The effect of retirement on health: a comparison of different panel data estimators

Abstract

This presentation provides new evidence on the question of whether retirement leads to improvements in health, using a difference-in-differences (DID) framework to estimate non-linear panel data models for workers in Germany between 1992 and 2017. Implementing DID in non-linear panel data models for self-assessed health and satisfaction with health, both ordered responses, or the number of doctor visits, a count, creates methodological challenges to which solutions are highly model specific. But there are some common themes: biases due to wrong parametric assumptions, the incidental parameters problem, the non-trivial task of deriving and estimating average partial effects. We review recent developments in the related econometric literature, and apply them in a unified framework and for a single dataset and parameter of interest, allowing to assess the robustness of results to specific modelling assumptions and to highlight the relative advantages of different approaches.

About the presenter

Rainer Winkelmann studied economics at the Universities of Konstanz, Paris, Washington University in St. Louis and the University of Munich. Since 2001, he is Professor of Statistics and Empirical Economic Research at the University of Zurich, and, since 2015, Chairman of the Economics Department. He also taught at Dartmouth College, USA, and the University of Canterbury, New Zealand, and was visiting professor at Harvard University, Syracuse University and UCLA. He has acted as an advisor to the German and New Zealand governments, on issues such as immigration policy, the international mobility of highly skilled workers, trade liberalization, and income inequality. He is chairman of the advisory boards of the DIW/SOEP Survey committee and the KOF Swiss Economic Institute.

His research interests include empirical labour economics, in particular unemployment, migration, and income distribution; and econometric models for count and panel data.