Pairwise Comparison Estimation of Censored Transformation Models

In this paper a pairwise comparison estimation procedure is proposed for the regression coefficients in a censored transformation model. The main advantage of the new estimator is that it can accommodate covariate dependent censoring without the requirement of smoothing parameters, trimming procedures, or stringent tail behavior restrictions. We also modify the pairwise estimator for other variations of the transformation model and propose estimators for the transformation function itself, as well as regression coefficients in heteroskedastic and panel data models. The estimators are shown to converge at the parametric (root-$n$) rate, and the results of a small scale simulation study indicate they perform well in finite samples. We illustrate our estimator using the Stanford Heart Transplant data and marriage length data from the CPS fertility supplement.