Unmasking Test For Multiple Outliers in an Exponential Sample

The discordancy test for many outliers in an exponential sample is very difficult and complicated because of the masking or swamping effect. The key to solve the question lies in the determination of k, the nunlber of contaminants that are discordant outliers in an exponential sample. The avaiable outlier test metbods often do not succeed in solving this problem. In this paper, a methods based on idea of Akaike’s information criterion of choosing variables for the detection of outliers is proposed, which does not require presetting k and highly complicated compution, can accurately detect k by maximizing statistic called modified Akaike’s in formation criterion （MAIC）. The statistic and formula calculated easily the significance level of test is also proposed. Finally, numerical examples are given, showing our method is very useful in practice.