A computer-based method using the Box-Cox transformation for identifying outliers

The Box-Cox transformation is studied with the objective of identifying outliers in a data set. It is important to determine whether the evidence for the particular transformation is spread evenly throughout the data or is limited to a few cases only. In addition to approximate measures for identifying outliers in literature, computation of the exact likelihood distance is made possible by the powerful processing capability of modem computers. The feasible solution algorithm (FSA) for least trimmed squares regression of Hawkins (1994) is adapted for Box-Cox transformed data to identify more than one outlier. The method is illustrated by applying it to real data.