Uncovering the latent structure of human time perception

Abstract

One of the ongoing controversies in interval timing concerns whether human time perception is subserved by multiple distinct mechanisms. The crux of this debate concerns whether timing over subsecond and suprasecond intervals can be attributed to a single semi-uniform mechanism or requires distinct or interacting mechanisms. Whereas past studies provide some valuable insights, our study overcomes previous limitations by adopting multiple convergent statistical approaches in a design with strong statistical power. We conducted two large online experiments in which participants reproduced temporal intervals ranging from 400 to 2400ms (Experiment 1; N=302) and 1000 to 2000ms (Experiment 2; N=302). We contrasted the application of exploratory factor analysis and structural equation modelling (ESEM; bifactor analysis) to distinguish different latent structures underlying duration reproduction patterns and compared model results against the results from subject-level changepoint analyses. In both experiments, the bifactor analysis model was the most parsimonious and best-fitting. This analysis yielded evidence for a two-factor model comprising a general timing factor spanning the full interval range and a second factor that quadratically scaled relative to middle intervals, which may reflect a Bayesian mechanism subserving the central tendency bias (reproduced intervals regressed to the mean of stimulus interval range). We observed a low proportion of changepoints in median reproduced intervals with the changepoint detection predicted by factor scores. Our findings contribute to the ongoing debate concerning the mechanisms in millisecond-to-second interval range and highlight the role of central tendency effects in investigating the changepoint and therefore the nature of continuity between timing systems.