Measuring and Improving Public Transit Accessibility Using Large-Scale Mobility Data

<p dir="ltr">Among transit practitioners, the consensus tenets for operationalizing transit equity are to 1) comprehensively identify factors that have contributed to transportation marginalization, 2) establish channels to solicit and incorporate feedback from populations impacted by transit projects, and 3) recenter planning and benchmarking around metrics that faithfully represent riders’ lived experiences on and resulting from transit. Researchers have successfully identified factors contributing to transport deprivation, formulated more accurate and informative accessibility metrics, and framed accessibility metrics as the ideal successors to traditional planning metrics. However, few of these advancements have meaningfully penetrated the practical sphere due to their complexity and prohibitive data requirements. To these ends, we present studies in three transit contexts–fixed-route systems under real-world service variability, demand-responsive systems in rural areas, and multi?modal first-mile station access. Each study proposes a novel, generalizable methodology designed to simultaneously solve a critical theoretical gap in existing accessibility analyses and facilitate adoptability by practitioners. </p><p dir="ltr">First, we introduce algorithms for fixed-route systems that sequentially process the scheduled General Transit Feed Specification (GTFS) and Automatic Passenger Count and Automatic Vehicle Location (APC-AVL) data to more realistically model the origin-destination (O-D) trip planning and travel experiences of riders. We produce O-D travel time estimates that account for real-world schedule deviations and use them as the foundation of the novel “achieved accessibility” metric, which quantifies the proportion of planned cumulative opportunity accessibility implied by the transit schedule that is actually realized when service variability is considered. In a case study, we estimate the achieved accessibility for riders traveling from socioeconomically disadvantaged locations in Pittsburgh, PA to local employment hotspots from 2016-2020. Findings collectively demonstrate the benchmarking utility of achieved accessibility and include O-D pairs realizing as low as 15% of planned accessibility in practice, and O-D pairs with low or decreasing achieved accessibility over time despite predominantly involving bus routes with high on-time performance. </p><p dir="ltr">Next, we address the absence of a widely-accepted accessibility metric for rural demand-responsive transit (DRT) systems by proposing the “rural accessibility index” (RAI), composed of three scores iv quantifying DRT service quality, socioeconomic disadvantage, and car-related economic stress (CRES) by origin census tract. To enable adoption of the RAI, we provide an algorithm that can be used to estimate O-D travel times for DRT service based on historical system trip data, as well as methods to estimate CRES prevalence using census data. Through a case study in Greene County, PA we demonstrate use of the RAI to prioritize tracts for system improvements and use of the component scores to identify specific causes of inaccessibility in each tract. We also benchmark the RAI against popular urban cumulative opportunity accessibility metrics and observe significant discrepancies, illustrating how urban metrics can break down in rural contexts and the need for a more context-sensitive metric for rural transit. </p><p dir="ltr">Lastly, we propose a framework for comprehensively modeling first-mile mode choice and predicting the mode share changes associated with an extensive set of best-practice accessibility improvements. Our model formulation accounts for traditional and emerging transport modes as well as route-level factors that can influence first-mile route and mode choice. We streamline adoption by intentionally extracting choice attributes from public data sources and providing a method to estimate the choice attributes associated with depersonalized/aggregated O-D trip records using open-source software. Empirical comparisons to best-in-class first-mile mode choice models indi?cate that our model formulation can yield predictive accuracy competitive with models that rely on either cost-prohibitive survey and GPS trace data or opaque deep learning formulations. We apply the model to a case study of planning the optimal accessibility improvements to shift rail riders from driving to active first-mile modes in Washington, DC. We illustrate the model’s ability to dramatically upscale transit agencies’ capacity to evaluate improvement scenarios and advocate for collaborations with other organizations. Scenario comparisons show the model’s potential to identify sets of complementary improvements that transit agencies, local government departments, and private transportation service providers can collectively make to generate surplus mode share shifts.</p>