD=122) ms. Despite losing two fairly slow participants between the experiments, this result was surprising2. The tasks in experiment 1 and 2

2

800

The mean time in session 10 of experiment 1 without the two participants was 830 ms.