Decoding load or selection in visuospatial working memory?

Abstract Flexible updating of information in Visual Working Memory (VWM) is crucial to deal with its limited capacity. Previous research has shown that the removal of no longer relevant information takes some time to complete. Here, we sought to study the time course of such removal by tracking the accompanying drop in load through behavioral and neurophysiological measures. In the first experimental session, participants completed a visuospatial retro‐cue task in which the Cue‐Target Interval (CTI) was manipulated. The performance revealed that it takes about half a second to make full use of the retro‐cue. In a second session, we sought to study the dynamics of load‐related electroencephalographic (EEG) signals to track the removal of information. We applied Multivariate Pattern Analysis (MVPA) to EEG data from the same task. Right after encoding, results replicated previous research using MVPA to decode load. However, especially after the retro‐cue, results suggested that classifiers were mainly sensitive to a selection component, and not so much to load per se. Additionally, visual cue variations, as well as eye movements that accompany load manipulations can also contribute to decoding. These findings advise caution when using MVPA to decode VWM load, as classifiers may be sensitive to confounding operations.