Relationship between video duration, perceptual change, saccades and duration estimates. 2A. Average saccade density across participants, by video duration, split by scene type. Error bars represent the between-participant standard error. Saccade density appears consistently higher for city scenes (for durations over 2 seconds) and decreases with video duration for all scene types. 2B. Normalized saccade density by scene type: city, campus/outside, office/café. Error bars represent 95% credible intervals for a Bayesian ANOVA on the effect of scene type on saccade density. As expected, saccade density is highest in city scenes, but its association with perceptual change is not straightforward for the other two scene types. 2C. Aggregated responses by video duration, split by saccade density. Error bars represent the average normalized response and between-participant standard error, according to video duration (horizontal axis). Each participant’s trials corresponding to each video duration are split into two categories, according to whether saccade density was below or above its average for that participant and video duration. Thus, the plot is split according to within-participant and duration statistics. No overall dependency between saccade density and response can be seen. 2D. Scatterplot for trial-by-trial normalized saccade density and response. According to Bayesian Pearson’s correlation, there is a weak positive association, although the Bayes factor does not provide support for either the alternative or the null hypothesis. 2E. Aggregated error size by video duration, split by dichotomized saccade density. Although not consistent, response error appears smaller in trials with more saccades (particularly clear for 12 and 16-second videos). This possible association was confirmed by a Bayesian correlation showing a negative trend between normalized saccade density and error size (2F).