The empirical economic growth literature: Robustness, significance and size

Henri de Groot, Free University, Amsterdam, the Netherlands

The empirical growth literature is criticized for its lack of robustness. For different definitions of robustness, conclusions vary from "almost every correlation is fragile" to "a substantial number of explanatory variables are robust." We re-analyze the empirical results of the economic growth literature for various alternative definitions of robustness using quasi-experiments. The analysis pertains to sign, size and significance, of the effects, and we relax the quasi-experimental procedure by no longer applying a set of 'fixed' variables. Response surface analyses of the quasi-experiments reveal that the number of robust variables is limited, the effects crucially depend on the specification of conditioning variables, and the default specification based on the convergence/catch-up model is associated with estimated effects of conditioning variables that constitute outliers.

Date: 22 April-00 0000