Sample Selection in Appalachian Research

Abstract

The Appalachian Regional Commission's definition of the Appalachian region is the one used most often by researchers, politicians, and the popular press. The uncritical use of this definition of Appalachia raises issues of both selection bias and excess heterogeneity in regression analysis of Appalachian income and growth. The ARC was created as part of President Johnson's war on poverty, and the geographical extent of its purview has been driven by politics and by the geography of poverty, neither of which is exogenous. It is well known that the use of endogenous variables to choose a sample creates bias and inconsistency in estimation of regression coefficients. To identify the counties that belong to the Appalachian region exogenously we use an algorithm based on three criteria: topography, contiguity, and prevalence of slavery in the 1860 census. We apply our sample to growth regressions using data from 1970 to 2008, addressing the question of the existence of a resource curse from coal extraction. For this model we find strong evidence of excess heterogeneity, but not bias.