本文利用martingale connections導出截斷迴歸方程式與非截斷迴歸方程 式分別在參數條件與半參數條件下，估計式之最小變異限值，並加以比較。該最 小變異限值可作為評估估計式優劣之準則，亦有助於導出有效性估計式。此外本 文利用數據實驗得知:在截斷迴歸方程式下，資料的截斷之比例愈大，其半參數 的最小變異限值失去的有效性愈小，此結果亦指出半參數估計方法在具高截斷比 例的迴歸方程式中，有利用之價值。

　　　　　This paper presents the calculation of parametric and semiparametric information bound for noncensored and censored regression models using martingale connections. The method is easy tocarry out in comparison with what Begun et. al. (1983) suggested.The examples shown in this paper indicate that there exists loss ofefficiency in semiparametric efficiency bound for noncensored andcensored regression models. In addition, in non-censored regressionmodels, the semiparametric efficiency bound can attain parametricefficiency bound in some cases. But in censored regression models,the semiparametric efficiency bound can not attain parametric efficiency bound. However, the numerical experiments show if censoring degree is high, the loss of efficiency decreases. It implies thatthe use of semiparametric estimation for censored regressionmodel with high censoring degrees suffers relatively less inefficiency from semiparametric methods.