A0835
Title: Post-empirical Bayes regression
Authors: Yu-Chang Chen - University of California, San Diego (United States) [presenting]
Sheng-Kai Chang - National Taiwan University (Taiwan)
Shuo-Chieh Huang - University of Chicago (Taiwan)
Abstract: Empirical Bayes (EB) methods are widely utilized in economics for estimating individual and group-level fixed effects across diverse contexts, including teacher value-added, hospital qualities, and neighborhood effects. While estimates generated by EB are often incorporated into other statistical analyses like regression models, the econometric properties of post-EB regression have not been thoroughly investigated. This knowledge gap is addressed through two key contributions. First, a unified framework is introduced for two-step EB methods that apply to linear and non-linear models, offering insights into their frequentist properties and assessing their robustness against model misspecification. Second, a critical evaluation of commonly used two-step EB methods is undertaken in existing empirical research. The analysis demonstrates that naive implementations of post-EB regression can introduce a systematic bias, particularly in non-linear models
