Figure 2
Serial bias builds up during a session. (a) Serial biases computed using first third (black) and second third (green) of the trials for two example experiments: Cam-Can40,41 and Foster et al. I38. Black bars on the top mark where curves are significantly different, p < 0.05, permutation test. (b) Both experiments show a significant increase in serial dependence through the session computed with a sliding window of 75 trials for Cam-Can and 200 for Foster et al. I. (c) For each subject, we computed the slope of serial dependence over the course of the session (without averaging). We found that serial-bias build-up was significant in two experiments (marked with black error-bars: Cam-Can, p = 0.008; Foster et al. I, p < 1e−6). Error-bars were calculated from bootstrap distributions and unless stated otherwise (all experiments in (c), red), are standard errors of the mean.
