Listy Biometryczne - Biometrical Letters Vol. 33(1996), No. 1, 33-43

Reconsidering a clustering procedure by Corsten and Denis (1990) to obtain a parsi-monious description of interaction in a two-way table the author introduces a different stopping rule based on the comparison of the mean square for negligible interactions with an external variance estimate. A more attractive idea of explaining interaction might be the data-analytic identification of exceptionally high deviations from additively. They are discovered in a selection process of indicator vectors, in principle forward (and perhaps secondarily backward). Termination of the selection is based on the comparison of the current residual mean square with an external variance esti¬mate (if absent with a substitute presented in the 1990 note). In a numerical example the latter proposal turns out to be more parsimonious and specific than the former.

interaction, two-way table, clustering, stopping rule, mean square for negligible interactions, individual extra-ordinary deviations, outliers, indicator vectors, forward selection, relative value of residuals