NONSTATIONARY DATA SHOULD NOT BE “CORRECTED”

Ellis (1979), in his study of interaction patterns in groups, discovered that his data did not satisfy the assumptions of a simple Markov model. In particular, he found that his data failed to satisfy the assumption of stationarity. In response to this, Ellis employed a new composite matrix procedure to generate a single set of predicted one‐step transition probabilities. This essay argues that this procedure (1) does not generate one‐step probabilities, (2) does not produce legitimately interpretable results, and (3) is a fundamentally inappropriate response to the discovery of nonstationary data. The composite matrix procedure used by Ellis is discussed and appropriate responses to the discovery of nonstationary interaction data are proposed.