Monte Carlo analysis of SP-off-RP data

SP-off-RP questions are a recent innovation in choice modelling that solicits information from respondents in a different way than standard stated-preference (SP) experiments. In particular, the alternatives and choice of a respondent in a real-world setting are observed, and the respondent is asked whether he/she would choose the same alternative or switch to another alternative if the attributes of the chosen alternative were less desirable in ways specified by the researcher and/or the attributes of non-chosen alternatives were more desirable in specified ways. This construction, called “stated-preference off revealed-preference” (SP-off-RP), is intended to increase the realism of the stated-preference task, relative to standard SP exercises, but creates endogeneity. In this paper, we present a series of Monte Carlo exercises that explore estimation on this type of data, using an estimator that accounts for the endogeneity. The results indicate that, when the variance in the processing error by respondents is the same for SP-off-RP data as for standard SP data, the two solicitation methods provide about the same level of efficiency in estimation, even though the SP-off-RP data contain endogeneity that the estimator must handle while the SP data do not involve endogeneity. For both solicitation methods, efficiency rises, as expected, as the variance of the processing error decreases. These results imply that, if respondents are able to answer SP-off-RP questions more accurately than standard SP questions (and hence have lower variance of processing error), then SP-off-RP data are more efficient that standard SP data. This implication needs to be viewed cautiously, since (i) the actual processing error for each solicitation method is not measured in the current study, and (ii) the results are for the specific data generation processes that are used in the Monte Carlo exercises.