Vol. 3, Issue 7, Part S (2017)

Abstract
Likes (1966) and Kabe (1970) have suggested a test statistic to detect single upper outlier for the exponential sample. Very few techniques of outlier detection are available for the extreme value distributions. In more recent works Lalitha and Tripathi (2016) have suggested a test statistic for the detection of outlying pair in a Gumbel sample with known scale parameter. But no such technique is available for the detection of a single outlier in a Gumbel sample. In this paper the procedure suggested by Likes (1966) and Kabe (1970) is suitably modified to obtain a test statistic for detecting discordant observation in both the cases namely upper and lower ones. The theoretical densities of these test statistics are obtained and compared with their respective simulated densities. Also their corresponding critical values are obtained. To highlight their usage some examples are discussed.