Economic Policy Simulation: Incentivizing Healthy Eating Through Changes in Supplemental Nutrition Assistance Program Policies
Research Center: Institute for Research on Poverty, University of Wisconsin-Madison
Investigator: Basu, Sanjay, Jay Bhattacharya, Christopher Gardner, and Hilary Seligman
Institution: Stanford University Medical School
1265 Welch Road
Stanford, CA 94305
Numerous editorial commentaries have debated whether the U.S. Department of Agriculture should restrict use of Supplemental Nutrition Assistance Program (SNAP) benefits for purchasing sugar-sweetened beverages (SSBs) and/or further incentivize the purchase of fruits and vegetables (F&Vs) as examined in the Healthy Incentives Pilot (HIP) study. The purpose of this evaluation was to use mathematical modeling methods to address three outstanding questions about these proposals:
- Given current physiological understandings of how changes in consumption of different foods relate to changes in body weight and the risk of type 2 diabetes, how much would consumption of SSBs and/or F&Vs need to change under such policies to generate significant changes in obesity and diabetes risk among the SNAP recipient population?
- Are such changes consistent with model-based effect size estimates of how much consumption is expected to change from a restriction on SSBs or a 30-cents-per-$1 subsidy of F&Vs?
- If such proposals are evaluated in randomized trials, what sample sizes would be required to detect anticipated effect sizes?
To estimate the changes in calorie intake and glycemic load required to significantly alter obesity and type 2 diabetes risk among SNAP recipients, two sets of models of metabolism and disease risk were used:
Values for age, sex, height, weight, and race/ethnicity among SNAP recipients were taken from the National Health and Nutrition Examination Survey (NHANES 1999–2010, N = 19,388 self-identified SNAP recipients).
- The National Institutes of Health (NIH) body weight simulator, which incorporates metabolic parameters to identify how much a person defined by age, sex, height, and weight would need to alter caloric intake to significantly reduce body weight; and
- The Stanford Project for Open Knowledge in Epidemiology (SPOKE) diabetes risk simulation, which estimates how changes in both calorie intake and glycemic load affect type 2 diabetes risk by age, sex, body mass index, and race/ethnicity.
To estimate potential alterations in calorie intake and glycemic load among SNAP participants following an SSB restriction, a modified Delphi method was employed, in which surveyed experts suggested three approaches to simulating SSB restriction effects:
- Calculating the marginal propensity to consume SSBs out of SNAP benefits based on SSB expenditure estimated from reported NHANES servings per day multiplied by item prices per serving, SNAP benefit levels self-reported in NHANES, and household income self-reported in NHANES;
- Simulating the reduction in SSB intake if SNAP recipients are subject to the sales tax associated with paying for SSBs out of pocket rather than being exempt from the tax by paying for SSBs out of SNAP benefits; and
- Simulating no change in SSB intake among infra-marginal SNAP participants and reductions in SSB intake among the remaining participants to levels of matched nonparticipants.
To find matched nonparticipants in the latter scenario, the novel method of near-far matching was employed, which uses the instrumental variables of State-level variations in SNAP (State differences in broad-based categorical eligibility, simplified reporting, and biometric information requirements) to find individuals who differ significantly in their instrumental variables, but are matched on observed characteristics (thereby accounting for both observed and unobservable individual-level covariates that differ between SNAP recipients and nonrecipients). To estimate substitution levels from SSBs to other foods, a quadratic almost-ideal demand system was used to estimate own- and cross-price elasticities among food groups (the system was estimated by linking the county-level geocoded NHANES to the Quarterly Food-at-Home Price Database).
This linkage also permitted the estimation of how much SNAP recipients would increase F&V intake given a subsidy (own-price elasticity of F&Vs), substitute other products for F&V (cross-elasticity), and utilize the additional benefits from the subsidy ($1 extra for each 30 cents of F&V purchased) across all foods. The estimated changes in consumption among food groups were converted to calorie intake estimates and glycemic load changes per capita per day for input into the metabolic models. Ten thousand repeated replications of the models were conducted while repeatedly sampling from the input data on consumption, price, elasticity, and metabolic change, to identify confidence intervals around modeled results.
Changes in SSB consumption that would be necessary to significantly reduce obesity and type 2 diabetes among SNAP recipients in the mathematical models were considerably smaller than anticipated. When incorporating the uncertainty ranges in metabolic parameters, and substitution of other beverages for SSBs, SNAP recipients would need to reduce SSB consumption by at least 16.7 kilocalories (kcals)/person/day (95 percent confidence interval (CI): 14.6-18.8) for a significant improvement at the p<0.05 level in terms of obesity prevalence and type 2 diabetes incidence; compensatory consumption of other beverages in place of SSBs would be anticipated to reduce the net kcal reduction to 7.9 kcals/person/day (95 percent CI: 6.1-8.0). The simulated effects of an SSB restriction averaged 24.2 kcals/person/day in less SSB consumption (95 percent CI: 21.2-27.3), which reduced to 11.4 kcals/person/day (95 percent CI: 8.4-14.5) after compensatory consumption of other beverages. Of note, the near-far matching method did not find a significant difference between SNAP recipients and matched nonparticipants when using instrumental variables to account for unobserved differences between recipients and nonrecipients (SNAP recipients who drank SSBs had an average of 276 kcal/person/day of SSB intake, versus 246 among matched nonparticipants; the averages including zero consumers were 157 and 140 kcal/person/day, respectively).
The changes in consumption under an SSB restriction would be expected to produce a significant but small effect, of -0.9 percent in population obesity prevalence (95 percent CI: -0.4 to -1.4 percent) and -8.5 per 100,000 in diabetes incidence (-2.4 to -14.6) among SNAP recipients. The diabetes outcome was driven more by changes in glycemic load than by changes in calorie intake. For a randomized control trial of the SSB restriction to detect the observed consumption changes, sample sizes would need to be N>1,890 in the intervention arm to achieve greater than 80 percent power.
Changes in F&V intake from a 30-cents subsidy were estimated to average 0.24 cup-equivalents per day (95 percent CI: 0.20-0.28). The increase in caloric intake and reductions in glycemic load were not statistically significant. Hence, there were not significant changes in associated obesity or type 2 diabetes in the modeled subsidy scenario, which would require more than twice the observed effect size. The sample size estimates for detecting the estimated effect sizes in a randomized trial would be N>1,650 in the intervention arm to achieve greater than 80 percent power, which is notably smaller than the HIP study’s population sample in Massachusetts (N=2,081 dietary recalls from 1,871 respondents).