The distributional consequences of a fiscal food policy: evidence from the UK

1. Introduction

One of the most pressing public health challenges is the prevalence of diet-related chronic disease resulting from poor dietary choices. For example, according to the 2004 Health Survey for England, which records body mass index (BMI), nearly 65 per cent of the adult population were overweight (BMI >25) while almost 25 per cent were obese (BMI >30). A great deal of interest and attention is focused on the use of fiscal interventions to correct poor dietary choices, in both Europe and the USA. Among the interventions that are proposed are taxes on foods which are deemed unhealthy, for example, those which are high in saturated fat or sugar and subsidies on foods that are deemed to be healthy, for example, fruit and vegetables. Much of the work to date focuses on overall effectiveness in reducing unhealthy food consumption and obesity rates. While many of the studies conducted by economists tend to show only a slight decrease in fat consumption resulting from a fat tax (Chouinard et al., 2005; Powell and Chaloupka, 2009), studies by health professionals show that fat taxes have meaningful impacts through reduced rates of cardiovascular disease (Marshall, 2000; Mytton et al., 2007). For example, Mytton et al. (2007) show that up to 3,200 deaths from cardiovascular disease could be avoided in the UK through a fat tax on a wide range of foods, and Marshall (2000) demonstrates that a tax on dietary saturated fat could avert up to 1,000 deaths a year in the UK.

In this paper, we consider the welfare implications of a fiscal food policy. While health and, therefore, welfare are likely to improve as a result of the policy, this will be countered by the fiscal burden. Although there is a great deal of attention on the use of fiscal interventions as a method of regulating diets and improving health, the welfare cost of such a policy remains a critical concern. From a classical welfare economics perspective, any policy which is based on food taxation will be regressive. Thus, through their effect on the fiscal burden, the impacts of the policy on welfare will be influenced by the magnitude of the tax and also, in a society which cares about distributional questions, who pays it. Addressing these issues requires a comprehensive social welfare framework which accounts for the impacts of changes in individual health and wealth interpersonally as well as individualistically. By presenting a formal analysis of the fiscal burden of tax- and subsidy-based food policy, this paper aims to extend the debate around the efficacy of such a policy to include consideration of the costs incurred. Distributional consequences are assessed using the distributional characteristic derived by Feldstein (1972) while the welfare impacts are measured using the method presented in Stern (1987) and Newbery (1995).¹

The paper is organised as follows. A review of the literature is in Section 2, with an emphasis on the welfare effects of fiscal food policies. Section 3 describes the theoretical and empirical methodology. The data are described in Section 4, as well as the empirical results. The final section concludes.

2. Fiscal food policies and social welfare

The existing literature measuring the welfare incidence of a fat tax tends to use money-metric measures based on compensating and equivalent variation. Leicester and Windmeijer (2004) simulate the effects of a UK tax based on four nutrients found to have adverse health consequences when consumed in large quantities (saturated fat, mono-unsaturated fat, sodium and cholesterol). The tax is levied at the rate of one pence per kilogram of the nutrient and, based on the simulated tax rate using data from the 2000 National Food Survey, is found to be highly regressive. The poorest 2 per cent of the 8,000 households in the sample spend 0.7 per cent of their income on the tax, while the richest pay only 0.1 per cent (the median income group pay ∼0.25 per cent). Chouinard et al. (2007) examine the effects of a fat tax on dairy products using data on weekly average household purchases for a sample of 23 US cities. They simulate the effects of a tax based on the fat percentage of different dairy products. Based on a 10 per cent tax rate, they calculate the equivalent variation as a measure of the welfare loss resulting from the tax. Their results find the tax to be highly regressive, with the burden falling mostly on low-income households: the burden is 0.24 per cent for households with an annual income of USD 20,000, but only 0.024 per cent for households with an annual income of USD 100,000. Allais, Bertail, and Nichele (2010) conduct an analysis similar to Chouinard et al. (2007) except they extend their model to include demand elasticities for a full range of food groups for a sample of French households using the TNS Worldpanel survey. Simulating a 10 per cent tax on foods in the cheese and butter category they find the tax regressive as well: the burden is 0.22 per cent for modest income households but only 0.068 per cent for well-off households. Okrent and Alston (2012) estimate an equilibrium displacement model to estimate the welfare effects of a retail food policy that accounts for the possible health benefits. Although a number of scenarios and fiscal policies are examined, based on a scenario of a USD 0.005 tax per gram of fat, they calculate the welfare change using a compensating variation measure and find that annual change in social welfare, excluding changes in public health-care costs, is a reduction of USD −1.937 billion to social welfare (Okrent and Alston, 2012). Once the annual change in public healthcare costs are included, however, the social welfare was raised by USD 1.675 billion.

