March 3, 2026Open Access

Identifying Latent Constructs Without Distributional Assumptions: A Probabilistic Model-Based Clustering Alternative to Traditional Factor Analysis

Key Points

E-ReMI effectively captures latent constructs without strict distributional assumptions, enhancing flexibility in analysis.
Performance comparison shows E-ReMI achieved a higher coefficient of determination against traditional factor analysis.
Evaluation was conducted using simulated datasets to assess variances and accuracy of clustering assignments.
This improvement supports better identification of underlying patterns, indicating a need for new approaches in data analysis.

Abstract

This study demonstrated that Extended Random Effect Maximal Interaction Two-Mode Clustering (E-ReMI), a probabilistic model-based clustering methodology for simultaneous grouping of both individuals and items, can capture latent constructs without decreasing the number of items or stringent distributional assumptions. The performance of E-ReMI relative to EFA was evaluated using simulated datasets to assess both the amount of variance accounted for by the reduced dimensionality of items, quantified by the coefficient of determination (R²), and the accuracy of item assignment to their latent constructs, determined by the Adjusted Rand Index (ARI).

Identifying Latent Constructs Without Distributional Assumptions: A Probabilistic Model-Based Clustering Alternative to Traditional Factor Analysis

Key Points

Abstract

Cite This Study