Domains of Behavior and Psychopathology Derived via Factor Analysis

 

Introduction

Predicting behavior from brain imaging data could advance the development of biomarkers for psychiatric conditions. For this, we need robust behavioral scores, combining data from questionnaires and psychological tests into meaningful summary variables.

Methods

We compared Factor Analysis (FA), Principal Component Analysis (PCA), and Independent Component Analysis (ICA) in how they decompose the Human Connectome Project's Behavioral Data to explore the latent structure of behavior and derive scores for prediction. The methods were compared in terms of model fit (variance explained), robustness (stability of factors/components over subsamples), and interpretability.

Results

PCA and ICA had slightly higher model fit than FA. FA and PCA were more robust than ICA. FA was robust for at least up to the first five factors. Using five factors/components gives a robust and meaninful decomposition with factors for Social Well-Being, Cognitive Accuracy, Psychopathology, Cognitive Speed, and Substance Use.