Data-Driven Extraction of a Nested Structure of Human Cognition

Abstract

Abstract:Decades of cognitive neuroscience research have revealed two basic facts regarding task-driven brain activation patterns. First, distinct patterns of activation occur in response to different task demands. Second, a superordinate, dichotomous pattern of activation/de-activation, is commonly observed across a variety of task demands. We explore the possibility that a hierarchical model incorporates these two observed brain activation phenomena into a unifying framework. We apply a latent variable approach, exploratory bi-factor analysis, to a large set of brain activation patterns to determine the potential existence of a nested structure of factors that underlies a variety of commonly observed activation patterns. We find that a general factor, associated with a superordinate brain activation/de-activation pattern, explained the majority of the variance (52.37%). The bi-factor analysis also revealed several sub-factors that explained an additional 31.02% of variance in brain activation patterns, associated with different manifestations of the superordinate brain activation/de-activation pattern, each emphasizing different contexts in which the task demands occurred. Importantly, this nested factor structure provided better overall fit to the data compared with a non-nested factor structure model. These results point to domain-general psychological process, representing a ‘focused awareness’ process or ‘attentional episode’ that is variously manifested according to the sensory modality of the stimulus and degree of cognitive processing. This novel model provides the basis for constructing a biologically-informed, data-driven taxonomy of psychological processes.