Advances and Applications in Statistics
Volume 9, Issue 1, Pages 109 - 125
(June 2008)
|
|
A COMPARISON OF DISTRIBUTION TEST STATISTICS FOR DETECTING OUTLIERS IN DATASET
D. K. Shangodoyin (Botswana), P. M. Kgosi (Botswana) and K. Setlhare (Botswana)
|
Abstract: Given that serial correlations exist between successive observations and that series are stationary and invertible, we evaluate by simulation method the performance of the maximum likelihood ratio statistic and z-statistic under some specified conditions. The comparison of the test statistics is achieved through the power function of the tests and the size of the observations. The results show that the choice of criterion for outlier detection depends on the value of the outlier and power of test of the criterion as well as the feasibility of the criterion for practical purposes; our findings reveal that for different sizes of the series, the standard normal test detects more outliers than the maximum likelihood ratio test. |
Keywords and phrases: maximum likelihood ratio, standard normal test, outliers and power function. |
|
Number of Downloads: 350 | Number of Views: 1235 |
|