Improvement of Forecasting Accuracy Using a Two-Stage Multinomial Choice Model Based on Physiological Responses

This empirical paper compares the forecasting accuracy of a two-stage MNL model with that of an ordinary MNL model. The explanatory variables used in this study include individual choice set and physiological responses of the subject. Designed experiment was conducted to acquire the choice set and physiological data of the subject. Based on the estimation from the consideration stage, the proposed model estimated a choice set, and further forecasted the final choice of the subject using a two-stage MNL model. By calibrating the threshold value of the consideration stage in in-sample, the two-stage model can on average outperform the accuracy of an ordinary MNL model. We find evidence that (i) an explicitly-staged model using a choice set lead to better forecasts; and (ii) influential factors are different in each stage and they exhibit different effectiveness.