Image Size vs. Scrolling in Photo Thumbnail Browsers

Results

Our program collected information on task performance time, error rates, and subjective satisfaction, which we imported into Microsoft Excel for analysis (see Appendix B). We did statistical analysis of time to correct completion and number of errors for the tasks. Each subject performed 3 tasks in each of the 6 treatments. We used the average of the sums of the times and the errors in each set of 3 tasks for the analysis. We also did a statistical analysis of subjective satisfaction using the results of the post-experiment questionnaire. All statistics are available in Microsoft Excel format in Appendix C.

Performance Time

A single factor analysis of variance (ANOVA) on performance time for the fixed size photos showed that as collection size increased from 24 to 36 to 48, the performance speed slowed down. This result was significant at alpha = .01 (f(2,69) = 17.45, p < .001). A set of paired t-tests were then run to determine whether the slow down was statistically significant between collections of size 24 and 36, 36 and 48, and 24 and 48 in the fixed treatment. All t-tests were significant at alpha=0.01, (t(23) = -6.70, p = 0.00000039), (t(23)= -3.14, p = 0.0023), and (t(23)= -4.84, p = 0.000035) respectively.

A single factor ANOVA on performance time for the variable size photos also showed that as the collection size increased from 24 to 36 to 48, the performance speed slowed down. This result was significant at alpha=0.01 (f(2,69) = 29.09, p < 0.01). A set of paired t-tests were then run to determine whether the slow down was statistically significant between collections of size 24 and 36, 36 and 48, and 24 and 48 in the variable treatment. The t-tests from 36 to 48 (t(23) = -5.32, p = 0.000011), and 24 to 48 (t(13) = -6.73, p = 0.0000004) were significant at alpha=0.01. The t-test for 24 to 36 was significant at alpha=0.05 (t(23) = -2.11, p = 0.02).

We ran another set of paired t-tests on performance time between the fixed and variable treatments for each collection size. The collections of 24 and 48 did not show any significant difference between the fixed and variable treatments. The collection of size 36 was significantly faster with variable sized photos at alpha=0.01 (t(23) = 5.23, p = 0.000013).

The means and standard deviations of the times in all six treatments are displayed in Figure 1 below. This information is presented graphically in Figure 2 below.

Error Rate

A single factor ANOVA on the error rate for the fixed size photos showed that as collection size increased from 24 to 36 to 48, the number of errors increased. This result was significant at alpha=0.05 (f(2,69) = 3.86, p < 0.05). A set of paired t-tests were then run to determine whether the increase was statistically significant between collections of size 24 and 36, 36 and 48, and 24 and 48 in the fixed treatment. The t-tests for all three were significant at alpha=0.05, with (t(23) = -2.00, p = 0.03), (t(23) = -1.81, p = 0.04), and (t(23) = -2.39, p = 0.013) respectively.

A single factor ANOVA on the error rate for the variable size photos also showed that as the collection size increased from 24 to 36 to 48, the number of errors increased. This result was significant at alpha=0.01 (f(2,69) = 7.58, p < 0.01). A set of paired t-tests were then run to determine whether the increase was statistically significant between collections of size 24 and 36, 36 and 48, and 24 and 48 in the variable treatment. The t-test between 24 and 36 was not statistically significant. The t-test between 36 and 48 was significant at alpha=0.05 (t(23) = -2.44, p = 0.011) and the t-test between 24 and 48 was significant at alpha=0.01 (t(13) = -3.29, p = 0.0016).

We ran another set of t-tests on error rates between the fixed and variable treatments for each collection size. There were no significant differences for any of the collection sizes.

The means and standard deviations of the errors in all six treatments are displayed in Figure 3 below. This information is presented graphically in Figure 4 below.

Subjective Satisfaction

A single factor ANOVA on the responses to the subjective satisfaction questionnaires showed that as the collections got larger, the preference for the fixed treatment (larger pictures with scrolling) increased. This result was significant at alpha=0.05 (f(2,69) = 4.16, p < 0.05). A set of paired t-tests were run to see if the preference was significant between collections of size 24 and 36, 36 and 48, and 24 and 48. The t-tests between 24 and 36 (t(23) = 3.73, p = 0.00054) and between 24 and 48 (t(23) = 2.82, p = 0.0049) were significant at alpha=0.01. The t-test between 36 and 48 was not significant.

Another set of t-tests was run between the treatment preference for each collection size and a value that would indicate no preference. For the collection of size 24, there was no significant preference. For collections of size 36 and 48, the fixed treatment was preferred in both cases. These results were both significant at alpha=0.05, with (t(23) = -2.34, p = 0.014) and (t(23) = -2.38, p = 0.013) respectively.

The means and standard deviations of the responses to the subjective satisfaction questionnaire for each collection size are displayed in Figure 5 below. This information is presented graphically in Figure 6 below.

Department of Computer Science: Direct questions and comments to the student editorial team