Alternative nested logit models: A response to comments by Andrew Daly on an earlier paper of Frank Koppelman and Chieh-hua Wen

This paper reviews and refutes three criticisms by Daly of earlier papers by Koppelman and Wen. The original papers identified important differences between two forms of the nested logit model. These differences between the Utility Maximizing Nested Logit (UMNL) and the Non-Normalized Nested Logit (NNNL) lead to important differences in model estimation, interpretation and prediction. We reaffirm the position that the NNNL model is not consistent with utility maximization unless normalized as recommended by Koppelman and Wen and challenge Daly's criticism of the earlier papers. First, Daly's hypothesis of the existence of an alternative behavioral theory, while correct in principle, is irrelevant to the question at hand, selection of a preferred model form, in the absence of an explicit theory. Second, the equivalence of the two models under special circumstances is consistent with Koppelman and Wen's recommendation for normalization of the NNNL model. Third, the argument that the NNNL has the advantage that it can estimate models which combine data from different sources is not correct as both model structures can estimate such models. Thus, we maintain the conclusion that the NNNL and UMNL models are distinct and the UMNL is preferred because of its consistency with established theories of behavior and social welfare.