Source: TEXAS TECH UNIVERSITY submitted to NRP
IMPACT OF FOOD RETAILERS` PRESENCE AND COMPOSITION ON NUTRITIONAL EQUITY AND HEALTH OUTCOMES IN THE UNITED STATES WITH MACHINE LEARNING
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
ACTIVE
Funding Source
Reporting Frequency
Annual
Accession No.
1028466
Grant No.
2022-68015-37142
Cumulative Award Amt.
$273,178.00
Proposal No.
2021-08220
Multistate No.
(N/A)
Project Start Date
Apr 15, 2022
Project End Date
Apr 14, 2026
Grant Year
2022
Program Code
[A1344]- Diet, Nutrition and the Prevention of Chronic Disease
Recipient Organization
TEXAS TECH UNIVERSITY
(N/A)
LUBBOCK,TX 79409
Performing Department
Agricultural and Applied Econo
Non Technical Summary
A number of studies find evidence of nutritional inequality in the United States--that is the association between low access to healthful food and low income or non-white population. However, the findings are not causal: we do not know whether the residents of the low-income non-white neighborhoods are choosing unhealthful diets because of the lack of access, or large food retailers with a variety of healthful options choose not to locate in the neighborhood due to lack of demand. The current project conducts an in-depth analysis for policy generation. First, we derive the conditional probabilities of each type of food retailer, e.g., grocery, fast-food restaurants, that locates in a certain region, given the demographic and regional features. We use cellphone mobility data to track available food options in a region, and machine learning models for improved prediction. Second, we calculate the rate of substitution between two types of food retailers to understand the dietary preferences of residents. If one type of food retailer is present, we assess how it affects residents' dietary and nutritional choices, such as fresh vegetable consumption. We use COVID-19 lockdown and store closures to obtain the causal inference. Household food purchase information will be obtained from scanner data. Finally, we evaluate the impact of the presence of various food retailers and their composition in the neighborhood on diseases that might be influenced by dietary choices, such as diabetes, obesity, and cardiovascular diseases. The project will contribute to retail food policy and health policy in the United States. Understanding the substitution between food retailers will help to design a resilient local food sector that might combat a future disease-induced pandemic.
Animal Health Component
100%
Research Effort Categories
Basic
0%
Applied
100%
Developmental
0%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
60750103010100%
Knowledge Area
607 - Consumer Economics;

Subject Of Investigation
5010 - Food;

