We compare the sensitivity analysis of cross-country growth regressions based on extreme bounds analysis to a more direct specification testing approach using non-nested hypotheses tests. The results suggest that those specifications that are adequate are also those that include two of the only...