Conducting Hypothesis Tests With Parallel Analysis and Likelihood Ratio Tests: A Comparison of Error Rates

• In Event: Alternative Methods for Data Analysis

Abstract

Parallel analysis (PA) assesses the number of factors in exploratory factor analysis. It involves conducting a series of tests of hypotheses that the data have no more than k factors. If the 95th percentile rule is used with PA, the hypothesis is being tested at the .05 level. Thus, PA can be viewed as an alternative to the likelihood ratio tests (LRTs). In our Monte Carlo study, we compared the error rates of PA and LRTs methods. We manipulated the number of observations, number of factors, number of variables per factor, size of loadings, and size of the correlations between factors. Across the 132 conditions, we assessed the error rates for eight variations of PA and the two LRTs.

Authors

• Samuel B. Green, Arizona State University

• Roy Levy, Arizona State University

• Marilyn S. Thompson, Arizona State University

• Wen-Juo Lo, University of Arkansas

• Yixing Liu, Arizona State University