A new statistic proposed by Tiku (1975) is compared with the Ferguson (1961) and Pearson-Chandra Sekar (1936) statistics. A simulation suggests that the new statistic is not superior to the Pearson-Chandra Sekar statistic when performance is assessed by the power of the test against the Dixon (1950) alternative In a recent paper, Tiku (1975) proposed a promising new family of statistics T for testing a normal sample for possible outliers. The family contains a test statistic corresponding to each r1 and r2, where r1 is the number of outliers suspected on the left of the sample, and the number on the right. Other possible outlier statistics include the sample skewness b1, (Ferguson 1961) Shapiro-Wilks W, and the Pearson-Chandra Sekar (1936) statistic L1 = (X(n) - X)/s. The first, two of these may be used to test for any number of outliers while LI is effective only for r1, = 0 r2 = 1.
- Dixon alternative
- Pearson-Chandra Sekar statistic
- Shaviro Wilks W
- Tiku’s T statistic
- extreme values