Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
In this article we propose an inferential procedure for transformation models with conditional heteroscedasticity in the error terms. The proposed method is robust to covariate dependent censoring of arbitrary form. We provide sufficient conditions for point identification. We then propose an estimator and show that it is √<italic>n</italic>-consistent and asymptotically normal. We conduct a simulation study that reveals adequate finite sample performance. We also use the estimator in an empirical illustration of export duration, where we find advantages of the proposed method over existing ones.