A Method of RP/SP Data Combination for Isomorphic Nested Logit Sub-models

To accommodate the similarity between alternatives, the Nested Logit (NL) model is usually applied to construct sub-models in RP/SP combined analysis. In previous literatures, however, RP and SP sub-models often adopt different hierarchical structures, which result in problems such as inconsistent comprehension for behavioral mechanisms, unclear corresponding relation between error-components and incorrectly description of the real hierarchical structure of SP data. This paper formulates a joint method, which considers hierarchical difference between error terms within a unified RP-NL/SP-NL framework and introduces fusion coefficients to rescale SP utility. Based on the survey data collected from new bus lanes on freeway in Beijing, three different RP-NL/SP-NL models, RP and SP independent NL model were estimated and compared. Taking into account the hierarchical relation between errors of RP and SP, the result implies that the proposed method can clearly reveal the choice process and hierarchical structure of SP data, reduce the impact of error terms and improve the explanatory power of variables to understand SP choice behavior better.