Field Of Science
3010 - Economics;
Goals / Objectives
The overarching goal of the proposed project is to examine and understand the impact of access to different food retailers and their regional distribution on nutrition equity and health outcomes in the United States. We use machine learning and artificial intelligence models to offer precise, local, and customized recommendations. The supporting objectives of the proposed project are to:1. Assess the equity of access to healthful foods. We will analyze the factors that explain location choice by food retailers, e.g., full-service and limited-service restaurants, supercenters, groceries, and convenience stores, and assess whether the location choices of these retailers are equitable across census tracts in the United States. We will evaluate the likelihood of access to healthful retailers, and how the distribution of the retailers changes across regional demographic features for lower-income neighborhoods, predominantly non-white neighborhoods, and rural areas.2. Estimate the causal impacts of various food retailers on nutritional choices. The variations in lockdowns associated with the COVID-19 pandemic provide an opportunity to obtain statistically valid estimates of the causal impacts of various food retailers on nutritional choices. The variations in lockdowns create natural experiments with treatment areas and control areas. We can check the difference in diet patterns within (before-after) restaurant closures in a neighborhood and/or between the control and treatment groups. The lockdown allows us to calculate the perceived substitutability between two store-types, e.g., supercenter versus fast-food restaurants.3. Analyze impacts of healthful food access on health outcomes. The third objective is to measure the association between healthful food access and the level of vulnerability to diseases that are partly influenced by diet patterns, e.g., heart disease, diabetes, and obesity.
Project Methods
We will combine several large data sets for this study at the census tract and county level. Dietary choices and demographic controls will be derived from the IRI Homescan data. The participating households use a barcode scanner to record their food purchases from stores, e.g., grocery stores, supercenters, club stores, drug stores, convenience, and health food stores. The data contain purchasing price, quantity, and product attributes. Household demographic characteristics are grouped and presented in the data as well. These variables will be used to understand the level of nutritional intake. For example, whether the household purchased green vegetables, consumed eggs and poultry, and energy-dense high sodium foods like potato chips or soft drinks. Household regional information will be used to merge household-level information with county-level data.Store closure and reopening data during COVID-19 (January-December 2020) will be collected from Google cellphone mobility data using Safegraph. We will merge the retailers' information with the Food Access Research Atlas (FARA) and Food Environment Atlas (FEA) by location for additional insights. Data on health and nutritional outcomes will be obtained from the Centers for Disease Control and Prevention (CDC). Census tract characteristics, racial distribution, and other sociodemographic information will be collected from the U.S. Census Bureau.We test the following three hypotheses from the objectives mentioned above:(1) H0: The composition of food outlets in a neighborhood is independent of the racial distribution, income, and urbanization level.(2) H0: Household dietary choices (therefore nutritional outcomes) are independent of the composition of food outlets in the neighborhood.(3) H0: The presence and abundance of less healthful food retailers (e.g., fast food restaurants) do not affect the diseases that may relate to dietary choices (e.g., diabetes).For hypothesis (1), we use logistic models, Naïve Bayes, eXtreme Gradient Boosting (XGB), and Artificial Neural Network to predict the location decision using socioeconomic and regional predictors, including the existing composition of food retailers. In particular, the use of spatial neural networks and naïve Bayes allows us to derive the conditional probabilities of site selection more precisely, given the racial and income characteristics of the region and the surrounding region.Greater foot traffic at a certain food retailer given other food retailers and the covariates indicates consumer preferences in a county.For hypothesis (2), we will measure vegetable, dairy, and protein intake from the IRI Homescan data for 2020. The effects of household and county characteristics and average foot traffic preferences will be controlled. Our key identification strategy is to use the exogenous shock of COVID-19. Due to the pandemic and its preventive measures,many stores were closed, which might have diverted consumers to other types of retailers. Lockdown policies and store closures changed the availability of food retailers, hence might have affectedfoot traffic.Forexample, closing a pizza place may increase foot traffic to a bigger grocery, and households may increase vegetable consumption. This logichelps us understand the effect of a food retailer's presence on household consumption choices. We will check the effect by income, race, and the level of urbanization.The third hypothesis (3) helps us understand the effects of nutritional (in)equityacross ethnic groups. For example, if an African-Americanhousehold is equally likely to choose nutritious foods given the location characteristics as opposed to the benchmark Caucasian household, but it could not because of low income, we find evidence in favor of the nutritional inequality literature. However, if the household is equally likely to choose nutritious foods given its income, but it did not, then the lack of access is attributed to dietary preferences and lack of demand in the region, and not because of systematic inequality. This logic will be applied in pairwise comparisons between Caucasian householdsand all other minority households. Either due to heterogeneous access or demand, food choices mayaffect household health outcomes,such as diabetes, obesity, and cardiovascular disease, which can be observed at the county level in CDC data.Many socio-economic and regional factors will be obtained from merging the data sets above, which creates a dimensionality problem. Our motivation for combining machine learning models with the traditional econometric model is to account for a large number of confounding control variables in the equation. The models are carefully chosen based on their predictive performance, flexibility, efficiency, and interpretability.

