Recipient Organization
UNIVERSITY OF FLORIDA
G022 MCCARTY HALL
GAINESVILLE,FL 32611
Performing Department
(N/A)
Non Technical Summary
Considerable prior research attempts to untangle the impacts of neighborhood food environments on diet and diet-related health disparities. Studies in the agricultural policy literature tend to focus on the presence and density of grocery stores, ignoring that the nutritional quality of offerings can vary considerably from store-to-store; studies in the nutrition literature analyze differences in food offerings at the store-level, but do so only for small geographic areas. Due to the high cost of constructing store-level measures of available foods, few national-scale studies account for the distributions of food retailers, their offerings, and prices.Consolidation and proliferation of non-traditional store formats such as supercenters, natural/gourmet groceries, and dollar stores in recent decades have changed the US food-at-home retail landscape. Recent studies indicate store offerings, prices, and consumer purchases vary across these formats, highlighting the need to account for differences in both store offerings and the distribution of retail formats when conducting policy-relevant research.To this end, the objects are to: 1) developstore-level measures of the healthfulness of food offerings that account for prices constructed using validated and reliable questionaries, store-level scanner data and product nutrition information; 2) identifymarket boundaries by store format using cellular data; 3) leveragefindings from 1 and 2 to evaluate how the healthfulness of offerings are associated with geography, market concentration, and sociodemographic characteristics; and 4) usevariation in the healthfulness of food offering in markets across time and space to causally identify how the neighborhood food-at-home retail environment impacts the healthfulness of food purchases and diet-related health outcomes.
Animal Health Component
25%
Research Effort Categories
Basic
(N/A)
Applied
25%
Developmental
75%
Goals / Objectives
In the US, 41.9% of adults are obese, putting them at risk for heart disease, stroke, type 2 diabetes, and certain types of cancer, with propensities of overweight and obesity and associated diet-related diseases being higher for minority and socioeconomically disadvantaged groups (Stierman et al. 2021). Considerable prior research has attempted to understand the role food environments play on diet and diet-related health outcomes and their contribution to observed health disparities. Studies find low-income and minority neighborhoods are disproportionately areas of low food access (Larson, Story, and Nelson, 2009; Hilmers, Hilmers, and Dave, 2012). To date, prior studies analyzing food access primarily consider the presence and densities of grocery stores and other food retailers, ignoring that the nutritional quality of food product offerings can vary considerably within retail formats. In fact, the USDA's Food Access Research Atlas defines areas of low access based on the presence of large supermarkets and supercenters(USDA-ERS, 2022). Those studies that do measure the healthfulness of food offerings at the store-level utilize only small samples of retailers in specific geographic areas (Cannuscio et al., 2013; Shaver et al., 2018) or focus on the availability of certain foods such as fruits and vegetables or specific product characteristics such as whole grains or added sugar (Chenarides and Jaenicke, 2019; Allcott et al., 2019).In Glanz et al.'s (2005) seminal study, the authors draw a distinction between two types of neighborhood food environments: community and consumer nutrition environments. Community nutrition environments account for the distribution of food retailers (number, type, location, and accessibility of food outlets consisting mainly of food stores and restaurants), while the consumer nutrition environment reflects the range of food choices, their nutritional qualities, and prices that consumers encounter within a retail food outlet. While there are numerous studies that analyze the impacts of community or consumer nutrition environments separately, there is a dearth of studies that analyze both at the national scale. Filling this gap, the long-term goal of this research project is to create and leverage a novel measure of the healthfulness of food product offerings constructed from store-level scanner data and combine this measure with information about the distribution of food retailers and shopping patterns to causally estimate the link between the neighborhood food-at-home (FAH) retail environment and dietary and diet-related health outcomes.An inherit assumption in studies focusing on the community nutrition environment is that similar food retailers in the same retail category (e.g., supercenters, supermarkets, etc.) offer similar food options to consumers at comparable prices. However, this is unlikely to be the case, particularly in the face of consolidation and diversification that has occurred within the food retailing industry over the last few decades. These changes in the US food retailing landscape highlight the need to understand how the nutritional quality of food offerings vary across and within retail formats.Glanz et al. (2005) laid the foundation for the development of the tool that has become the most commonly used measure of the consumer food environment - the Nutrition Environment Measures Survey (NEMS) (Glanz et al., 2007). McKinnon et al. (2009) and Lytle and Sokol (2017) provide systematic reviews of the measures used to assess a variety of consumer food environments including food-at-home retailers. Their reviews include measures constructed using interviews, inventories, checklists, and market basket instruments and numerous other methodologies. Although Lytle and Sokol (2017) acknowledge that scanner data can