Evaluation

In an experiment we tested the suitability of the aforementioned techniques for controlling a remote pointer on an external display. Participants of our study acquired targets of different sizes and positions using each of the techniques. This section briefly describes the experimental evaluation and its results regarding selection time and error rate. A detailed description of the experiment and its results can be found in [BJB09].

Figure 4.6: Procedure of a single trial: before the start of a trial the display shows both

the start button and the target (a). After starting the trial, the start button disappears and the target is rendered in solid red (b). Hovering over the target causes it to turn yellow (c).

To identify potential strengths and weaknesses, we asked participants to acquire targets of three different sizes, two different distances (from the display’s center), and eight different directions (i.e., the eight generic orientations). These parameters were chosen in a way that targets never reached the screen’s boundaries to observe potential overshooting effects. The overall task has been modeled as a multidirectional tapping task [DKM99]. At the beginning of each trial, par- ticipants had to select a start button which was rendered as a solid red square in the display’s center (see figure 4.6a) causing it to disappear (see figure 4.6b). Then they could move the re- mote pointer freely towards the target using the current technique. Visual feedback was provided when the pointer was inside the target’s boundaries by turning the target into a yellow square (see figure 4.6c). The trial was completed, when the participant pressed the device’s select but- ton while the pointer was inside the target. Pressing the button while the pointer was outside

4.2 Pointing on the Target Display 75

the target increased the error count. In contrast to other Fitts’ law experiments (cf. [FBB+05]), participants had to press and release the button while the pointer was inside the target. We asked our participants to acquire the targets as fast as possible while maintaining a low error rate. During the test we recorded both the task completion time (i.e., the time from pressing the start button until a successful target selection) as well as the error rate (i.e., the number of unsuccess- ful target selections). We further recorded the pointer’s trace for each trial to identify possible overshooting effects. The results show that Scrollwas – as expected – the overall slowest tech- nique whileMovewas the fastest. For small target sizesTiltperformed worse than the other two techniques which may be explained by overshooting the target leading to increased movement times. For medium-sized and large targets (48 and 72 pixels respectively), Movehad a mean selection time of 1401 ms (Tilt: 1868 ms) while Scrollhad a significantly higher selection time of 2023 ms. The selection time improved by 31% for Moveand 8% for Tilt respectively. The improvement of Move for small targets (24 pixels) was considerably lower (13% compared to

Scroll). Surprisingly, Tiltwas slower than Scrolland resulted in a loss of performance by 15%.

The results are consistent for both close and distant targets (see figure 4.7a).

24 48 72 0 20 40 60 80

Mean Error Rate (in

%

)

100

Target Size (in pixels)

Tilt Move Scroll Long Short 24 48 72 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Mean Select

ion Time (in

seconds)

4.5

Target Size (in pixels)

Tilt Move Scroll

a

b

Figure 4.7: Evaluation of task times and error rates of relative pointing: a) shows the mean

task times, b) denotes the average error rates. Error bars represent 95% confidence intervals. As shown in figure 4.7b, Scroll was by far the least error prone technique. Especially for small- and medium-sized targets,MoveandTiltshowed similar error rates (i.e., no significant difference) which were significantly higher compared toScroll. The distance between the start button and the target also had an effect on the selection time for small- and medium-sized targets. It can be stated, that the closer a target is, the less error prone is the technique. This can again be explained with potential overshooting effects. While this holds true forMoveandTilt, the error rate ofScroll

was consistent regardless of the target’s size and distance. For large distances,Movehad an error rate of 46% for small targets and 21% for medium targets respectively. For large targets the error rate decreased to 9%. Tiltshowed nearly the same error rates. An explanation for the high error rate of smaller targets is due to slight phone movement while pressing and releasing the phone’s select button. For short distances,Movehad error rates of 23% for small targets (Tilt: 29%) and 6% for medium-sized and large targets (Tilt: 13% for medium-sized, 7% for large targets).

Figure 4.8:Traces of relative pointing: Left (right) shows traces for all participants for short (long) distances. The targets are placed diagonally at 225◦(direction: NW).

The extremely high error rates in conjunction with high selection times for small targets ofTiltcan be explained with the overshooting effect as shown in figure 4.8. During our study, we observed this effect numerous times. Participants moved the pointer towards the target and accelerated it (i.e., increasing the tilting angle of the device) to get there faster. However, the pointer eventually became too fast to control causing it to overshoot the target. Participants had to go back in order to correctly select the target and complete the trial. While this seems obvious for targets far away, the effect also occurred for short distances. Here we assume that participants were not able to smoothly accelerate the pointer causing a short but fast movement. As denoted in figure 4.7c, the overshooting effect occurs for all techniques but is significantly worse forTilt.