|
Abstract:
|
The increasing amount of two-way data in diverse areas has lead to renewed interest in robust methods for detecting outlying cells. One such method, Tukey's median polish, was introduced in the 1970s as a general statistical tool. To overcome the potential lack of uniqueness of the L1 solution, Terbeck and Davies (1998) have developed methods based on M-estimators. The statistical literature provides no technical guidance, however, on how these algorithms should be adapted to handle replicates. A simulation study was performed to compare four options for handling replicates: median polish applied to individual values, median polish applied to cell medians, and two modifications of Terbeck and Davies (1998) methods also applied to cell medians. Results show that median polish applied to individual values had the best performance in detecting an outlying cell in a two-way layout.
|
