This paper uses census tract-level data to determine the presence, density, and popularity of U.S. food retailers. We merge census tract-level demographic data from ACS, neighborhood amenities from heterogenous sources, and Point of Interest (POI) data from anonymized cellphone GPS ‘pings’ to identify food retailer location and foot traffic information.

Sequence: ACS -> Merge -> Regressions

Downloads ACS 2010 5-year average census tract level demographic data + 2010 dicennial census data. Merges census tract-level data on walkability scores, job density, transit availability, road network, local sales tax and county-level crime statistics.

Same as above but with 2019 data.

Merges with Safegraph point-of-interest data (store count) and store visit frequency data.

• Correlation plot
• t-test
• Logit
• Gradient Boosted Logit
• Gradient Boosted Logit w/ balancing
• OLS
• Gradient Boosted OLS

• Negative binomial regression
• Zero-truncated negative binomial regression

Crime data

Total list of variables in final data