Grocery shopping behavior in Detroit neighborhoods experiencing disinvestment and decline: An empirically grounded agent-based model of data-scarce communities

This paper presents a data-driven agent-based model that simulates the weekly grocery shopping behavior of disadvantaged consumers in highly segregated lower eastside neighborhoods of Detroit, Michigan. We focus on neighborhoods experiencing severe disinvestment to analyze the shopping behavior of residents after all major regional and national supermarket chains abandoned the city. The presented model is unique in that it utilizes detailed shopping behavior data collected to examine travel in marginalized communities, specifically among residents in severe poverty who are often overlooked in the travel behavior literature. The research shows that in extreme socio-economic decline, sociodemographic variables (such as class) can become more relevant than the built environment (land-use mix, density, and street connectivity) in determining access and influencing mobility. After identifying unique groups of household agents, we design rules that utilize probability distributions generated from survey responses. The decision-making of agents that emulate households is habitual rather than utility driven. Modeled behavior is designed based on stated preferences, which may contradict premises such as the “shortest distance to the nearest shop” approach, a common assumption in the literature. We also report on three what-if scenarios to evaluate how major population changes would affect the results.