Food and health responses to government policy

Abstract

This dissertation includes three chapters on food and health responses to government policy. The first chapter investigates how participating in the Supplemental Nutrition Assistance Program (SNAP) affects bulk purchasing behavior. SNAP issues monthly lump-sum payments. In theory, these payments help households overcome liquidity constraints and allow households to purchase bulk items. Purchasing products in bulk sizes decreases unit price and can increase consumption while holding expenditures constant. I show that SNAP households can save a considerable amount of money by purchasing bulk relative to non-bulk items. To estimate a causal relationship between SNAP participation and bulk purchases, I use within-household variation in the timing of program recertification as a source of exogenous variation in the decision to participate in SNAP. I find that adopting SNAP increases the expenditure share of bulk purchases for all groceries by five percentage points. This finding suggests that relaxing budget constraints enables households to exploit savings from shopping efforts that require lump-sum liquidity. Investigating Ready To Eat (RTE) cereal and yogurt purchases show that households shift their expenditures toward food categories that encourage bulk purchases.

The second chapter investigates the effect of SNAP benefit disbursement on intramonthly household-level purchases made from a supermarket retailer. Using variation in the timing of benefit receipt, we find that food spending in week one of the benefit cycle is 27 percent higher than in week four of the benefit cycle. However, we find little evidence for cyclicality in the healthfulness of food purchases. This paper also compares and contrasts estimates that use variation in the indicator for benefit receipt (benefit receipt estimates) to estimates that utilize variation in the probability of SNAP benefit receipt (likelihood of benefit receipt estimates). We find that the likelihood of benefit receipt estimates are statistically distinguishable from and 2.1 to 2.8 times larger than the benefit receipt estimates for the outcome of spending. For measures of healthfulness, we continue to find discrepancies between the two sets of estimates; however, the estimates are often statistically indistinguishable from each other. We decompose the difference between the two sets of estimates and find that all of the difference is due to endogenous measurement error, present only in the likelihood of benefit receipt estimate. We provide guidance to researchers in the event that only the likelihood of benefit receipt is known.

Finally, the third chapter bounds the true COVID-19 infection rate among students at Iowa State University. Knowing the true COVID-19 infection rate has been vital to understanding the virus’s scope and severity and properly navigating public health policies. Precise estimates of the true infection rate using testing data, however, have been difficult to obtain because of a measurement error problem caused by imperfect test accuracy, a missing data problem caused by individuals self-selecting into testing, and a repeated testing problem. This paper makes credible assumptions about the test accuracy and applies monotonicity assumptions about how individuals select into testing and the infection rate to bound the true cumulative COVID-19 infection rate among students at Iowa State University. Combining on-campus testing data from Iowa State University and testing data from Iowa State University student-athletes, who are subject to repeated testing, we illustrate how sensitive the bounds are to various assumptions used to address the measurement error and selection problem. Our preferred estimates imply that the true cumulative infection rate is in the interval [0.109, 0.444]. Accounting for repeated testing, the width of the bound increases, and the true cumulative infection rate is in the interval [0.096, 0.444].