Nonlinear Censored Regression Using Synthetic Data

The problem of estimating a nonlinear regression model when the dependent variableis randomly censored is considered. The parameter of the model is estimated by leastsquares using synthetic data, that is a suitable transformation of the response variablesthat preserves the conditional expectation. Two such transformations are considered.Consistency and asymptotic normality of the least squares estimators are derived. Theproofs are based on a novel approach that uses i.i.d. representation of synthetic datathrough Kaplan-Meier integrals. The asymptotic results are completed by a small com-parative simulation study.