On the use of mixed RP/SP models in prediction: accounting for random taste heterogeneity

A basic assumption in mixed revealed preference (RP)/stated preference (SP) estimation is that both data sets represent basically the same phenomenon. Thus, we would expect individuals to show the same tastes regardless of the tool used to elicit their preferences. However, different and significant parameters are often found in each case. Although this is not an issue from an estimation standpoint, understanding why differences appear is crucial in forecasting because the model structure used in that case differs from the estimated one. This problem is compounded if differences between both data affect their ability to reproduce systematic or random taste variations because (i) microeconomic conditions on individual behaviour are more difficult to fulfil, and (ii) an erroneous specification may have a major impact on the predicted results. Problems associated with using joint RP/SP models in forecasting have received scant attention and no studies have examined the case where both types of data show different systematic or random heterogeneity. We review the problem from a theoretical viewpoint and suggest analyses that could aid decision taking in this context. Using real data, we provide evidence on the effects of using different joint RP/SP models in forecasting and highlight the importance of performing these analyses.