A MAXIMUM-LIKELIHOOD APPROACH TO CORRELATIONAL OUTLIER IDENTIFICATION

This article introduces a maximum likelihood approach to correlational outlier identification and compares it to the Mahalanobis D squared and Comrey D using a Monte Carlo simulation. The performance measures used were the hit rate and bias in correlation estimates resulting from the application of each technique. The results indicate that identification performance depends heavily on the nature of the correlational outliers and the performance measure used, but that the maximum likelihood approach exhibits the most robust performance across conditions.