Progress 04/15/23 to 04/14/24

Outputs
Target Audience:Previously, the initial results of our project were showcased at a prominent annual meeting organized by the American Economics Association (AEA) in January 2023. This event draws a diverse audience including social scientists, students, and industry professionals from various fields. During the reporting period, we further advanced our research and presented updated findings at a significant conference dedicated to addressing economic inequities within agricultural, food, and environmental systems. This conference, held at the University of Minnesota, Minneapolis, is a collaborative effort involving major academic and professional organizations in agricultural economics. The conference was jointly hosted bythe American Journal of Agricultural Economics and the Agricultural and Applied Economics Association. Our presentation, titled "Substitutions or Complements? The Effects of Opening a Food Store on Customer Visits to Neighborhood Food Retailers," explored the impact of new food retailers on existing businesses and local food access. The findings sparked interest among attendees, comprising both academicians and industry stakeholders. The discourse fostered at this event offered valuable insights into the potential policy implications of our research, particularly in the context of enhancing nutrition security across the United States. Changes/Problems:We have encountered several challenges that impacted the progress of our project, particularly in data access and manpower. Firstly, we experienced a setback when one of our graduate students discontinued their program in the summer of 2023, which created a delay in our timeline. Additionally, the ownership of the cellphone location data we relied on shifted from SafeGraph to Dewey, leading to a temporary loss of access to this crucial dataset from the end of 2023 to the beginning of 2024. However, we successfully negotiated a new contract with Dewey and have resumed work on the second objective of our project. Furthermore, we have faced ongoing challenges in securing access to the IRI data set from USDA officials. While various factors have contributed to these difficulties, we remain hopeful for a resolution soon. Access to the IRI data is vital for providing comprehensive nutritional information and is integral to the completion of Objective 2 of our project, as it contains essential details about household food purchases. Currently, our analysis is limited to observing changes in foot traffic at food retailers, and without the IRI data, we cannot assess the impact of these retailers on the nutritional choices of consumers. This limitation underscores the importance of obtaining the IRI data to ensure a more thorough and effective analysis. We remain optimistic and are actively collaborating with the USDA and other stakeholders to secure the necessary data access. We appreciate their continued understanding and support in these matters. What opportunities for training and professional development has the project provided?The project involving researchers and graduate students from Texas Tech University and Washington State University has significantly contributed to their training and professional development through a variety of opportunities, enriching their educational and career trajectories. Key areas of development include: Big Data Handling: The students gained practical skills in managing and analyzing large datasets, including cellphone mobility data, American Community Survey data, and USDA data. This training is crucial for addressing complex issues related to food access and nutritional inequality, providing them with a robust foundation in data-intensive research. Machine Learning Methods: Exposure to advanced machine learning techniques and artificial intelligence models has equipped the students with the necessary tools to decipher complex relationships within the data. This expertise not only enhances their research capabilities but also boosts their professional value in the fields of data analysis and modeling. Application to Real-World Problems: The project offered hands-on experience in applying these sophisticated analytical skills to real-world problems, particularly those involving food access and nutritional disparities. Such experiences are invaluable, deepening the students' understanding of the practical applications of their academic work and preparing them for impactful careers in academia, industry, or policymaking. Collaboration and Networking: Participation in this project facilitated collaboration with researchers from various institutions, helping the students to build professional networks. These connections are crucial for fostering future research collaborations and opening up new career opportunities. Presentation and Communication Skills: The researchers also had the opportunity to present their research findings at the Conference on Economics of Inequity in Agricultural, Food, and Environmental Systems at the University of Minnesota, which was hosted by the American Journal of Agricultural Economics and the Agricultural and Applied Economics Association in fall 2023. This experience was instrumental in developing their presentation and communication skills, allowing them to effectively convey their research insights, engage with feedback from seasoned experts, and further refine their scholarly contributions. How have the results been disseminated to communities of interest?Earlier in January 2023, we initially showcased our project results at the American Economics Association (AEA) annual meeting, which is well-attended by a wide range of professionals including social scientists, students, and industry experts. We subsequently advanced our research and shared updated findings at a major conference focused on economic disparities in agricultural, food, and environmental systems, held at the University of Minnesota, Minneapolis. This event was a collaborative initiative supported by key academic and professional bodies in agricultural economics, including the American Journal of Agricultural Economics and the Agricultural and Applied Economics Association. At this conference, we presented our study titled "Substitutions or Complements? The Effects of Opening a Food Store on Customer Visits to Neighborhood Food Retailers." Our research delved into how new food retailers affect existing ones and local food access. The presentation generated significant interest from both academics and industry stakeholders, providing deep insights into the implications of our findings for policy, especially in terms of improving nutrition security across the United States. What do you plan to do during the next reporting period to accomplish the goals?For the upcoming phase of the project, we plan to submit the prepared manuscripts to academic journals. Work on Objective 2 is currently ongoing; this includes the extraction of store closure data during the COVID-19 pandemic. The other objectives have largely been completed. Once the manuscripts are finalized, we will present the results to a broader audience.

