Optimal design of stated preference experiments when using mixed logit models.

The optimal design of Stated Preference (SP) experiments has shown the feasibility of improving the experimental designs such that better estimatescan be obtained at the end of the process, when compared with models estimated with data gathered using standard experimental designs. This optimization approach has been used to design SP experiments to value environmental aspects, to detect the presence of income effect in mode choice and to assess the subjective value of time components. In all cases Multinomial Logit models have been used to model the demand or choice process. In this paper the attention has focused on changing the demand model, such that taste variation can be detected in waiting or travel times. A Mixed Logit model was chosen to find out the presence of this condition. Given the open form of the demand model, then a combination of optimization and simulation had to be used when looking for the best SP experimental design. Software codes were prepared such that the optimization procedure could be carried out, since the minimization of the coefficient variances required some numerical analysis. These codes were linked with standard simulation Mixed Logit routines, such that the optimal designs could be obtained. This two part procedure was used to design a SP experiment dealing with the improvement of a rail service in the Great Concepcion, Chile. Cost, waiting and travel times were the explanatory variables. The choice process was presented as a binomial one, comparing the present service with an improved one. Standard experimental designs were also applied, for comparison. Preliminary results have shown that better estimates can be obtained when applying the optimal designs, with respect to the standard designs. In some cases, taste variation could be detected more robustly, whereas in other cases the occurrence of taste variation could be rejected more strongly. At the moment a more extensive survey is being carried out, such that more conclusive results can be presented. For the covering abstract see ITRD E135582.