Association for Public Policy Analysis & Management

The food environment is complex in terms of where and how households access food, the characteristics of households and stores, and many important factors to determine the household food purchase behaviors. An agent-based model, Food Accessibility Agent-based Model in Central Columbus, Ohio (FAAMC) was developed to evaluate the food availability of households and to test the impacts of interventions to increase food availability.

 To recreate real-world patterns of household food purchase behaviors, we first created a synthetic micropopulation dataset simulating all the households in Central Columbus, Ohio at the census tract level combining The Ohio State University’s Mapping Food Environment Survey (MFES) and the U.S. Census Bureau’s American Community Survey (ACS). Although MFES contains all the detailed information on each household’s food purchase patterns, household addresses were suppressed to protect the survey respondents’ confidentiality. To address this problem, we used an iterative proportional fitting process (IPF)-based spatial microsimulation to replicate the MFES respondents as many as the total population in the ACS at each census tract. The ratio of marginal totals of each common variable both in the MFES and the ACS iteratively acts as new expansion weight in the replication process. Then, we randomly allocated this synthetic population to obtain the latitude and longitude information using geographic information system (GIS). Finally, an agent-based model was built to simulate the household food purchase determined by household income level, household size, private vehicle ownership, the types of food stores, the preference of store types, the distance to food stores, and the proportion of the United States’ Department of Agriculture Thrift Food Plan (USDA TFP) items available at each store. Three categories of the policy interventions to increase food availability were evaluated in a single and/or combination manner(s): (1) reducing the preference to convenience stores/partial markets; (2) increasing the availability of the USDA TFP items; and (3) increasing income for households with lowest income levels as universal basic income. Community experts and stakeholders also participated in designing and validating the FAAMC through group model building.

 The FAAMC estimated that the mean food availability for food secure households is 17% higher than for food insecure households. When implemented independently, increasing income of poorest households can contribute to 10% higher food accessibility for food insecure households. The increase in the availability of the USDA TFP items in convenience stores/partial markets and in supermarkets can respectively achieve 3% and 5% higher food availability for food insecure households. Additional 3% food availability can be achieved when reducing the preference to convenience stores/partial markets. The combination of the increase in the availability of the USDA TFP items and the income increase for households with lowest income levels is the most impactful intervention, with which food insecure households can obtain 37% higher food availability.

 This study exemplifies the integration of both survey and census data regarding food security and the use of interdisciplinary data analytics methods. It also provides an example of how a systems approach can serve as an effective and efficient tool for food environment policy.