This paper proposes a proxy-variable search procedure, based on a sensitivity analysis framework, aiming to provide a useful tool for the applied researcher whenever he faces measurement or proxy-variable uncertainties. Extending from the sensitivity analysis literature it proposes two main methodological innovations. The first relates to the usage of a proxies grouping process to obtain averaged coefficient estimators for theoretical explanatory variables that have more than one possible measure. The second is a proposal of using the actual empirical distribution of the available data to base the inference over the confidence probabilities in choosing each possible measure as proxy for a theoretical variable. This is done using the widely known bootstrapped residuals technique. Besides the methodological main focus, an empirical application is presented in the context of cross-country growth regressions. This empirical application provided favorable evidence to the neoclassical view about the specification of the human capital effect on growth. The results also emphasized how neglecting educational quality differentials might lead to wrong conclusions about the robustness of the relationship between human capital accumulation and economic growth.
