Regularizing Threshold Priors With Sparse Response Patterns in Bayesian Factor Analysis With Categorical Indicators

• In Event: Bayesian Techniques in Structural Equation Modeling

Abstract

Using instruments comprising Likert-type response items is ubiquitous for studying constructs of interest. However, using such an item format may lead to items with response categories infrequently endorsed or unendorsed completely. In maximum likelihood estimation, this results in technical difficulties in effectively estimating factor analysis or structural equation models. This work implements a Bayesian estimation approach, which uses regularizing priors on thresholds to account for potentially sparse categories. The proposed approach to accounting for sparse responses is expected to result in more efficient posterior sampling, leading to a stronger statistical foundation for making inferences. In addition, models utilizing the proposed approach are more structured, resulting in inferences that more naturally align with research hypotheses.

Authors

• R. Noah Padgett, Harvard University
  Presenting Author

• Grant B. Morgan, Baylor University
  Non-Presenting Author