Misspecification of Space: An Illustration Using Growth Convergence Regressions

I illustrate the importance of choosing the correct space in empirical applications of spatial econometric models. I consider different spatial weighting matrices in an SAR(1) model -- contiguity matrix, distance based matrix and their variants adjusted for size of each observation. I show formally that only the modified weighting matrices imply specifications that are robust to changes in the sample size. I demonstrate the effect of spatial misspecification by presenting simulations of a regional convergence model. The different specification of space are also estimated using European regional data. The results confirm the sensitivity of the conclusion with respect to the choice of the space.