Summary of Occlusion Expectation Results

In document An attentional theory of continuity editing (Page 149-153)

Chapter 4: Cuing a Cut

4.3 Experiment 1: Results

4.3.1 General statistics

4.3.2.1 Summary of Occlusion Expectation Results

When the Exit condition is unpredictable (Random Exit subject group) subjects wait until after the object is 50% occluded by the screen edge before preparing a saccade. This is manifest as a significant decrease in correct response rates at cue position -3 (167ms before the cut) and a rapid recovery back to full accuracy after cue position 1 (42ms after the cut). If the cut occurs when the object is not occluded (0% Exit), or half occluded (50% Exit) by the screen edge there is no sign of predictive saccading and the subject takes correspondingly longer to recover full accuracy after the cut, not until after cue position 5 (208ms) and 3 (125ms) respectively. This absence of saccade preparation prior to the cut indicates that subjects would have the potential to be aware of the visual disruption caused by the cut.

When the correct response rates are aligned according to screen position rather than their position relative to the cut, a single pattern of saccade preparation incorporating all three Exit conditions can be seen. Subjects’ expect a cut to occur only when the object is fully occluded, not before.

4.3.3

Occlusion advantage hypothesis

Viewers will find it harder to adapt to cuts that occur before the focal- object is fully occluded.

149 This hypothesis is intended to test whether subjects find it easier to withdraw attention from the focal-object in preparation for a saccade when the focal-object is occluded. The advantage of 100% Exit shown in the previous section should still be seen even when the Exit conditions are predictable. Over repeated presentations it is expected that 100% Exit group will show a greater increase in performance compared with 0%, 50%, or Random Exit group.

To test this hypothesis we first need to split the data into the four Exit subject groups: 0%, 50%, 100%, and Random. The Random Exit group will be kept in this analysis as a control group as the subjects were unable to predict when the cut would occur so they should show less withdrawal of attention pre-cut compared to the other three Exit groups. Subjects were allocated to one of these groups at the start of the experiment and every cut they observed was edited with this degree of Exit. Because the onset of the cuts were as predictable across all subject groups, if there is no advantage of 100% Exit, all Exit groups should show the same degree of predictive saccading. To ensure that subjects performed a saccade as quickly as possible any experimental conditions with a Time Gap larger than 0 was omitted (conditions 1, 3, 5, 7, and 9). The conditions left in this analysis were the same as in the occlusion expectation analysis: 2 (0% Entry), 4 (50%), 6 (100%), and 8 (Random).

The mean correct response rate for each Exit subject group (listed in decreasing size) was for 100% Exit 87.18% (s.d. 1.86), 50% Exit 82.31% (s.d. 10.03), Random Exit 81.35% (s.d. 7.25) and for 0% Exit 77.88% (s.d. 2.87). Performing a one-way ANOVA76 between the groups shows there is no significant main effect of Exit for correct response rate (F=2.005, df=3, p=.144, not sig.). However, there is a significant difference between 0% and 100% Exit (Independent Samples t-test: t= - 6.661, df=10, p<.000) and 0% and Random (t=1.908, df=10, p<.05, one tailed).

76 Parametric tests are used as the distribution of correct response rates was previously shown to satisfy the tests of normality and the data being used in this stage of analysis has an N or 21.

150 Figure 4-15: Correct response rates (y-axis: %) for each cue position (x-axis: frames relative to cut) across the four Exit groups (0% blue, 50% green, 100% red, Random black).

Splitting the data across cue positions (Figure 4-15) the familiar trend of a decrease in correct response rate immediately surrounding the cut can be seen. However, the pattern varies across Exit conditions. To investigate the pattern of attention withdrawal within each Exit group repeated-measures ANOVA comparing correct response rate across the 13 cue positions.

Within the 0% Exit subject group (blue line Figure 4-15) there is a main effect of cue position (Greenhouse-Geisser: F=27.656, df=3.879, p<.000) which can be attributed to a large decrease in correct response rate at cue positions 1, 3, 4, and 5. This decrease is significant when compared with cue position -377 (position 1: mean difference = 75.0, p<.000; position 3: mean diff. = 26.67 p<.05; position 4: mean 77 Cue position 3 was used as a “baseline” comparison as it has a correct response rate representative of most pre-cut cue positions and is similar across all Exit groups (see Figure 4-15).

151 diff.=19.167, p<.05; and position 5: mean diff.= 11.667, p<.01). Correct response rate returns to normal between cue positions 5 and 7 (between 208ms and 250ms after the cut). There is no significant decrease in accuracy before the cut. This is a similar pattern as found for edits with 0% Exit within the Random Exit group (see section 4.3.2).

Within the 50% Exit subject group (green line Figure 4-15) there is also a main effect of cue position (GG: F=15.760, df=1.590, p<.01) which can be attributed to a significant decrease in correct response rate at cue positions 1 and 3 only (position 1 compared with -3: mean diff. = 67.86, p<.000; position 3 compared with -3: mean diff. = 24.29, p<.05). Cue position 4 shows no significant difference even though its mean appears lower than the pre-cut average. This lack of effect can be attributed to a high degree of variance. The period of decreased correct response rates for 50% Exit can be interpreted as 125ms.

Within the 100% Exit subject group (red line Figure 4-15) there is a main effect of cue position (GG: F=35.015, df=3.199, p<.000) and this can be attributed to a slight, but significant decrease in correct response rate before the cut at cue position 0 (compared to position -3, mean diff.=-8.33, p<.05) and a large significant decrease in correct response rate at position 1 (significantly lower than all other cue positions, p<.000). There are no other significant differences. This indicates that the decrease in correct response rate lasts 82-125ms (until between positions 1 and 3).

By comparison, Random Exit, which contains all the other three Exit conditions, also shows a significant main effect of cue position (GG: F=19.216, df=1.904, p<.000) but this is due to decreased accuracy at cue positions 0 (p<.0578), 1 (p<.000), and 3 (p<.05). This indicates that the Random Exit group shares the same signs of predictive saccading as 100% Exit but takes longer to recover after the cut. The decrease in correct response rates lasts 167ms.

78 Compared to cue position -6. Cue position -3 could not be used in this Exit group as there was uncommonly large variance in the correct response rates at cue position -3.

152 This comparison of the duration and position of the decrease in correct response rates around the cut indicates that only 100% and Random Exit show any signs of preparation for the saccade before the cut and that 100% Exit produces the shortest period of attention withdrawal: 100% (decrease in correct response rates lasts 82- 125ms), 50% (125ms), Random (167ms), 0% Exit (208-250ms).

In document An attentional theory of continuity editing (Page 149-153)