The Comparison between the Simulation Variance for Censored and Uncensored Data for Maximum likelihood Normal Regression Model

Abstract

A normal linear regression model is considered in which we have data for n+m individuals. For the first (n) individuals the values of the response variable, say y1, y2 , … yn represent uncensored observations while for the remaining (m) individuals, the values denote by yn+1, yn+2… yn+m represent right-censored observations. Maximum likelihood estimation of the linear regression coefficients and residual variance for the normal case with censored and uncensored data is derived and assessed through simulation studies. The main findings result of the comparison between the simulation variance for censored and uncensored data is that for estimation of β0 and β1 for n=5, 10, the variance of the ml estimator had larger values than for estimation of β0 and β1 for n=5, 10 when there was censoring.