Errors in Variables and the Empirics of Economic Growth

We examine cross-sectional empirical evidence on the determinants of economic growth in light of an instrumental variables estimator, based on sample moments of order higher than two, which does not require extraneous instruments and which remains consistent, under quite reasonable assumptions, when measurement errors affect the explanatory variables. We focus on several in‡fluential papers — Barro (1991), Mankiw, Romer, and Weil (1992), Sachs and Warner (1997a), Easterly and Levine (1997), Levine and Zervos (1998)— and find that many of their results are “fragile”. We argue that the application of our estimator to cross-sectional empirical studies of the determinants of growth yields important insights which may qualify previous findings in the literature, especially given the errors in variables problems which are known to plague commonly used cross-sectional datasets.