| Abstract: | In this paper we show how utility, lottery dependent utility, and weighted utility models can be calibrated using algebraic and statistical techniques. The models are empirically compared in laboratory (student subjects) and real settings (sales force personnel of the Los Angeles Times). In our empirical comparison we evaluate two aspects: the extent to which observed preferences are consistent with each model, and predictive accuracy of the models on a holdout sample. The results indicate that only about 20% of the observed choice patterns in our experimental design are consistent with the expected utility model, 50% with the weighted utility model, and 90% with the general lottery dependent utility model. On individual level predictions to the holdout sample, however, the expected utility model does as well as the other two models. This latter finding is robust across different measurement and estimation methods and student and non-student subjects. |
