Modeling Dynamic Effects in Repeated-Measures Experiments Involving Preference/Choice: An Illustration Involving Stated Preference Analysis

Preference structures that underlie survey or experimental responses may systematically vary during the administration of such measurement. Maturation, learning, fatigue, and response strategy shifts may all affect the sequential elicitation of respondent preferences at different points in the survey or experiment. The consequence of this phenomenon is that responses and effects can vary systematically within the data set. To capture these structural changes, the authors present a maximum likelihood-based change-point multiple regression methodology that explicitly detects discrete structural changes at various points in time/sequence in regression coefficients by simultaneously estimating the number of change points, their location and duration in the sequence of data points, and the respective regression coefficients for each subset of the data defined by the change points. An application involving a stated preference or conjoint analyses study of student apartment choices illustrates that the structure of preferences changes significantly over the sequence of profile responses.