A G-to-S Approach to Unobserved Heterogeneity in “Short” Dynamic Panel Models

• In Event: Research in Causal Inference and Applications

Abstract

Researchers using panel data must think carefully about the model and estimation technique they use, accounting for the dynamic structure of their data and controlling for individual effects, if necessary. While excellent advice on addressing unobservable individual effects is provided in the political methodology literature, advice for doing when using dynamic models is incomplete. This paper fills in some of the gaps by doing the following three things: 1) outline a general-to-specific model selection and estimation procedure to address issues of individual effects when there is a concern that a dynamic model may be necessary; 2) test hitherto neglected “transformed likelihood” estimators for dynamic panel models; and 3) do these two things for the type of short panel data (T<10) typically used by those studying public opinion and political behavior.

Author

• Mark A. Pickup, Simon Fraser University