Maximum likelihood estimation of the truncated and censored normal regression models
Hartley, Michael J. | Swanson, Eric V.
This paper provides a general treatment of the problem of estimating the parameters of a regression model in the presence of truncation and censoring. The maximum likelihood estimator (MLE) and its covariance matrix for both the censored and truncated models are presented. Since the likelihood equations are nonlinear, solutions must be obtained by iterative methods. Four computer algorithms for obtaining the MLE are given and compared. In addition, a method of modifying the algorithms so as to improve their rate of convergence is given. Finally Monte Carlo experiments are used to compare the consistent initial estimator with the convergent MLE and to provide guidelines for the selection of an algorithm.
显示更多 [+] 显示较少 [-]