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

Fluctuations in strategy, attention, or motivation can cause large variability in performance across task trials. Typically, this variability is treated as noise, and assumed to cancel out, leaving supposedly stable relationships among behavior, neural activity, and experimental task conditions. Those relationships, however, could change with a participant’s internal cognitive states, and variability in performance may carry important information regarding those states, which cannot be directly measured. Therefore, we used a mathematical, state-space modeling framework to estimate internal states from measured behavioral data, quantifying each participant’s sensitivity to factors such as past errors or distractions, to predict their reaction time fluctuations. We show how modeling these states greatly improves trial-by-trial prediction of behavior. Further, we identify EEG functional connectivity features that modulate with each state. These results illustrate the potential of this approach and how it could enable quantification of intra- and inter-individual differences and provide insight into their neural bases.Statement of RelevanceCognitive behavioral performance and its neural bases vary both across individuals and within individuals over time. Understanding this variability may be key to the success of clinical or educational interventions. Internal cognitive states reflecting differences in strategy, attention, and motivation may drive much of these inter- and intra-individual differences, but often cannot be reliably controlled or measured in cognitive neuroscience research. The mathematical modeling framework developed here uses measured data to estimate a participant’s dynamic, internal cognitive states, with each state derived from specific factors hypothesized to affect attention, motivation or strategy. The results highlight potential sources of behavioral variability and reveal EEG features that modulate with each state. Our method quantifies and characterizes individual behavioral differences and highlights their underlying neural mechanisms, which could be used for future targeted training or neuromodulation therapies to improve cognitive performance.