Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
We quantify the informational content of special regressors in heteroskedastic binary response models with median-independent or conditionally symmetric errors. Based on Lewbel (1998), a special regressor is additively separable in the latent payoff and conditionally independent from the error term. We find that with median-independent errors a special regressor does not increase the identifying power by a criterion in Manski (1988) or lead to positive Fisher information for the coefficients, even though it does help recover the average structural function. With conditionally symmetric errors, a special regressor improves the identifying power, and the information for coefficients is positive under mild conditions. We propose two estimators for binary response models with conditionally symmetric errors and special regressors.