Statistical Regularities Do Not Facilitate Encoding but Induce Biased Guessing in the Attribute Amnesia Task

While some studies have shown memory enhancement for items with statistical regularities, whether the effect persists when people are not expecting to report the probed information remains unclear. Here, we investigated whether an influence of statistical regularities on working memory encoding can be observed using the attribute amnesia task, a working memory task that employs a surprise test to probe people’s memory for an attended item. Participants completed an epoch of 320 trials in which they reported the location of a target letter among three distractor digits. In 70% of those trials, targets were presented in one color (frequent color), while the remaining 30% involved targets displayed in one of three alternative colors with equal probability (infrequent colors, 10% each). Notably, for the 321st trial (surprise trial), the color of the target was unexpectedly probed prior to the location inquiry. In this trial, half of the participants had the target rendered in the frequent color (frequent group), while the other half had the target rendered in an infrequent color (infrequent group). Analysis revealed a significant difference in surprise trial color reporting accuracy, with the frequent group (45%) outperforming the infrequent group (16.7%). However, both groups exhibited a consistent bias towards selecting the frequent color as the target color during the surprise trial (frequent group: 45%, infrequent group: 50%). This suggests that, rather than an enhancement in working memory encoding specific to the frequent color, the observed between-group difference in surprise trial performance can be fully attributed to a bias to endorse the frequent color as the target color. In summary, our findings indicate that, while no enhancement in working memory encoding attributable to statistical regularities is observed, participants to some degree extract the summary statistics of the attributes of targets across trials and these summary statistics can inform behavior.