In Chapters 2–5, we saw that the extensive research on interaction adaptation patterns is replete with seemingly conflicting findings and that few unequivocal generalizations have emerged so far. In Chapters 6 and 7 we saw that part of the difficulty has been the mix of conceptual and operational definitions that have been employed. In this and the next two chapters, we consider the role that statistical analysis plays in confounding the picture. As will be apparent, conclusions about interaction adaptation patterns are tied to the type of statistical analysis procedure used. The problem is not that researchers have been using inappropriate methods but rather, they have often failed to recognize how their choice of analysis influenced conclusions about adaptation. Thus, studies using one method might find “reciprocity” while studies using another might find “compensation.” Instead of this being a case of contradictory findings, it may merely reflect two different methods – for example, use of a between–subjects versus within–subjects design.

Our intent in reviewing statistical analysis alternatives, then, is not to identify which methods are inherently “better” but rather to distinguish different methods according to their objectives, the forms of measurement for which they are applicable, the restrictiveness of their assumptions, the kinds of estimates they supply, and so forth, so that results emanating from different analyses can be interpreted properly.

That conclusions are linked to analysis choices is not a new idea. Several past investigations have compared methods in an effort to determine empirically how they differ.