Censored Normal Regression with Measurement Error on the Dependent Variable
当删失正态回归模型的非极限观测因变量存在零均值测量误差时，传统Tobit最大似然估计不一致。本文提出八种基于期望函数的估计量，数值实验表明它们在存在测量误差时比传统估计更有效，无误差时也表现良好。
When zero mean measurement error is added to the dependent variable for the nonlimit observations of the censored normal regression model, the conventional maximum likelihood estimator (Tobit) is inconsistent. Correct maximum likelihood estimation appears to be computationally difficult under various specifications for the distribution of the measurement error. Estimators based on either the expectation function or the conditional expectation function for uncensored observations remain consistent in the presence of measurement error. Eight such estimators are examined. The results of a numerical experiment suggest that several of these estimators are substantially more efficient than the conventional maximum likelihood estimator when measurement error exists and that they also will do reasonably well when it does not.