be used to assess food environments, none of the measures included in their review used store-level scanner data. Rather, the instruments designed for assessing store environments were typically paper-based forms completed by trained data collectors with the most commonly used measure being the NEMS. Since it was first developed, this survey has been adapted to include stores, corner stores, and restaurants, modified to reflect a variety of diets including the Australian, Mediterranean, and Costa Rician diets (Golfin et al., 2017; Martínez-García et al., 2020; Whelan et al., 2018), and employed as the measure (or a modified version) of the consumer nutrition environment in over 190 peer-reviewed studies (Glanz et al., 2023).The Nutrition Environment Measures Survey in Stores (NEMS-S) and the Nutrition Environment Measures Survey in Corner Stores (NEMS-CS) are accurate, yet costly, measures of the healthfulness of food items available at food retailers. To construct these measures, trained data collectors physically visit the store of interest and survey the products available on the shelf, recording information on the availability of products and healthier alternatives, prices, and quality of fresh fruits and vegetables. Due to the high cost of constructing the NEMS, studies that employ this measure do so for relatively small samples of stores in close proximity to each other.Even with small samples, there is evidence that NEMS-S varies considerably across and within retail formats. A study of 373 stores in an 18-square-mile section of West and Southwest Philadelphia finds the average NEMS-S for large chain supermarkets was 38.4 (st. dev. = 7.6), while the average NEMS-S was 21.6 (st. dev. = 4.75), and 11.1 (st. dev. = 6.10) for medium-size grocery stores and corner stores (including chain pharmacies and dollar stores), respectively (Cannuscio et al., 2013). Possible NEMS-S scores range from -9 to 54; higher scores indicate more diverse and healthful food items. The large standard deviations suggest significant variation in the healthfulness of product offerings across stores within the same store format, especially within the group of corner stores. These findings further suggest that store-level measures are necessary to truly understand food availability and access. Furthermore, there is evidence that the healthfulness of offerings varies across sociodemographic characteristics of neighborhoods within the same city. Shaver et al. (2018), using a modified version of NEMS, find the average modified-NEMS score in predominantly Caucasian neighborhoods of Flint Michigan is 29.7 and 24.1 in predominantly African American neighborhoods, with a larger standard deviation for African American neighborhoods.We propose four interrelated objectives. In Objective 1, we propose constructing NEMS-SCAN scores for retailers in the IRI OmniMarket Core Outlets (formerly InfoScan) database. In Objective 2, we proposed using SafeGraph cellular phone data to examine shopping patterns to determine market boundaries by store format. In Objective 3, we propose using the NEMS-SCAN scores generated in Objective 1, market boundary data from Objective 2, market concentration data, sociodemographic data, and machine learning techniques to impute the NEMS-SCAN scores for each retailer in the NielsenIQ TDLinx database. While the IRI database contains detailed product-level sales data, it covers only a fraction of all FAH retailers. TDLinx coverage is superior, however, product-level data are not reported. In Objective 4, we propose using the imputed NEMS-SCANS scores from Objective 3 and market boundary information from Objective 2 to construct market-level measures of the healthfulness of foods available at FAH retailers; these measures will capture both the community and consumer neighborhood FAH retail environment. We then propose using these measures to causally estimate the impacts of the neighborhood FAH environment on the healthfulness of food purchases and diet-related health outcomes.
Project Methods
Objective 1: Develop a new store-level measure of the healthfulness of food offerings, NEMS-SCAN, based on the Nutrition Environment Measures Survey (NEMS) that is constructed using retail-level scanner data.Using the NEMS instrument, IRI OmniMarket Core Outlets data and the Purchase to Plate Crosswalk (PPC), we will develop two versions of the store-level measure of the healthfulness of food offerings constructed from weekly store-level scanner data. The two versions, NEMS-SCAN-S and NEMS-SCAN-CS, correspond to the current NEMS-S and NEMS-CS data collection and scoring procedures, respectively. NEMS-S was designed to assess the nutritional environment at grocery stores, while the NEMS-CS was designed to assess smaller footprint corner stores. Given the proliferation of dollar stores and other non-traditional formats, the NEMS-CS measure is better equipped to discern variability in the food offerings across this growing segment of the food retail space.IRI OmniMarket Core Outlets dataset reports weekly sales value and quantities of food sold by item and by retail store for the years 2008 to 2021. Grocery, club, supercenter, convenience, dollar, and drug stores are represented in the dataset. The Purchase to Plate Crosswalk (PPC), created by the USDA, allows IRI OmniMarket Core Outlets data to link to the Food and Nutrient Database for Dietary Studies (FNDDS) and the National Nutrient Database for Standard Reference by UPC (Carlson et al., 2022). The FNDDS is a database containing detailed information about the nutrient values of foods and beverages.Objective 2: Use foot traffic data to examine shopping patterns and the characteristics of shoppers at the various stores to determine market boundaries by store format. We will leverage foot traffic data to examine characteristics of shoppers across different stores in two