Group Comparisons Using ANCOVA: A Comparison of Five Analytic Approaches

• In Event: Methods for Causal Inference and Estimation of Treatment Effects

Abstract

This study aims to investigate ANCOVA-based methods for conducting group comparisons. We propose a Monte Carlo study to compare the performance of five analytic methods under undesirable data conditions that are analogous to empirical data: conventional ANCOVA, ANCOVA using weighted sum scores, ANCOVA using factor scores, path model with correction for unreliability, and SR model. The manipulated factors are loadings of indicators measuring a latent variable, degree of model misspecification, main effect of the grouping variable, and sample size. Results from preliminary analysis suggest that no one method outperforms the other methods in terms of estimates and standard errors of grouping effect. The issue of inadmissible solutions is not a concern. All methods show some inflation of type I error rates.

Authors

• Lu Liu, Florida State University

• Yanyun Yang, Florida State University