We revisit the growth empirics debate using a novel limited-information Bayesian model averaging framework in short T panels that addresses model uncertainty, dynamics, and endogeneity. We construct an estimator without restrictive distributional assumptions, illustrate its performance using simulations, and apply it to the investigation of growth determinants. Once model uncertainty, dynamics, and endogeneity are accounted for, we identify several factors that are robustly correlated with growth. We find the strongest support for the neoclassical growth variables including initial income and proxies for physical and human capital accumulation, as well as evidence in favor of both fundamental and proximate factors including macroeconomic policy, geography, and ethnic heterogeneity. In addition, we demonstrate that applying methodologies that do not account for either dynamics or endogeneity yields different sets of robust determinants.
