Modeling choice behavior with data from multiple sources has received attention in recent years as an alternative way to cope with weaknesses resulting from using a single data set. This paper presents two alternative structures, the non-normalized nested logit (NNNL) and the utility-maximizing nested logit (UMNL) formulations, for estimating the nested logit (NL) models with combined revealed preference (RP) and stated preference (SP) data. The paper demonstrates how to correctly set up artificial tree structures for estimating mixed RP/SP NL model using the NNNL and UMNL formulations and provides formulas for recovering utility function, dissimilarity and scale parameter estimates. The estimations and correction procedures for mixed RP/SP NL models are empirically illustrated. The proposed estimation structures and correction procedures can be applied to other nested structures with multiple data sources.
Samenvatting