Agreement between a web collaborative dataset and an administrative dataset to assess the retail food environment in Mexico

Abstract Background Latin American countries are often limited in the availability of food outlet data. There is a need to use online search engines that allow the identification of food outlets and assess their agreement with field observations. We aimed to assess the agreement in the density of food outlets provided by a web collaborative data (Google) against the density obtained from an administrative registry. We also determined whether the agreement differed by type of food outlet and by area-level socioeconomic deprivation. Methods In this cross-sectional study, we analyzed 1,693 census tracts from the municipalities of Hermosillo, Leon, Oaxaca de Juarez, and Tlalpan. The Google service was used to develop a tool for the automatic acquisition of food outlet data. To assess agreement, we compared food outlet densities obtained with Google against those registered in the National Statistical Directory of Economic Units (DENUE). Continuous densities were assessed using Bland–Altman plots and concordance correlation coefficient (CCC), while agreement across tertiles of density was estimated using weighted kappa. Results The CCC indicated a strong correlation between Google and DENUE in the overall sample (0.75); by food outlet, most of the correlations were from negligible (0.08) to moderate (0.58). The CCC showed a weaker correlation as deprivation increased. Weighted kappa indicated substantial agreement between Google and DENUE across all census tracts (0.64). By type of food outlet, the weighted kappa showed substantial agreement for restaurants (0.69) and specialty food stores (0.68); the agreement was moderate for convenience stores/small food retail stores (0.49) and fair for candy/ice cream stores (0.30). Weighted kappa indicated substantial agreement in low-deprivation areas (0.63); in very high-deprivation areas, the agreement was moderate (0.42). Conclusions Google could be useful in assessing fixed food outlet densities as a categorical indicator, especially for some establishments, like specialty food stores and restaurants. The data could also be informative of the availability of fixed food outlets, particularly in less deprived areas.