Impacts
What was accomplished under these goals? Objective 1 was accomplished. Our machine learning model distinguishes between low-income, low-access areas (previously termed food deserts)--rural regions with sparse populations and minimal ethnic diversity, and food swamps--densely populated urban areas with significant non-white populations lacking vehicle access. Key factors influencing access to healthful food include population density, ethnic demographics, property values, and income levels. Objective 2 was partly accomplished. We faced a data constraint as the ownership of the data changed in 2023. However, using store opening information, we were able to assess that the impact of new store openings on existing businesses varies significantly. For instance, while grocery store openings tend to disrupt customer traffic across multiple store types, coffee shops show minimal impact on neighboring businesses. Furthermore, the effects of new openings are not symmetric; the influence of one type of store on another can differ markedly in the reverse scenario. Another interesting finding is the varied responsiveness of different demographic groups. Black and low-income visitors are more likely to alter their shopping behaviors in response to new store openings compared to other groups, with Black visitors showing significant shifts in their shopping patterns, especially in response to new specialty stores. This suggests that store openings may offer preferred options to these groups, potentially enhancing accessibility to desired goods and services. Additionally, the presence of substitution effects within and sometimes between similar store categories highlights competitive dynamics, particularly among smaller store formats like drug and convenience stores. Yet, in some cases, new store openings can lead to business creation rather than just competition. Supercenters, for example, are found to contribute positively to the business activity of nearby incumbent stores, indicating a potential for new business generation alongside competitive pressures. Grocery stores stand out for their strong impact on various incumbent store categories, significantly affecting the business of restaurants, snack places, and specialty food stores, although less so for fast food outlets. This underscores groceries' competitive nature and their considerable influence on local retail ecosystems. Overall, these findings provide a nuanced understanding of how new store openings influence local economic dynamics, offering valuable insights for businesses, policymakers, and community planners. The third objective investigates the impact of local food environments on health outcomes, utilizing Double Machine Learning (DDML) to control for endogeneity between geographical characteristics and health. The methodology differentiates the effects of sensitive variables, such as the Modified Retail Food Environment Index (mRFEI), from nonsensitive predictors, isolating the genuine impact on health outcomes. Findings suggest that access to healthful food retailers generally correlates with better health outcomes, such as reduced obesity and high blood pressure rates. However, these effects are moderated by individual lifestyle choices such as smoking, physical activity, and alcohol consumption, indicating that unhealthy lifestyle habits can diminish the positive effects of healthier food environments.

Publications

  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Amin, M.D., Badruddoza, S., and McCluskey, J.J. 2023. "Substitutions or Complements? The Effects of Opening a Food Store on Customer Visits to Neighborhood Food Retailers." Conference on Economics of Inequity in Agricultural, Food, and Environmental Systems, University of Minnesota, Minneapolis, MN. Hosted by the American Journal of Agricultural Economics and the Agricultural and Applied Economics Association.


