The influence of walking accessibility on station-to-station passenger flow and its interaction with metropolitan race/class segregation: A case study of MARTA’s heavy-rail network, Atlanta (USA)

Mass transit is a key transport strategy in helping cities decarbonize, adapt to an era of rapid climate change, and guide rapid urbanization. Central to transit planning is the ability to accurately estimate demand for an effective, efficient, and equitable infrastructure and services. Instrumental to this effort is direct-demand modelling (DDM), which has evolved to become more nuanced in predicting ridership at station-level and station- to -station levels and in shedding light on key ridership and performance determinants. Local and Metropolitan accessibility as predictors of transit patronage has been shown significant in recent DDM studies at station-level, with an apparent synergistic relationship. This, however, has not been explored on a station- to -station passenger flow level. In several ways this is a more valid unit of analysis for rail ridership studies as it captures critical factors between and at both ends of the trip that are experienced by the passenger. It is also well documented that the sensitivity of passengers to key ridership determinants varies across income levels. In some jurisdictions income level strongly correlates with race/ethnicity and/or class, due in part to historical legacies of classism and/or racism. Segregation because of class and/or race prejudice, often found in US cities, might yield spatial heterogeneity in whole-network DDM model parameters and introduce bias that could potentially mislead transit analysts, policy makers, and systemwide effectiveness. We explored and tested these possibilities and considered modelling and policy implications as we leveraged Atlanta’s legacy of racial and income segregation in studying MARTA’s Origin-Destination (O-D) passenger flow patterns. First, a potential synergistic relationship between origin-stations’ and destination-stations’ walking accessibility levels was tested. Disparities, if any, in this and other ridership determinants were then explored between two distinct sets of O-D pairs whose origin Pedsheds accommodate majority-white or majority-nonwhite residents. Comparison and testing using generalized crossed-effects modelling reveals important differences in fit, magnitude, and significance of some parameters across submodels and as compared to the whole-network model. We also identified distinct moderating effects of distance between O-D pair stations and walking accessibility levels across submodels. In racially- and/or class-segregated cities planners would benefit from developing race- and/or class-based DDM submodels that would likely yield less biassed parameters; improve our understanding of rail transit patronage determinants; and help in crafting more effective and equitable transit and land-use policies.