ways. First, we will analyze the distances traveled by patrons to stores to estimate travel associated with food shopping and the market boundaries of store types. Second, we will analyze same-day and same-week related brands to identify the extent to which patrons visit multiple food outlets in a single day or single week. We refer to this phenomenon as "multi-store shopping." We will use data on food establishments from SafeGraph "Places" and "Patterns" data products for 2018 - 2021 to complete this portion of our analysis. SafeGraph is a private company that uses cell phone data, aggregated to the establishment level, to identify variables related to these points of interest. Key variables in the "Places" data are the location of points of interest (latitude and longitude as well as street address), North American Industry Classification System (NAICS) code for the point of interest, and hours of operation. Key variables in the "Patterns" data include hourly, daily, and weekly foot traffic from a select panel, home census block groups of visitors from the panel, and same-day and same-week brands of establishments visited by the same panel visitors who visited the point of interest. To identify distances traveled for food store visits, and to therefore determine the market boundaries or "draw areas" of different store types, we will employ the SafeGraph "Patterns" data and summarize the distributions of travel distances from visitor census block group centroids to the exact locations of the establishments they visit. We will consider different store types, e.g., traditional grocery stores, supercenters, club stores, dollar stores, convenience stores, or drug stores. We will identify the market boundaries by store type characterized by median distance traveled as well as the market boundaries characterized by maximum distance traveled.To determine the extent to which shoppers visit a multitude of food stores, we will examine the same-day and same-week related brands variables. We will use a sample of local and regional grocery stores and tally the number of visitors who also visited large, national chains. We will repeat this exercise for large, national chain visitors who visit other large, national chains. Large, national chains will be determined by the store count and geographic coverage in the SafeGraph "Places" data.Objective 3: Employ NEMS-SCAN to examine how the healthfulness of food offerings varies across retail formats, market concentration, demographics, geography, and SNAP-authorization status.Leveraging the NEMS-SCAN scores calculated using the IRI OmniMarket Core Outlets data as well as indicators computed from several different data sources described below, we will explore the associations between the healthfulness of store-level food offerings as measured by NEMS-SCAN and local market structure, demographics, SNAP participation, and geography. Each store will have an individual NEMS-SCAN score and will also be assigned market structure, demographic, SNAP, and geographic characteristics based on location using U.S. Census Bureau's (2023a) American Community Survey (ACS) data. Market structure characteristics will depend on concentration as measured at the market level; we will use NielsenIQ TDLinx data to construct these measures. NielsenIQ's TDLinx dataset, covering the years 2004 - 2022, is a geocoded census of US food retailers. TDLinx is the most effective and comprehensive resource of which we are aware for estimating market concentration, market share, and changes in market structure over time.To test the robustness of our findings, we will use a variety of definitions of markets including those prevalent in the extant literature, those based on distances or census block groups, and those developed from Objective 2's findings. Since the IRI data only covers a fraction of stores, we propose developing a predictive model of NEMS-SCAN scores for stores in the IRI data and using the resulting model to impute NEMS-SCAN scores for stores in TDLinx. The predictive model will exploit associations between NEMS-SCAN and store format, market concentration, sociodemographic characteristics, and geography.?Objective 4: Employ NEMS-SCAN and our understanding of shopping patterns to estimate the causal effect of the neighborhood FAH environment on food purchases and diet-related health outcomes.We propose using household-level panel data contained in the IRI Consumer Network database, imputed store-level NEMS-SCAN scores from Objective 3, and our understanding of shopping patterns to analyze the causal relationship between the healthfulness of food offerings and food purchases. A subset of IRI Consumer Network panelist complete surveys containing medical questions. We will use these data to estimate the causal relationship between the healthfulness of food offerings in the neighborhood food environment and health outcomes. Since individual medical data are only available for a subset of households, we will analyze how our census-tract NEMS-SCAN measures relate to incidence of obesity, high cholesterol, high blood pressure, diabetes and heart disease in adult populations (≥ 18 years old) using publicly available data. The 500 Cities Project, a collaboration between the Center for Disease Control (CDC), Robert Wood Johnson Foundation, and CDC Foundation, provides city- and census tract-level estimates for chronic disease risk factors and health outcomes for the largest 500 cities in the US (CDC, 2021 & 2023). In December 2020, the 500 Cities Project was replaced by the PLACES project. The PLACE Project expands the data coverage to counties, places, census tracts, and ZIP Code Tabulation Areas (ZCTA) across the entire US. These projects datasets contain health outcomes including rates of obesity, heart disease, diabetes, and high cholesterol for adults at the census-tract level for the years 2016 - 2022.