Predicting Fixed Effects in Panel Probit Models: With Two Applications to Heath Care Demand

Abstract

We present a novel method to estimate and predict fixed effects in a panel probit models when N is large and T is small, and when there is a high proportion of individual units without variation in the binary response. Our approach builds on a bias-reduction method originally developed by Kosmidis and Firth (2009) for cross-section data. In contrast to other estimators, our approach ensures that predicted fixed effects are finite in all cases. Results from a simulation study document favourable properties. First, we use the estimator to predict period-specific fixed effects for the extensive margin of health care demand, using German panel data for 2000-2014. We find negative correlation between fixed effects and observed characteristics. Although there is some within-individual variation in fixed effects over sub-periods, the between-variation is four times as large. Second, in an application to hospital readmission rates in a panel of hospitals following the 2010 Affordable Care Act, we find that hospital fixed effects are strongly correlated across different treatment categories and on average higher for privately owned hospitals, a difference that is completely crowded-out by competition.

Our STATA command brglm, can be downloaded in STATA via “ssc install brglm”.

About the presenter's visit

Johannes Kunz will be visiting the School of Economics on 1 August 2019. While here he will be using room 520A Colin Clark building. If you would like to meet with him or have lunch or dinner with him please contact Dr Rigissa Megalokonomou who will be his host while at The University of Queensland.  Dr Megalokonomou can be contacted on r.megalokonomou@uq.edu.au.