Potentials and Limitations of Large-scale, Individual-level Mobile Location Data for Food Acquisition Analysis

Duanya Lyu, Luyu Liu, Catherine Campbell, Yuxuan Zhang, Xiang YanarXiv 2025 · DOI · arXiv

Introduction

Understanding food acquisition patterns is critical for addressing food insecurity. While traditional methods like surveys and interviews provide deep insights, they are often limited by recall bias, small sample sizes, and high implementation costs. The emergence of large-scale mobile location data (GPS traces) offers a high-resolution alternative for tracking shopping behavior. This paper evaluates the potential of using this "big data" source, specifically examining how data processing choices—such as the spatial criteria used to define a "visit"—impact the resulting food acquisition metrics.

Methodology

The researchers developed a multi-step pipeline to transform raw GPS pings into meaningful food acquisition metrics. The process begins with inferring home locations and activity "stays" using the Trackintel package. Food outlets are then classified based on both the types of products sold and "behavioral confidence," which distinguishes between primary food-selling locations (e.g., supermarkets) and mixed-purpose locations (e.g., gas stations).

Methodological pipeline for processing GPS data into food acquisition metrics.

A core component of the methodology is the Buffer-Based Visit Identification. To determine if a recorded "stay" constitutes a store visit, the researchers applied varying search radii based on store size—for instance, 50 meters for small outlets and up to 200 meters for big-box stores.

Illustration of buffer-based visit identification using varying search radii.

Data and Study Area

The study utilized a case study of Jacksonville, Florida, incorporating:

  • GPS Data: 286.4 million records from approximately 94,000 devices (Sept 1 – Oct 15, 2022).
  • Food Outlet Data: A comprehensive database of retail outlets from the University of Florida GeoPlan Center.
  • Demographics: U.S. Census Bureau ACS data used to evaluate the representativeness of the GPS sample.

Results and Discussion

The analysis successfully identified over 852,000 food outlet visits. Key findings include:

  • The "Bypass" Phenomenon: Data indicated that individuals frequently bypass the nearest food store to shop elsewhere, suggesting that factors like price, quality, or cultural preference often outweigh simple proximity.
  • Algorithmic Sensitivity: Visit identification for small stores proved extremely sensitive to the search radius. Expanding the radius from 50m to 100m often introduced "noise" from passersby or visitors to adjacent shops.
  • The Underestimation Gap: GPS data captured roughly 1.4 visits per week per person, a rate significantly lower than those reported in traditional surveys (e.g., USDA FoodAPS). This suggests that while GPS data is excellent for identifying spatial patterns, it may underestimate absolute visit frequencies.
  • Trip Characteristics: The average network distance to a visited store was 7.61 km. Only 18% of visits were "home-based," highlighting a limitation where GPS data may miss trip origins if devices are not consistently active.

Limitations and Conclusion

The study identifies several critical hurdles for GPS-based analysis, including sampling bias (under-representation of elderly and disadvantaged populations) and the challenge of tracking household-level activities with individual-level device data. The authors conclude that a "mixed-methods" approach—combining large-scale GPS data for spatial patterns with high-fidelity surveys for ground-truthing—is the most robust path forward for food acquisition research.

Summary updated

BibTeX
@misc{lyu2025potentials,
  author = {Duanya Lyu and Luyu Liu and Catherine Campbell and Yuxuan Zhang and Xiang Yan},
  title = {{Potentials and Limitations of Large-scale, Individual-level Mobile Location Data for Food Acquisition Analysis}},
  year = {2025},
  eprint = {2503.18119},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2503.18119},
}
Figure