A machine learning approach for the factorization of psychometric data with application to the Delis Kaplan Executive...

Area of Science:

• Psychometrics
• Cognitive Neuroscience
• Machine Learning

Background:

• Replicability crisis in psychological sciences highlights issues with multivariate data factorization.
• Exploratory Factor Analysis (EFA) is standard but yields divergent models for executive functioning.
• Heterogeneity in EFA findings limits generalizability and replicability.

Purpose of the Study:

• Propose Orthonormal Projective Non-Negative Factorization (OPNMF) for robust psychometric data factorization.
• Leverage internal cross-validation to enhance generalizability of findings.
• Compare OPNMF with EFA and Principal Component Analysis (PCA) for executive functioning.

Main Methods:

• Applied OPNMF, EFA, and PCA to executive functioning scores from 334 adults (Delis-Kaplan Executive Function System - D-KEFS).
• Utilized internal cross-validation for generalizability assessment.
• Evaluated factorization replicability across gender and age subsamples.

Main Results:

• OPNMF and PCA converged on a two-factor model, distinguishing low-level and high-level executive functions.
• This two-factor model showed strong support in subsamples.
• EFA yielded a five-factor model, poorly supported in subsamples, indicating limitations in generalizability.

Conclusions:

• OPNMF provides a conceptually meaningful, technically robust, and generalizable factorization for psychometric tools.
• The parsimonious two-factor model from OPNMF encompasses EFA's complexity while improving generalizability.
• OPNMF offers a superior alternative for analyzing executive functioning data.