Agent-based modeling insights into the optimal distribution of the Fresh Fruit and Vegetable Program

Abstract

The Fresh Fruit and Vegetable Program (FFVP) program provides a free fresh fruit or vegetable snack to children in low-income schools between two and five times a week. This is an important nationwide program, but current funding levels are insufficient to reach all eligible schools. Thus, there is a need to develop insight on contextual factors that may impact the effectiveness of FFVP in facilitating the development of childhood food preferences. An agent-based model of preference formation is used to understand how exposure to FFVP in early elementary school may affect preferences for healthy food by 6th grade. Preferences are modeled using the temporal difference learning algorithm used by Hammond et al. (2012). This model simulates habit formation in the context of food by modeling preference formation in terms of positive or negative feedback about the food most recently consumed. The model is designed to provide insights into how different patterns of exposure to FFVP affect preferences, and how living in a food desert changes the effect of FFVP. We use data on 35,981 students from Arkansas elementary schools (from 2008 to 2016) and information about the commercial food environment in Arkansas communities to populate the model. We find that early FFVP exposure is more beneficial than late exposure conditional on the number of years exposed. We also find that FFVP is more beneficial for children who grow up in environments lacking healthy food.