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

Abstract

Preference structures that underlie survey orexperimental responses may systematically varyduring the administration of such measurement.Maturation, learning, fatigue, and responsestrategy shifts may all affect the sequentialelicitation of respondent preferences at differentpoints in the survey or experiment. Theconsequence of this phenomenon is that responsesand effects can vary systematically within the dataset. To capture these structural changes, the authorspresent a maximum likelihood–based change-pointmultiple regression methodology that explicitlydetects discrete structural changes at various pointsin time/sequence in regression coefficients bysimultaneously estimating the number of changepoints, their location and duration in the sequenceof data points, and the respective regressioncoefficients for each subset of the data defined bythe change points. An application involving astated preference or conjoint analyses study ofstudent apartment choices illustratesthat the structure of preferences changessignificantly over the sequence of profileresponses.