Path Analysis With Mixed Categorical and Continuous Data: Maximum Likelihood, Least Squares, and Bayesian Estimations

• In Event: Methods for Multilevel, Categorical, and Multidimensional Data

Abstract

In practice, it is common that a path analysis is conducted with mixed categorical and continuous variables to predict a binary outcome in fields such as education, social/behavioral sciences and medicine. The goal of the study is to compare the performance of robust maximum likelihood estimation that treats the binary outcome as categorical (MLR-probit and MLR-logit), weighted (WLSMV) and unweighted (ULSMV) least squares methods with mean and variance adjusted, and Bayesian estimation with two priors in parameter estimates and standard errors associated with each estimation method for a path analysis model with varying types and distributions of data. Findings from this study will hopefully provide some guidelines to empirical researchers who conduct these types of studies.

Authors

• Paula Castro Abbott, University of Arkansas at Fayetteville

• Xinya Liang, University of Arkansas

• Chunhua Cao, The University of Alabama

• Wen-Juo Lo, University of Arkansas

• Nnamdi Chika Ezike, University of Arkansas at Fayetteville

• Ji Li, University of Arkansas for Medical Sciences