Built environment effects on travel behaviors of internal migrants in urban China : a comparative perspective

Abstract

Hundreds of millions of internal migrants reside and work in cities in the developing world. Accurately understanding their travel behaviors and disentangling the effects of their influencing factors have become increasingly indispensable for formulating and executing socially inclusive transportation plans and policies. However, these tasks are challenging. In developing countries, inclusiveness and social justice tend to be overshadowed by the economic growth imperative, and voices of the disadvantaged, such as internal migrants, are liable to be neglected in local (transportation) policy-making processes. Moreover, in academia, although researchers, especially those in the West, have paid increasing attention to migrants’ travel behaviors concerning their demands, preferences, problems, and determinants, as well as a distinctive process of “transportation assimilation,” scholarship on internal migrants’ travel behaviors remains scarce, particularly in the context of developing country. Therefore, adopting a comparative perspective, this study aims at revealing internal migrants’ travel behaviors and their built-environment determinants, as well as their differences with those of local residents (referred to as locals hereafter) in urban China. It proposes a central hypothesis: in terms of internal migrants and their counterparts in the same city, i.e., locals, considerable differences exist in not only their travel behaviors but also in the relationships between their travel behaviors and the influencing factors. Three research questions are proposed: a) How does the built environment affect travel behaviors of internal migrants and locals differently, in terms of statistical significance? b) What are the differences in the magnitudes of effects of the built environment on travel behaviors between internal migrants and locals? c) What are the differences in the shapes of those effects on travel behaviors of such two groups? This thesis conducts three empirical analyses in Xiamen, China, which accommodates a large number of internal migrants. The data used in the analysis include built-environment, socio-demographic, and travel behavior data. The latter two types of data are obtained from a large-scale survey conducted in Xiamen in 2015. Employing a multilevel zero-inflated negative binomial regression model (MLZINB), it first investigates the trip frequencies of four major travel modes of internal migrants and locals and comparatively reveals their respective built-environment determinants. Then, comparing a state-of-the-art machine learning method, i.e., light gradient boosting model (LightGBM) and a type of discrete choice model, i.e., multinomial logit model (MNL), it predicts the commuting mode choice of such two groups and examines the effect size of built-environment variables. Finally, relaxing the linear or generalized linear assumptions about relationships between the built environment and travel behaviors and employing the random forest method (RF), it delves into the shapes of effects of built-environment variables on commuting duration of the two groups. The key findings tell a comprehensive story concerning the travel behaviors of internal migrants in urban China. First, the migrants obviously have a lower socioeconomic status than locals, and they make fewer daily trips and are much more dependent on active and public transport. While locals’ trip frequencies are affected by both the socio-demographic and built-environment variables, those of migrants are mainly influenced by the socio-demographics; in addition, effects of built-environment variables on migrants’ travel behaviors are much less significant. Second, for the commuting mode choice of both migrants and locals, the built environment as a whole plays a more important role than the socio-demographics. Moreover, the built environment has larger effects on commuting mode choice of locals than on that of migrants, whereas the opposite applies to the effects of socio-demographics. Third, on average, internal migrants have slightly shorter commuting duration than locals. Moreover, all the built-environment variables influence commuting duration of both groups in non-linear patterns, and threshold effects obviously exist. Differences are observed in the shapes, gradients, and threshold values of effects of built-environment variables on commuting durations of internal migrants and locals. This study sheds light on the differences in travel behaviors of internal migrants and locals and the effects of the built environment, which can have significant implications for researchers and policy makers in addressing the transportation inequality issues, accommodating transportation demands of both migrants and locals, and building more socially inclusive and resilient cities.