SYSTEMATIC BIASES IN CODED SOCIAL INTERACTION DATA

Abstract

Sequential techniques to analyze social interaction (lag-sequential analysis, Markov chains, semi-Markov models) are newcomers to the methodological arsenal of communication researchers. As with any new methodology, these have many as yet unvoiced presumptions whose violation can affect materially the outcome of their applications. I identify one such presumption—“the perfect measurement presumption”—and demonstrate that violations of it can lead to serious systematic bias in parameter estimates. In addition, I discuss briefly a set of palliatives for this source of systematic bias.