中文摘要：

LS估计对粗差非常敏感。在粗差的探测与识别理论体系中,通常采用正态检验、学生氏t检验以及τ检验等,笔者对此进行比较分析。标准化局部敏感度指标与标准化LS残差均可用来作正态检验,但研究表明,当观测量相关时,前者的检验功效大于后者。先验单位权方差因子未知时,可依据内部学生化残差及外部学生化残差分别进行τ检验和学生氏t检验。与此对照,构造了内部学生化敏感度指标及外部学生化敏感度指标以代替标准化敏感度指标。由于τ检验理论本身存在固有缺陷,而学生氏t检验或将造成＂纳伪＂错误的增加。为此,较为稳妥的方案是仍然采用正态检验,但将标准化局部敏感度指标中的单位权中误差以其抗差LMS估计代替。

英文摘要：

The least-squares（LS） adjustment approach is very susceptive to outliers.A comparative analysis of statistical tests used for detection and identification of outliers,including Gaussian normal test,Tau test and Student＇s t-test,were addressed in details.Both the standardized local sensitivity indicator and the standardized LS residual can be served as Gaussian normal test statistics for outlier detection.However,the former is superior to the latter one for correlated observations in the sense of detection power.When the variance factor is not known,the internally Studentized residual and externally Studentized residual can be employed to perform Tau test and Student＇s t-test,respectively.In a parallel manner,the internally Studentized local sensitivity indicator and externally Studentized local sensitivity indicator were constructed and investigated.Since the probability of not rejecting the null hypothesis when it is false（type Ⅱ error） may be too high by using Student＇s t-test,and since the Tau test has a limitation in itself,both of them are not appropriate statistical tests.To circumvent this difficulty,the standard deviation involved in the standardized local sensitivity indicator can be replaced by its normalized robust least median of squares estimator,and then perform the Gaussian normal test for detection and identification of outlying observations.