Bayesian Inference for Dynamics of Slowly Changing Variables in Time-Series Cross-Sectional Data Analyses

Abstract

The time-invariant and/or rarely changing explanatory variables are of interest to political scientists, including both their short- and long-run effects. However, estimating these effects in the analysis of time-series cross-sectional (TSCS) data by the conventional estimators may be problematic when unit effects are included in the model. This paper discusses the advantages of using Bayesian multilevel modeling to estimate the dynamic effects of these slowly changing explanatory variables in the analysis of TSCS data and applies a Bayesian dynamic multilevel model to analyzing the effects of political regime on social spending in Latin America.