Supplementary Figure 9: Labeled-line models of how serial dependence may arise from changes in single-unit orientation tuning.

a) Gain model: channels tuned to recently-seen stimuli temporarily become more responsive. This increased sensitivity shifts the population response to a current stimulus toward the orientations of previous stimuli. b) Shift model: single-unit orientation tuning is temporarily shifted away from the orientations of recently-seen stimuli. Shifts in orientation tuning skew the population response to a current stimulus toward the orientations of previous stimuli. c) We fit the models to subjects' responses from the fully randomized version of Experiment 1 using one half of the data (odd numbered runs; 412 trials per subject) and tested model performance on the remaining (independent) half of the data (even numbered runs; 412 trials per subject). Both the gain model (blue data) and shift model (red data) performed significantly above chance in predicting subjects' errors on the left-out data (permutation tests; shaded region shows the permuted chance distribution). While the gain model produced somewhat better fits in three of four subjects, both models provide plausible accounts of how serial dependence in orientation perception might arise from simple changes in single-unit tuning.