Getting a healthy start: The effectiveness of targeted benefits for improving dietary choices

There is growing policy interest in encouraging better dietary choices. We study a nationally-implemented policy – the UK Healthy Start scheme – that introduced vouchers for fruit, vegetables and milk. We show that the policy has increased spending on fruit and vegetables and has been more effective than an equivalent-value cash benefit. We also show that the policy improved the nutrient composition of households' shopping baskets, with no offsetting changes in spending on other foodstuffs.

Take-up and use of Healthy Start Vouchers is high. An estimated 79-80% of all eligible households receive the vouchers, of which 90% are used (Department of Health, 2009). Many retailers accept Healthy Start Vouchers, including supermarkets, corner shops, milkmen, chemists, market stalls and greengrocers. Households can find information about which retailers accept vouchers on the Healthy Start website, or by calling their helpline.
All applications to the Healthy Start scheme have to be countersigned by a health visitor or midwife, who is also expected to provide information and advice on breastfeeding and healthy eating. Eligible households are sent four vouchers per 4-week period per child (eight vouchers for children aged between 0 and 1). Vouchers are only valid for this four week period, after which they can no longer be used.
To date, there have not been any large-scale evaluations of the Healthy Start scheme, though there are three small (qualitative) studies. One focuses on the views and experiences of parents, professionals and small retailers (Lucas et al., 2013), whilst the others examine whether it affected food consumption. They find an increase in reported fruit and vegetable consumption (Hills et al., 2006) and intakes of energy, calcium, folate, iron and vitamin C (Ford et al., 2009), but sample sizes are small (n = 58 and 336 respectively). Figure A1 illustrates the standard economic incentive effects of targeted benefits such as the Healthy Start scheme. For simplicity we consider the case without the Welfare Food scheme.
The line A-A' represents the initial budget constraint, where the household decides between spending on fresh fruit, vegetables and milk and spending on other goods. The introduction of the vouchers shifts the budget constraint outwards, but also introduces a kink, since the extra income can only be spent on the targeted goods, shown in line A-B-B'. The budget constraint for an equivalent value cash benefit without constraints on spending would be B''-B'.
Consider first "distorted households", represented by 1 , who would spend less than the value of the vouchers on fresh fruit and vegetables if they were given cash. Receipt of the vouchers leads these households to move out to 'the kink' (point B), increasing spending on the targeted good by more than if they were given a cash benefit (Southworth, 1945). For "infra-marginal households", represented by 2 , however, the effect of the vouchers is the same as cash benefits.
They use the voucher to cover existing spending and then re-allocate their (non-voucher) income among other goods. These consumers are also predicted to increase their spending on the targeted good, but the magnitude is in line with a standard income effect.

Milk spending
The pre-existence of the Welfare Food scheme complicated the analysis slightly. We focus on fruit and vegetables. We argue that households consume milk in relatively fixed quantities, depending on household size. Our analyses in Table A1 below support this, showing the estimates of a simple difference-in-difference analysis, comparing milk spending of eligible and ineligible households before and after the introduction of the scheme (we discuss the estimation in more detail in Section 4). Column 2 presents the estimates of a triple differences analysis, additionally distinguishing between distorted and infra-marginal households. Column 3 and 4 show the same for the quantity of milk purchased (in litres). These analyses show no evidence that milk spending changed with the introduction of Healthy Start Vouchers. Figure A2 below shows the distributions of monthly spending on milk and on fruit and vegetables, conditional on the number of adults, the number and ages of children, year, month and household fixed effects. This shows that the distribution of monthly milk expenditures is very concentrated, and much more concentrated than monthly spending on fruit and vegetables, with no large differences between eligible, ineligible, distorted, or infra-marginal households.
Analyses (not shown here, but available from the authors upon request) show that the standard deviation of milk spending is significantly smaller than the standard deviation of spending on fruit and vegetables.
We also confirm findings from the UK Committee on Medical Aspects of Food and Nutrition Policy (Department of Health, 2002), who state that the vast majority of households did not consume the amount of milk they could purchase with a token under the Welfare Food scheme.
Our findings suggest 85% of households in our sample belong to this group, with little difference between eligible, ineligible, distorted and infra-marginal households. Finally, examining the effect of the introduction of the Healthy Start scheme on milk spending shows no significant effects, with all estimates close to zero.

