Comparative analysis of Multinomial logit (MNL) and Nested logit (NL) modeling approaches for mode choice decisions.

Abstract

Number of methods have been used globally for predicting the travel mode choice, including Multinomial Logit (MNL), Nested Logit (NL), Multinomial Probit (MNP), Generalized Extreme Value (GEV), Mixed Multinomial Logit (MMNL) and Artificial Neural Network (ANN) modelling approaches. Among them, MNL and NL approaches are predominant. However, the assumptions in MNL regarding the independence error components might not hold in developing countries like Sri Lanka. NL partially relaxes this assumption. The study gathered trip-specific, socio-economic, and household data via online and face-to-face questionary surveys. After meticulous parameter selection and cross-sectional analysis, both MNL and NL models were developed and compared. Buses (46%), motorbikes (19%), trains (12%), and cars (11%) were the dominant modes. Notably, total travel time significantly influenced mode choice, with NL model exhibiting superior accuracy over MNL. Recognizing the weight of this attribute informs urban planning, policy formulation, and transportation system optimization.