The SHAC estimator in panel data with group-specific spatial lags

An adaptation of the spatial heteroskedasticity autocorrelation consistent (SHAC) estimator from the context of a single cross section to a panel data environment is presented. The spatial model involved with the SHAC estimator is generalized to allow spatial lag parameters and spatial weights to vary across groups of cross section members. A model of earnings growth in the 48 contiguous states of the U.S. grouped into eight subnational regions, is used as an illustration. The traditional production inputs of labor, physical capital, and human capital are significant determinants of earnings growth. Significant differences between the spatial lag parameters of subnational regions are found. Copyright Springer-Verlag Berlin Heidelberg 2014