Selection of Explanatory Variables for Linear Regression Models Estimated on Regional Panel Data

Abstract

One of the problems in estimating of a single-equation linear regression model is the selection of explanatory variables. While many methods of selecting variables for models estimated on the basis of time series or cross-sectional data have been developed, there are no such methods of selecting variables for panel data models. The lack of an appropriate method for selecting variables for linear panel data models may lead to incorrect parameter values for some variables, which makes it difficult and sometimes even impossible to interpret the results of estimated model. Methods of selecting variables for panel data models cannot be based on the Pearson linear correlation coefficient. Therefore, a three-step procedure of variable selection for linear panel data models has been proposed, providing the correct parameter signs for all selected variables. The procedure is illustrated with the selection of variables for panel data models with fixed effects of the average annual unemployment rate according to Labor Force Survey (LFS) in Polish voivodships in the years 2010-2021 (balanced panel consisting of 192 observations).