2.3 Experiment 2 –Extended durations up to five minutes

To further analyze the stability of the observed time compression effect over extended periods, we tested temporal frequency (FRQ) effects at trials of longer duration. We only considered temporal frequency as a visual feature in Experiment 2 to keep the size of the experiment tractable, since we found no significant differences between the tested visual features in Experiment 1. To maintain participant engagement and for ecological validity at longer durations, Experiment 2 employed sequences of short video clips arranged into “movies”, instead of static images. The trials tested in Experiment 2 varied in duration, including: 30s, one minute, three minutes, and five minutes.

Participants and apparatus. In the follow-up replication study (Section 2.2) we verified that similar results could be found both in HMDs and CDs when using analogous experimental set ups. Given that Experiment 2 deals with longer temporal durations, to avoid a potential confounding effect of fatigue caused by prolonged exposure [25] in HMDs, we use CDs. The stimuli were presented on a Samsung display (S24F350FHU, 1920x1080 spatial resolution, 60Hz), at a distance of 60cm from the viewer. Fifty-one participants took part in Experiment 2 (20 female, mean age 22.9 years).

Stimuli. The videos used in Experiment 2 consisted of 700 three-second video clips from the Moments dataset [26]. The set of videos was manually curated to exclude high arousal actions, synthetic content, text, or manipulated playback speed. Randomly ordered sequences of videos were arranged to form each trial. Since video clips had a fixed duration of 3s, the number of video clips per trial was a function of the length of the trial (10 clips for 30s trials, 20 clips for one minute trials, 60 clips for three minute trials and 100 clips for five minute trials). No transition effects were applied between different clips inside a given trial. Videos were played as continuous “movies” composed of back-to-back 3s clips.

Temporal frequency. To induce a high temporal frequency in sequences of videos, we subdivided each 3s clip into 1s cuts that were then reshuffled, effectively increasing temporal changes in visual content without changing the totality of visual information presented.

Procedure. Experiment 2 was carried out using half-duration judgements for measuring the perception of time, following the procedure outlined in Experiment 1 (see Section 2.1). Participants were randomly assigned to one of the four possible duration conditions, and completed all the trials with the same presentation duration to avoid bias effects due to differences in trial durations. For simplicity, we only used the 55% sampling point (prompting participants to make temporal judgements after 16.5s elapsed within 30s trials, after 33s for one minute trials, 99s for three minute trials and 165s for five minute trials), which means each duration had only two possible conditions (2 magnitude levels x 1 sampling point x 1 visual feature). Each participant completed 6 trials, for a total duration between three minutes (in the case of 30s trials) and 30 minutes (for five minute trials).

Statistical analysis. Each of the sampled trial durations was analyzed separately, testing for significant differences in high vs low magnitude levels of FRQ with Chi-square proportions tests. The answer variable was binary, as in Experiment 1.

Results. With a significance level established at p = 0.05 the duration of H-FRQ stimuli was significantly underestimated for trial durations up to three minutes (higher p_under,55 for H-FRQ, see Fig 5): 16.3% of trials in L-FRQ condition vs. 38.9% in H-FRQ at 30s (χ² = 13.27, p = 0.001, post hoc power = 0.83, ES = 0.586); 18.7% L-FRQ vs 68.7% H-FRQ at one-minute (χ² = 50.99, p = 0.001, post hoc power = 0.99, ES = 0.635); 25% L-FRQ vs. 52% H-FRQ at three-minute trials (χ² = 15.39, p = 0.001, post hoc power = 0.88, ES = 0.579). At five-minute trials, however, the effect was no longer present: 16.6% L-FRQ vs. 16.6% H-FRQ (χ² = 0, p = 1).

Y axis: % of answers. X axis: Temporal duration of the trials (30s, 1min, 3min, 5min). The displayed bars correspond to p_under,55 at the 55% sp (underestimation). The percentage of correct answers for each case is complementary to the displayed value in the graph (p_corr,55+p_under,55 = 100%). Like in Experiment 1, underestimation is more frequent in high magnitude levels, in this case in trials with a duration of up to three minutes.