A procedure for the detection of outliers that use the difference between the log likelihood of the complete data set and the log likelihood when a particular observation is removed. If the difference is large then the observation involved is considered an outlier.