Fine-Tuning Parallel Analysis to Make More Nuanced Decisions About the Number of Factors

• In Event: 34.079 - Roundtable Session 14
  In Roundtable Session: 34.079-10 - Innovative Research on Factor Analysis and Dimensionality

Abstract

Past research suggests that revised parallel analysis (R-PA) is an accurate method to assess the number of factors relative to other parallel analysis approaches. We argue that R-PA should be modified to (a) allow for factor models with imperfect fit and (b) include effect size statistics to assess the statistical strength of factors. We offer modifications to R-PA that meet these goals. To assess the efficacy of these modifications, we manipulate six dimensions in a Monte Carlo study: type of factor model, factor loadings, factor correlations, model misfit for sample data, sample size, and model misfit for comparison datasets. The results of this study should offer insights into how parallel analysis can be adapted to improve its accuracy and interpretability.

Authors

• Samuel B. Green, Arizona State University

• Yuning Xu, SRI International

• Marilyn S. Thompson, Arizona State University