Progress 04/15/22 to 04/14/23

Outputs
Target Audience:During this reporting period, our primary target audience consisted of participants at the American Economic Association (AEA) annual meeting held in January 2023. We presented our preliminary findings in a poster titled "Presence of food retailers and nutrition security in the United States," which was co-authored by Co-PIs Modhurima Dey Amin, Syed Badruddoza, and Jill J. McCluskey. The AEA meeting attracts approximately 13,000 economists and social scientists from around the globe, showcasing the latest research and developments in the field of economics. As targeted by the AEA, our presentation reached a diverse and inclusive audience, encompassing individuals from various ethnicities, genders, income levels, sexual orientations, and demographic backgrounds. The conference attendees included undergraduate and doctoral students, faculty members, researchers, and industry stakeholders, both from economics and other related disciplines. By presenting our research at the AEA meeting, we ensured that our findings were shared with a wide range of professionals, fostering knowledge exchange and promoting interdisciplinary dialogue. In addition to academics and researchers, our target audience also included policy-makers, non-governmental organizations, and industry professionals involved in addressing food retailer presence and nutrition security issues. The AEA meeting provided a platform for these stakeholders to gain insights into our research, which could potentially inform their decision-making processes and contribute to improving nutrition security in the United States. To summarize, our efforts during this reporting period reached a diverse and interdisciplinary audience, including academics, researchers, students, and industry stakeholders, all of whom have a vested interest in understanding and addressing the challenges related to food retailer presence and nutrition security. Changes/Problems:Despite our persistent efforts to communicate with USDA officials, we have unfortunately been unable to obtain access to the IRI data set thus far. We understand that various factors may have contributed to this situation, and we remain hopeful that a resolution can be reached soon. The IRI data is vital for acquiring comprehensive nutritional information and completing Objective 2 of our project, as it contains crucial details about household food purchases. In the unfortunate event that we are unable to gain access to the IRI data, our analysis will be limited in scope. While we will still be able to report on changes in foot traffic at food retailers, we will regrettably be unable to assess the impact of these retailers on the nutritional choices made by consumers. This limitation underscores the importance of obtaining the IRI data to ensure a more complete and robust analysis of the project objectives. We remain optimistic and committed to collaborating with the USDA and other relevant stakeholders to secure the necessary data access, and we appreciate their understanding and support in this matter. What opportunities for training and professional development has the project provided?The project has provided several opportunities for training and professional development for the two graduate students involved at Texas Tech University and Washington State University. These opportunities include: Big Data Handling: The students have been trained in handling and processing large-scale datasets, such as cellphone mobility data, American Community Survey data, and USDA data, which are essential for the analysis of food access and nutritional inequality issues. Machine Learning Methods: The graduate students have been exposed to advanced machine learning techniques and artificial intelligence models, equipping them with the skills to analyze complex relationships and patterns in the data. This training will enhance their ability to conduct sophisticated research and contribute to their professional growth in the field of data analysis and modeling. Application of Methods to Real-World Problems: The project has given the students an opportunity to apply their newly acquired skills to real-world issues related to food access and nutritional inequality. This hands-on experience will enable them to develop a deeper understanding of the subject matter and its practical implications, thereby preparing them for careers in academia, industry, or policy-making. Collaboration and Networking: Working on this project has allowed the graduate students to collaborate with researchers from different institutions and develop professional networks. These connections will be valuable for future research collaborations and job opportunities. Presentation and Communication Skills: The presentation of a poster at the AEA meeting in January 2023 provided the graduate students with an opportunity to showcase their research findings, receive feedback from experts in the field, and develop their presentation and communication skills. Overall, the project has offered numerous opportunities for training and professional development to the graduate students, enhancing their skills in big data handling, machine learning, and real-world problem-solving. These experiences will contribute to their future success in the fields of research, academia, or policy-making. How have the results been disseminated to communities of interest?During this reporting period, the results have been disseminated to communities of interest through the following means: Conference Presentation: The preliminary findings of the project were presented at the American Economic Association (AEA) annual meeting held in January 2023. The conference attendees, which included economists, social scientists, students, and industry stakeholders from around the world, had the opportunity to learn about the research, engage in discussions, and provide valuable feedback. A poster titled "Presence of food retailers and nutrition security in the United States" was showcased at the AEA meeting, which allowed for visual representation of the research findings and facilitated further discussions among attendees interested in the topic. Networking and Collaboration: During the conference, the researchers had the opportunity to connect with experts in the field, discuss their work, and explore potential collaborations for future research. These connections help to disseminate the findings more broadly within the academic community and promote further research on the topic. University and Departmental Activities: The researchers have shared their findings with colleagues and students at their respective universities through seminars, workshops, and informal discussions. This dissemination helps to raise awareness about the research and its implications among the academic community. What do you plan to do during the next reporting period to accomplish the goals?During the next reporting period, we plan to take the following steps to accomplish our goals: Complete Objective 3: We will focus on analyzing the association between the presence of food retailers and health outcomes in an area. This will involve collecting and processing relevant data, conducting statistical analyses, and interpreting the results in the context of our research objectives. Prepare a Manuscript: After completing the analysis for Objective 3, we will work on preparing a manuscript that consolidates our findings from all objectives. The manuscript will be structured and written following the guidelines of top-tier agricultural economics journals. Journal Submission: Once the manuscript is ready, we will submit it to a top-tier agricultural economics journal for peer review and potential publication. This will help disseminate our findings among academicians, policymakers, and industry stakeholders. Presentations at Research Seminars: We plan to present our findings at high-profile research seminars hosted by different universities and institutions. These presentations will further disseminate our research findings and foster discussions among academicians and industry stakeholders. Networking and Collaboration: We will continue to engage with experts in the field and explore potential collaborations for future research related to food access, nutritional inequality, and health outcomes. Ongoing Training and Professional Development: We will continue to provide training and professional development opportunities for the graduate students involved in the project, ensuring that they gain valuable skills and experience in handling big data, machine learning methods, and their applications in our research area. Monitoring Progress: Throughout the next reporting period, we will closely monitor the progress of our research activities, making necessary adjustments and refining our strategies to ensure that we accomplish our goals and objectives. Communication and Dissemination: We will maintain our commitment to disseminating our research findings through various channels, including conference presentations, workshops, policy briefs, and reports, as well as leveraging online presence and social media platforms to reach a wider audience. By following these steps, we aim to make significant progress towards accomplishing our goals during the next reporting period.

