Since the social learning variables are highly significantly correlated with SRI use, as will be discussed below, we maintain these variables. But note that any endogenous component of these regressors generates a correlation between covariates and the error term akin to the selection-onunobservables problem. Thus, tests for the robustness of results in the face of prospective selection-on-unobservables provide an effective check on the qualitative impact findings, even if some of the first-stage regressors are endogenous.