Universidad de los Andes, Santiago, Chile
October 09, 2015, 12:15, Room GC C3 30 (click here for the map)
Empirical evidence suggests that, under some circumstances, the introduction of a new option in a choice-set can increase the choice probability of other alternatives. This result, known as the decoy effect, defies the basic regularity assumption, which is at the root of standard models of choice that are based on a compensatory approach under the Random Utility Maximization (RUM) framework. The goal of this research was threefold. First, we worked toward the development of a practical probabilistic choice-model that could account for the decoy effect, building upon various types of choice behaviors that that been described in cognitive psychology. Then, we used the proposed choice model to study, with Monte-Carlo simulation, the power of different statistical tests for detecting the presence of this phenomenon. Finally, we designed and applied a Stated Preferences (SP) survey to detect and to characterize the decoy effect in route choice. Results of this research showed first that all the decoy effect types that have been described in the literature, can be replicated by the Random Regret Minimization (RRM) discrete-choice model. Regarding statistical testing for the presence of the decoy effect, we found that McNemar and Proportions tests showed larger power when the effect size was modeled as RRM. Finally, four conclusions were driven from the application of the SP survey. The first was that the decoy effect was present in route choice, but that it was hard to detect it in the context of commuting trips or when alternatives were far from the true trade-off line. The second result of the SP experiment was that the magnitude of the average sample effect obtained from it was coherent with a data generation process based on the RRM model. Third, the SP survey showed that the larger decoys found were of the compromise type, and that the more robust ones were those of the range type. Finally, the SP survey indicated that, although an emergent-values Logit model showed slightly better fit, the RRM had substantially superior performance in outer-sample forecasting. This final result suggests that the RRM does capture, to some extent, the underlying behavior that is causing the decoy effect, but that this choice-model may still be somehow incomplete for this purpose. Four future steps of this line of research can be identified. The first is to improve the RRM model. The second step corresponds to the design and application of a Revealed Preference (RP) experiment to detect the decoy effect in real transportation behavior. The next, is to deepen the analysis of the circumstances under which the decoy effect occurs. The final step corresponds to the study of possible transportation public policies that can benefit from the decoy effect, such as seated-only buses to favor the use of public transportation or different pricing strategies.
C. Angelo Guevara is associate professor at Universidad de los Andes in Chile; research affiliate of the Intelligent Transportation Systems (ITS) laboratory at the Massachusetts Institute of Technology (MIT); and external affiliate of the Choice Modelling Centre (CMC) at the University of Leeds. He holds an MSc in transportation from Universidad de Chile, as well as an MSc and a PhD in the same area from MIT. He has been awarded the Fulbright and the Martin-Family fellowships, as well as the honorable mention of IATBR's Eric Pas dissertation prize. His main research interest is in the modeling of choice behavior, with recent contributions on endogeneity, sampling of alternatives, behavioral economics.