The Impact of Informative Priors for Parameter Recovery in Bayesian Confirmatory Factor Model

• In Event: Measurement Models

Abstract

The study conducted the sensitivity analysis to examine the robustness of posterior distributions with varying degrees of prior information under the CFA model within the Bayesian framework. In particular, the proposal investigated the recovery of population parameters, especially the factor loading parameter in CFA model with Bayesian estimation method. Different specifications of prior information were manipulated to investigate the known parameter recovery. Several estimation conditions were compared: the Bayesian framework implementing noninformative priors, accurate informative priors, inaccurate informative priors and wrong priors. Results showed that optimal parameter recovery could be achieved with accurate informative priors, and the overall capability of parameter recovery varied greatly based on different specifications of priors.

Authors

• Dingjing Shi, University of Virginia

• Xin Tong, University of Virginia