Revised July 2010
Economists frequently use instrumental variables to demonstrate a causal relationship between some trait of a country and economic growth, but concern is growing that many instrumental variables used in widely cited growth regressions may be weak, invalid, or both. Attempts to remedy this problem have been inadequate. Samuel Bazzi and research fellow Michael Clemens show that top journals continue to publish growth regressions that rest on problematic instruments. They demonstrate that instruments used in many prominently published growth regressions may be either demonstrably invalid or too weak to test hypotheses about the causes of growth; their continued use risks pushing the literature further from science and closer to irrelevance.
Bazzi and Clemens urge researchers to take three steps to overcome the shortcomings: grounding research in somewhat more generalized theoretical models, deploying the latest methods to test sensitivity to violations of the exclusion restriction, and opening the “black box” of the Generalized Method of Moments (GMM) with supportive evidence of instrument strength.