Although most readers of this book will be familiar with analysis of variance procedures, the topic of the interpretation of interactions requires special attention for three reasons: (a) interaction effects are frequently obtained in research on nonverbal behavior, (b) interaction effects are very important in much research on nonverbal behavior, and (c) interaction effects are frequently misinterpreted in research on nonverbal behavior and, indeed, in behavioral research in general.

Referees and advisory editors for various journals in the behavioral sciences, and consultants in research methods, find the misinterpretation of interaction effects to be one of the most common of all methodological errors. The nature of the error is almost always the same: the effects of the interaction are not distinguished from the main effects. The cautionary note is best sounded in the warning: If we're looking at the means we're not interpreting the interaction.

An illustration

Suppose we have employed female and male encoders to encode various nonverbal stimuli to female and male decoders. Table 7.1 shows the resulting table of mean accuracy scores and the table of variance.

In the published report an investigator might accurately state that there was a significant effect of sex of encoder such that decoders decoding females performed better than those decoding males.