The reform of the scheme
Estimating the effect of the introduction of Healthy Start Vouchers is complicated by the preexisting Welfare Food scheme. We assume that consumers maximise their utility, which is a function of their spending on milk ( 1 ), fruit and vegetables ( 2 ) and other food ( 3 ): = ( 1 , 2 , 3 ). This is subject to a budget constraint, given by where denotes (other) income and denotes income from benefits.
We assume milk is purchased in fixed quantities, depending on household size, the number and age of children: 1 = ̅ 1 = 1 ( , , ). In other words, the quantity purchased does not depend on price or household income. This is consistent with the fact that the distribution of milk spending (conditional on household characteristics) is very concentrated, as shown above. If benefit income is paid in cash, households purchase a fixed quantity of milk and then allocate their remaining budget between fruit and vegetables and other food: * = ( 2 , 3 , − 1 ̅ 1 ), with n = 2, 3 and where the superscript * indicates the optimal spending.
Under the Welfare Food scheme, households receive welfare tokens of value that can only be spent on milk. For the majority of households, the report by the UK Committee on Medical Aspects of Food and Nutrition Policy indicates that that < ̅ 1 1 (Department of Health, 2002). In other words, households are "distorted", they but are assumed not to locate at the kink because the amount of milk that can be purchased with the tokens is greater than the maximum they want to purchase. Because of the distortion, spending on fruit and vegetables and other food for household receiving welfare tokens is given by ̂=̂( 2 , 3 , − ) ≤ * , with n = 2, 3.
In November 2006, welfare tokens are replaced with Healthy Start Vouchers, which are of roughly equivalent monetary value. However, the new vouchers can be spent on milk, fruit and vegetables: = ̅ 1 1 + 2 2 . The effect of introducing the vouchers is similar to the case with no welfare tokens (as shown in Figure A1), but the value of the extra benefit is lower: ′ = − ̅ 1 1 .
We can distinguish between two groups. First, those who are infra-marginal under Healthy Start Vouchers: ≤ ̅ 1 1 + 2 * 2 , and who therefore choose optimal spending ̅ 1 , 2 * , 3 * . Following introduction of Healthy Start Vouchers, there is no change in spending on milk, while spending on fruit and vegetables and other food increase in line with ′ (from ̂ to * ) due to an income effect.
This predicts that the increase in spending on fruit and vegetables is greater among distorted than among infra-marginal consumers (from ̂2 to ̌2), whilst the increase in spending on other foods is less among distorted than infra-marginal consumers (from ̂3 to ̌3).
Before the introduction of the reform, we observe ̅ 1 1 +̂2 2 . Using the cut-off will therefore cause some distorted consumers to be included in the infra-marginal group, leading to an overestimate of any change in fruit and vegetable spending for the infra-marginal group.

Figure A2: Conditional distributions of monthly expenditures on milk and on fruit and vegetables
Note: Expenditure is conditional on the number of adults, the number and ages of children in the household, year, month and household fixed effects.

Appendix B: Identifying households on benefits in the Kantar data
We define a household as being "on benefits" if both the head and main shopper are not in work, unemployed, in education, or work less than 8 hours a week. To assess how well this simple rule does in predicting which households are on benefits, we look at data from the Expenditure and Food Survey (EFS; a repeated cross-sectional study of households in the UK), which contains both hours worked and actual benefits receipt. Using this definition, we find that 17.8% of all households with a child aged 0-8, or where the woman is pregnant in the EFS are on benefits, compared to 20.7% in a similar sample in the Kantar data. In addition, we compare our definition of being on benefits (based on hours worked) in the EFS to the share of these households that are truly on benefits. We examine the robustness of our findings to an alternative definition of benefit receipt, where rather than using hours worked, we predict the probability of receiving benefits as a function of covariates in the EFS. Table B2 shows the marginal effects of a probit regression using the EFS for the period December 2004 to November 2008. This suggests that we predict benefit receipt well: a simple probit using the cross-sectional EFS provides a pseudo R 2 of 0.55. Figure B1 compares the distribution of the predicted probability of being on benefits in the Kantar data (predicted using the estimates from the EFS) with that from the EFS data. This shows that they line up well: the majority of individuals are predicted not to be on benefits, with the largest densities for probabilities of benefit receipt less than 0.4. The density increases again for probabilities above around 0.7. In our robustness analysis, we use Pr(benefits)>0.7 as the cut-point to define a household as being on benefits, though the results are robust to using different cut-points ranging from 0.6 to 0.9. Comparing this predicted benefit receipt (based on Pr(benefits)>0.7) to the actual benefit receipt in the EFS, we find that, among those we define as being on benefits, 92% receives benefits. Figure B1: The densities of Pr(on benefits) from the EFS and Kantar data      (1) and (3) show esimates of the coefficients from equation (3), columns (2) and (4) show estimates of the coefficients from equation (4) Table 3 for comparison. Column 2 drops households that do not record purchases of loose fruit and vegetables. Column 3 defines the value of the voucher as the value minus spending on milk. Column 4 uses the share of spending on fruit and vegetables (x100) as the dependent variable. Column 5 also includes households whose benefit status changes over time. Column 6 uses households with a predicted probability of receiving benefits (using the EFS) over 0.7. Column 7 uses these predicted probabilities as weights. * p<0.10, ** p<0.05, *** p<0.01.