Performance summaries

In Figures 6.1, 6.2, 6.3 and 6.4, a summary of the experimental results is shown graph- ically. In these figures, the mean time to complete a trial and the mean number of errors incurred during the trial are shown as bars on the graph for each of the groups mentioned in Table 6.1. The mean time is shown as the left-hand bar for each group and the mean number of errors is represented by the right-hand bar. The mean time is in seconds. Each figure shows the means for a device type and a trial type (e.g. standard (USB) mouse and a straight tunnel trial in Figure 6.1). After the summary graphs, tables are given comparing

different groups using t-test results and Bonferroni p-values to establish confidence levels.

From these t-test and p-value results, conclusions are made about the significance of the

differences between groups and the effectiveness of the Virtual Dynamic Tunnel algorithm is determined.

In Figure 6.1, it can be seen that the difference between mean trial time between Group 1 and Group 2 is close to zero. This implies that for the standard (USB) mouse with a straight tunnel and training, the dynamic tunnel algorithm had little or no impact over training. Comparing Groups 1 and 3, however, show that for standard (USB)/straight, training with the algorithm was better than the algorithm alone; the difference between Groups 2 and 4 also support the effect of training on time. Lastly, comparing Groups 3 and 4 indicates that the algorithm actually impedes performance for standard (USB)/straight when used without training. The best performance in time and number of errors is when the algorithm is used with training for the standard (USB)/straight trials.

1 2 3 4 0 1 2 3 4 5 6 7 8

Standard (USB) mouse with straight target

Group number

Seconds or number of errors

Seconds to complete trial Number of errors during trial

Figure 6.1. Standard (USB) mouse and straight trial.

Figure 6.2 shows the mean results for the standard (USB) mouse with a circular trial. In this case, the difference between Groups 1 and 2 indicate that in groups given training, the algorithm makes performance worse except with respect to the number of errors, which are lower in the intervention group. This can be explained by noting that since a circular tunnel requires more fine motor skills, a trade-off between the number of errors and the completion time becomes more apparent (i.e. as a user tries to minimize errors, she/he may have to slow down and thereby increasing completion time) than in straight tunnel trials. However, in the non-trained groups (i.e. 3 and 4), both mean trial time and number of errors are worse in Group 3, which has the algorithm applied. As far as training is concerned, for standard (USB)/circular trials, there is no significant advantage to training. This can be explained by noting that typical user movement is not a complete circle, but in a semi-circular fashion and that the training phase, consisting of only one extra trial may not be sufficient for the user to get accustomed to this new way of moving the mouse cursor.

1 2 3 4 0 2 4 6 8 10 12 14 16

Standard (USB) mouse with circular targets

Group number

Seconds or number of errors

Seconds to complete trial Number of errors during trial

Figure 6.2. Standard (USB) mouse and circular trial.

The mean results for the head-tilt mouse using a straight trial are show in Figure 6.3. In this figure, it can be seen that the mean trial time for Group 1 is less than Group 2 and also less than Group 4. This indicates that training with the algorithm for the head-tilt mouse and straight trials improves completion time performance over the control group. The error counts across all groups remained flat in this case.

The last summary bar graph is shown in Figure 6.4 and represents the results for the head-tilt mouse with a circular trial. In this figure, Group 1 and Group 4 have similar performance for both mean trial time and mean number of errors. This suggests that for circular trials with the head-tilt mouse, the algorithm with training performs no better than the control group. Also, Group 3’s low performance indicates that the algorithm impedes performance if training is not used in this case.

The preceding graphs show the results summary for the experiments with some con- jecture about the meaning of those results. In the following section we further analyze

1 2 3 4 0 2 4 6 8 10 12 14

Head tilt mouse with straight target

Group number

Seconds or number of errors

Seconds to complete trial Number of errors during trial

Figure 6.3. Head-tilt mouse and straight trial.

the results using t-tests betweens groups and establish confidence levels using Bonferroni

1 2 3 4 0 5 10 15 20 25 30 35 40 45

Head tilt mouse with circular target

Group number

Seconds or number of errors

Seconds to complete trial Number of errors during trial

Table 6.2. T-tests of trial time comparing Group 3 to Group 4.

Device Trial type t-test Bonferroni p-value Reference figure

standard mouse straight 1.18 0.292 Figure 6.1

standard mouse circle 1.98 0.104 Figure 6.2

head mouse straight 0.48 0.653 Figure 6.3

head mouse circle 2.31 0.069 Figure 6.4