Order Effects in Stated-Choice Experiments

Abstract

Results are presented from a study of possible order effects in stated-choice experiments tested in the context of transport mode choice decisions. A stated-choice experiment with three alternatives—car, public transport, and bicycle—is used to analyze whether the sequence in which choice sets are presented influences the estimated utility functions of transport modes. The data were collected by means of a self-administered mail questionnaire. More than 950 respondents participated in the experiment. The analyses show that order effects are relatively small but significant. The appearance of the public transport alternative before the bicycle in the choice task influences the utility of public transport significantly. Order effects depend on trip purpose.

Optimal design of factorial paired comparison experiments in the presence of within-pair order effects

“…In each of these applications, evidence of order effects was found. Such evidence was also found by van der Waerden, Borgers, Timmermans, and Bérénos (2006) in choice experiments with three alternatives per choice set for studying transport mode decisions, and by Chrzan (1994) in choice experiments when investigating brand preferences in marketing. Matthews and Morris (1995) took into account a potential order effect in a paired comparison experiment for the measurement of pain, but did not obtain convincing evidence of its existence.…”

“…For the linear model, this modification has been considered by Scheffé (1952) and, for the BradleyTerry model, it gives rise to the model of Davidson and Beaver (1977) with multiplicative order effects. The published empirical work, however, uses order effects that are common for every paired comparison or choice set (see Davidson & Beaver, 1977;Matthews & Morris, 1995;Wickelmaier &Choisel, 2006 andvan der Waerden et al, 2006). Therefore, we focus on model (1) rather than on a model with one order effect for each paired comparison.…”

“…The authors report a satisfactory model fit, despite the model simplification. Similarly, van der Waerden et al (2006) use a model with order effects that are common for every choice set. In this paper, we will therefore also focus on the situation where one order effect is common to all the pairs presented in the paired comparison experiment.…”