Premium and protection of a response estimation procedure for two-way tables when outliers occur (Q2266314)

The authors briefly recapitulate a procedure for testing of the presence of \(K=1\) and \(K=2\) outliers in two-way tables by the \(Q_ K\)-test statistic [cf. \textit{J. A. John} and \textit{N. R. Draper}, Technometrics 20, 69-78 (1978; Zbl 0395.62030)]. Then so detected spurious observations are deleted and the model (linear regression scheme) fitted. The authors examine what kind of protection against a particular type of spuriosity is afforded by such a procedure and they determine the price that must be paid for using this procedure when, in fact, no spurious observations are present in the data set. Simulations of a number of specific two-way tables are carried out and the simulation method is described in detail. At the end the discussion of simulation results shows, how difficult it can be to detect one or two outliers in a data set, when the size discrepancies are small and moderate and that the protection afforded is surprisingly insensitive to the test level used for the \(Q_ K\) statistic.