A simple Monte Carlo approach to examine sample robustness in growth regressions

Growth regressions are often influenced by extreme observations in the sample. We demonstrate the usefulness of a simple Monte Carlo method as a diagnostic for sample robustness. We apply the technique to a data set used by Mehlum et al. (2006), who show that institutional quality is decisive for growth in resource rich countries. Monte Carlo sampling reveals that this result hinges crucially on the inclusion of Malaysia in the sample of countries. Inclusion of Malaysia yields robust, significant estimates of the key interaction term between resource abundance and institutions, whereby strong institutions can turn resource abundance into a blessing. Exclusion of Malaysia yields robust, insignificant estimates of the interaction term, whereby institutions cannot overcome the resource curse. Further explorations find that the remaining results in Mehlum et al. (2006) are similarly sensitive to the sample of countries included. We argue that the Monte Carlo method utilised provides easily interpretable representations of the robustness of estimates to an arbitrary sample of countries and should become standard practice in growth regression diagnostics, similar to robustness testing of alternative specifications and additional covariates.