Set size and delay duration interact in influencing serial dependence in visual working memory

Serial dependence can be found in observers' responses in a series of visual working memory (VWM) tasks (e.g., Fritsche et al., 2017; Bliss et al. 2017; Pascucci et al., 2019). To assess how memory load influences serial dependence in VWM, we manipulated memory load by separately varying set sizes and delays in the present study. On each trial, observers viewed a set of 1, 3, or 6 Gabor patches (between-subjects, n=20 for each set size). Then, after a random delay (sampled from a normal distribution with M = 2,000ms, SD = 500ms), one of the Gabors’ locations was cued as the target for response. Observers first reported the target Gabor orientation by matching it with an on-screen, rotatable line, then indicated their confidence rating on their orientation-judgment task on a slider bar. The difference in target Gabor orientation between every two consecutive trials was randomly chosen from 11 values, ranging from -75 to +75 degrees (15 degrees apart). We quantified serial dependence through bootstrapping: For each of the 1,000 bootstrap samples (20 observers resampled with replacement), we estimated the amplitude parameter by fitting a derivative-of-Gaussian (DoG) function to the grouped-average response errors, separately on the short- and long-delay sets of trials (median split) for each set size. Overall, consistent with previous findings, we found a significant attractive serial-dependence effect. Interestingly, attraction was the strongest for the set size of 3 (amplitude of around 2.5 deg), followed by 6 (~1.25 deg), and then by 1 (~0 deg), suggesting that serial dependence may be facilitated by an intermediate load in a memory task. Furthermore, set size and delay interacted in influencing serial dependence: for the set size of 3, attraction was stronger in short delays than in long delays, but, for the set size of 1, the effect direction reversed.