Food environment data sources in Berkeley, California - A spatial analysis of four data sources

INTRODUCTION: Certain restaurant densities are used as proxies for exposure to healthy or unhealthy foods. While different datasets have been used to quantify restaurant density, the differences between these datasets have not been systematically examined. We compared four sources of "restaurant" listings in Berkeley, California, assessing their correlation, and the variability in their associations with local aspects of the social environment as quantified by census data.

METHODS: Current restaurants from the Yellow Pages (YP), Yelp, Dun & Bradstreet (D&B), and parcel data were geocoded to the 63 census tracts in Berkeley, California. For each data source, restaurant density (number per square mile) was calculated for each tract. Census data on population density, age, proportion of minority, and worker density were also assigned to tracts. Linear regression was used to investigate the associations between restaurant density estimated from different sources and census data.

RESULTS: We found large variations by data source in the total number of restaurants, with 423 identified from Yelp, 478 from YP, 696 from D&B, and 56 from parcel. Despite these differences, all four sources were correlated (Pearsons correlation >0.68 between restaurant densities). Consistent associations were found between restaurant densities and social factors for the different data sources.

DISCUSSION: The four databases varied in the number of restaurants identified, likely representing different sources of bias. Despite estimates being correlated, researchers should be clear about the databases used and their biases when quantifying measures like restaurant density. Without a "gold standard" database, studies may benefit from sensitivity analyses.