Stated-preference surveys provide a good opportunity for obtaining information for the construction of choice models, but they run the risk of asking questions that are too complex for respondents to answer meaningfully, especially when many attributes and alternatives are considered in each question. An alternative technique is the use of adaptive conjoint analysis or adaptive stated-preference (ASP) surveys, wherein simpler questions are used, each game is dependent on the answers given to previous games, and parameters are estimated at the level of the individual. A simulation study used to test the ability of ASP to produce unbiased estimates of the mean and standard deviation of the value of time (VOT) is described. It is found that this is possible under certain conditions, namely, that a seven-point rating scale is used to obtain responses, each respondent plays three games for each parameter to be estimated, the starting estimate of VOT used should be higher than the expected final mean value of VOT, and a logit model sensitivity parameter of c = 5 is used in the binary logit model used to convert probabilities of choice into utility differences. The results of this simulation give reassurance that in addition to being easier for respondents, the ASP survey method produces unbiased estimates of the distribution of VOT (or any other parameter under consideration). Given the extreme flexibility of the disaggregate results produced by the ASP survey method, the demonstration of unbiased estimates is an important confirmation of the usefulness of this method.
Abstract