Abstract
Previous studies demonstrated that visual representations in working memory (WM) are biased by three factors: 1) the categorical structure of the stimulus space (i.e., categorical bias), 2) the other representations held in WM (i.e., inter-item bias), and 3) the previous-trial stimulus/response (i.e., serial dependence). Because the majority of previous studies have investigated these biases independently, it is unknown how they might interact in WM. The present study investigated the interaction between the three sources of WM bias. Experiment 1 used an orientation delayed estimation task for two items to investigate the interaction between categorical bias and inter-item bias. In the analysis, I measured the categorical bias separately for the trials where the two biases repulse a given item in the same direction (i.e., congruent trials) and the trials where they repulse a given item in the opposite direction (i.e., incongruent trials). I found that the categorical bias was completely canceled out by the inter-item bias in incongruent trials. Experiment 2 used a set-size one orientation delayed estimation task to investigate the relationship between categorical bias and serial dependence effect. Similar to Experiment 1, I assessed the serial dependence effect separately for the trials where the direction of the two biases was congruent and for the trials where they were incongruent. I found that the serial dependence effect was exaggerated in congruent trials but was reduced or even canceled out in incongruent trials. Together, the present study demonstrates that the three sources of bias influence each other, producing a systematic pattern of biases depending on specific stimulus features and their relationship. These results imply that WM may involve a process that integrates multiple sources of biases to support the current goal of the behavior and suggest that theories in WM cannot speak away these biases by appealing to random trial-to-trial variability.