Final Analytical Comparison of Aggregate and Disaggregate Mode Choice Models Transferability

Abstract

Transportation models as tools for transportation planning are critical to such related decisions. Considering the high cost of calibrating and validating such models, effective alternatives are highly sought for; one such alternative being the use of models calibrated for other cities. This calls for transferability analysis which has not been the subject of many researches. Due to criticality of aggregate and disaggregate data in transportation models, this paper tries to compare transferability of models calibrated with data of both groups. Mode choice models for daily work trips in two real-sized cities of Qazvin and Shiraz are analyzed. Models are calibrated employing multinomial logit structure with four modes of private car, taxi, bus, and 2-wheelers. In order to increase reliability of results, the top five best models are selected for each city-data category to be transferred. Based on transferability test statistics, transfer index, and goodness-of-fit of transfer models, aggregate models are not transferable and their results are deceptive. Transferability measures of these models are not in acceptable range; whereas transferability of disaggregate models have relative proper response. According to transfer index and goodness-of-fit of origin models operate similar to destination models. However transferability test statistics rejects the assumption of equality coefficients in both cities models. Using personal variables helps to effectively transfer origin models in addition to improve them.