Urban Mobility and Urban Form: Understanding Mobility Patterns of the New York City Taxi System Using Latent Factor Models

Abstract

Understanding human movement through urban spaces can offer actionable insight for various policy outcomes and is increasingly critical as the future of cities see heightened demand for urban space. This research develops a latent factor model to analyze how taxi travel reflects New York City urban relationships, and investigate the association between racial and socioeconomic similarities and affinity between urban spaces. Urban mobility was modeled through interpretable latent parameters including: travel distance cost, desirability of destinations, and affinity between locations. The model was optimized using Stochastic Gradient Descent, validated through parameter recovery analysis on a constructed synthetic data set, and ultimately fit to almost 300 million trips from the New York City Taxi System. The learned model parameters represent a novel data source of latent factors of human travel which were analyzed alongside New York City geospatial data and US Census Data. The desirability and popularity of New York urban spaces were found to be distinct, and several areas within the exterior regions of The Bronx and Queens were identified as potentially underserved by current taxi supply. In addition, correlations between the affinity of two locations and similarities in socioeconomic and racial indicators were strongest for travel between populations that were the least-educated, lowest-income, and had the highest portions of African American and Hispanic residents; little evidence of association between social indicators and location affinity was found for other populations. Overall, similarities in racial composition, income level, and population density were the most important in the model, after accounting for proximity and desirability, to understand the strength of connection between urban spaces in New York City.