Dynamic Cognitive States Predict Individual Variability in Behavior and Modulate with EEG Functional Connectivity during Working Memory

Abstract

Short-term fluctuations in strategy, attention, or motivation can cause large variability in cognitive performance across task trials. Typically, this variability is treated as noise when analyzing the relationships among behavior, neural activity, and experimentally structured task rules and stimuli. These relationships are thought to remain consistent over repeatedly administered identical task conditions (e.g. trial types and stimuli) while the variability is assumed to be random and to cancel out when averaged across trials and individuals. We propose that the variability carries important information regarding a participant's internal cognitive states, and could provide insights into both intra- and inter-individual differences in performance and its neural bases. However, these states are difficult to quantify, as they are not directly measurable. Therefore, we use a mathematical, state-space modeling framework to estimate internal cognitive states from measured behavioral data to predict each participant's reaction time fluctuations. We can quantify each participant's sensitivity to different factors (e.g. previous performance or distractions) that were predicted to affect cognitive states, and thus become sources of variability. By including a participant's states in the behavioral model, we improved model performance by a factor of 10, over a model with only experimental task parameters. We show how the participant-specific states reflect neural activity by identifying EEG functional connectivity features that modulate with each state. Overall, this approach could better quantify and characterize both individual and population behavioral differences across time, which could improve understanding of the neural mechanisms underlying the interactions among cognitive, strategic and motivational processes affecting behavior.