Modified Semiparametric Maximum Likelihood Estimator in Linear Regression Analysis With Complete Data or Right-Censored Data

This is the authors' abstract. We don't add key points for this paper.

Consider a linear regression model where the response variable may be right-censored. The standard maximum likelihood estimator (MLE)-based parametric approach to estimation of regression coefficients requires that the parametric form of the error distribution be known. Given a dataset, we may not be able to find a valid parametric form for the error distribution. In such a case the error distribution is unknown and arbitrary, and a semiparametric approach is plausible. A special modified semiparametric MLE (MSMLE) of the regression coefficients is proposed. Simulation suggests that the MSMLE is consistent is asymptotically normally distributed and may be efficient. The new procedure is applied to engineering data.