Impacts
What was accomplished under these goals? Under the overarching goal of examining and understanding the impact of access to different food retailers and their regional distribution on nutrition equity and health outcomes in the United States, several accomplishments were achieved in line with the supporting objectives: Assessing the equity of access to healthful foods: Our research provided insights into the association between the demographic features of an area, such as income, race, and level of urbanization, and the location choices of food retailers. We found that 7 out of 10 retailers' location choices could be predicted by these features.Our preliminary results suggested that the presence of most store types is statistically greater in white-majority tracts than in black-majority tracts. Estimating the causal impacts of various food retailers on nutritional choices: We performed an event analysis using the Callaway and Sant'Anna (2021) model to derive the average treatment effects of the treated (ATT) on consumer visits to food stores when a new food store opens in a census tract. By analyzing cellphone mobility data for 2018-2019, we identified substitution and complementarity between food retailers within a census tract.Furthermore, opening a limited-service restaurant increased foot traffic to other store types in the neighborhood, but this effect was negative on the foot traffic to large groceries and convenience stores in black neighborhoods and positive for white neighborhoods. This indicates a substitution effect between fast-food restaurants and large grocery stores only in black-majority census tracts. Analyzing impacts of healthful food access on health outcomes: While our current study has not yet directly assessed the impact of healthful food access on health outcomes, we have laid the groundwork for future research by examining the distribution and consumer preferences for different food retailers across income and racial groups. This information will be valuable in generating customized food, health, and urban development policies that can address the disparities in access to healthful food options and their potential impacts on health outcomes. In conclusion, our study has made significant progress in understanding the distribution and consumer preferences of food retailers in the United States, which will inform further research on the impact of access to healthful food options on nutrition equity and health outcomes. By incorporating machine learning and artificial intelligence models, we aim to offer precise, local, and customized recommendations that will promote inclusive food and health policies.

Publications

  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Amin, M.D., Badruddoza, S., and McCluskey, J.J. 2023. Presence of food retailers and nutrition security in the United States. Selected Poster. American Economic Association Annual Meeting. New Orleans, LA.