Substantive conclusions are dependent upon techniques of measurement

Two recent reviews of mathematical modeling in social psychology (Abelson, 1967; Rosenberg, 1968) indicate that efforts to develop formal analyses of sequential social behavior will most likely accelerate in coming years, as will research devoted to determining which of the several competing models provides the best representation of the way social interaction is sequentially organized. The purpose of our paper is to highlight two methodological difficulties which may beset the investigators in such research. We have evidence that apparently trivial decisions—such as how to operationally define one's unit of observation or how often to observe the units so defined—may have profound effects on the conclusions drawn from the data. The formal model chosen as best fitting the data may be rejected or found adequate depending upon the measurement decisions adopted. Two empirical analyses are cited below to support this contention: The first shows how closely predictions of behavior based on past performance are dependent upon the level of complexity of the units utilized in the analyses. The second shows that a Markov model of sequential dependencies in conversational speech was supported using one set of observational sampling rates but rejected when different rates were used.