To understand how food stamps affect food spending, nonexperimental research typically requires some source of independent variation in food stamp benefits. Three promising sources are examined: (a) variation in household size, (b) variation in deductions from gross income, and (c) receipt of minimum or maximum food stamp benefits. Based on results of a linear regression model with nationally representative data, 90% of the total variation in food stamp benefits is explained by gross cash income, and household size variables alone. This finding raises concern about popular regression approaches to studying the Food Stamp Program.