TESTS FOR DISCORDANCY OF SUSPECTED OBSERVATIONS IN A RAYLEIGH SAMPLE.

The Rayleigh distribution is often used in life testing experiments, applied statistics, operations research, agriculture and reliability analysis. It is well known that the inclusion of spurious observations or outliers affects adversely the data analysis and hence wrong conclusions can be made. In this paper, two test statistics T_k and Z_k for testing multiple upper outliers in a Rayleigh sample are considered. The test statistic Tk is a ratio of weighted sum of spacings Y_j's (Y_j = X_(j) - X_(j-1); j = 2, 3, …, n) including upper outliers X_(n-j+1),…, X(n) to all Yj's. The Zk is obtained from the ratio of two maximum likelihood estimates of parameter. The critical values for are obtained from the exact null distribution, whereas the same for Tk are approximated by a curve. Also, the test T_k is location and scale invariant. A simulation study is carried out to compare the performance of both tests with maximum likelihood ratio test. [ABSTRACT FROM AUTHOR]