Bayesian Estimation of Confirmatory Factor Analysis for Count Data With Extensions for Cross-Classification

• In Event: Division D Graduate Student in Progress Research Gala

Abstract

In this paper, we develop a confirmatory factor analysis (CFA) model under the assumption that the data are Poisson. The estimation method is Bayesian, and we provide prior specifications for model parameters. Bayesian Poisson CFA of a single simulated dataset (n = 300) shows that the method performs acceptably in terms of parameter recovery, and the parameter estimates are similar to maximum likelihood estimates. We extend the Poisson CFA to cross-classified data. Analysis of a single simulated dataset (n = 300), with two crossed hierarchies shows acceptable parameter recovery for the Bayesian approach. These preliminary results suggest that the Bayesian approach we developed may be adopted for CFA of count item indicators, if the data are single or multilevel.

Author

• James Ohisei Uanhoro, The Ohio State University - Columbus