The studies cited above do not have the analysis of welfare changes as their primary motivation and there are a number of reasons to be dissatisfied with them in this aspect. Informal welfare measures of social welfare, such as the money-metric equivalent and compensating variations used in these studies, are far from ideal. Money metric utility measures assume that the social marginal utility of income is equal to one in every household (Banks, Blundell, Lewbel, 1996). This is a limiting assumption since it does not allow the analysis to distinguish between policies that have differing implications for the distribution of the fiscal burden across income groups. Moreover, Blackorby and Donaldson (1988) show that money metric measures violate concavity of social orderings over optimal commodity allocations unless household preferences are homothetic.² Thus, while data requirements tend to be modest for money-metric measures, they only represent valid measures of welfare under very strict assumptions which are not normally descriptive of actual behaviour (Blackorby and Donaldson, 1988). Considerations of distributional equity can only be assessed under a rigorous social welfare model, which allows the social marginal utility of income to differ between households and permits social judgements on equality. This is accomplished by specifying a social welfare function with heterogeneous marginal utilities.

This paper examines the impact of fat taxes and thin subsides within a formal and rigorous welfare economics framework. The analysis provides two distinct economic measures which contribute to an analysis of welfare change. Firstly, the distributional characteristic is computed providing information on the degree of concentration of different food items in low-income households and makes clear the extent to which taxes and subsidies will impact the poor. Secondly, using first- and second-order welfare differentials the overall impact of fat taxes and thin subsidies on social welfare is assessed. The next section discusses the interpretation of the distributional characteristic and welfare approximations in the context of fiscal food policies.

3. The welfare impact of price changes

3.1. The distributional characteristic

The distributional characteristic, introduced by Feldstein (1972), is commonly used to examine distributional consequences from indirect taxes and price changes (see, for example, Newbery, 1995; Ray, 1999 and Liberati, 2001).⁵ It provides information on the distributional impacts of indirect price changes, like those caused by a fiscal food policy, by identifying the goods for which price changes will have the greatest impact on the less well-off.

3.2. Welfare approximation

3.3. Evaluating the social weights

The inequality aversion parameter determines the extent to which a marginal increase in expenditure to the poor is worth more than to the rich. Thus, values of $ρ$ that are successively greater than zero lead to greater weight being applied to the welfare of poorer households. As a result, the relative weight given to consumption of low-income households in the distributional characteristic (i.e. $βh$⁠) is also increased. At one extreme, when $ρ=0$ then $W=∑Vh$ is the Benthamite pure utilitarian case of no inequality aversion and $βh$ collapses to unity meaning all households are applied the same social weight so distributional issues are of no concern. At the other extreme, when $ρ→∞$ then $W=min{V1,…,VH}$ is the Rawlsian maximin case in which only the utility of the poorest household matters. A range of values for $ρ$ are used in our empirical analysis to assess whether conclusions are robust to different distributional judgements.

4. Data and food groups

Data on food expenditures and quantities are from the UK government's Expenditure and Food Survey (EFS) for 2003–2004. The 2003–2004 dataset is based on 7,014 households in 672 postcode sectors stratified by Government Office Region in England and Wales. Participating households voluntarily record food purchases for consumption at home for a 2-week period using a food diary for each individual over 7 years of age. The data collected by the food diaries are supplemented with the use of till receipts. The EFS identifies four major categories of interest: food (subdivided into 55 categories), non-alcoholic drink (subdivided into seven groups), alcoholic drink (subdivided in four groups) and catering services (split into three categories).

Individual food items are converted into aggregate food groups. The aggregation is necessary in order to estimate the demand system that is required for the computation of the second-order welfare effects. A two-level hierarchy is adopted for the purpose of demand system estimation. In the top level, foods are aggregated into: dairy and eggs; meat and fish; staples and starches; fruit and vegetables; fats and sugars; beverages and hot take-outs. At the second level, these broad categories are disaggregated into the groups listed in Tables 1, 2 and 3. The rationale adopted in the second-level aggregation is to identify separable groups which have similar fat contents in order that a similar fat tax would be applied to all goods within the group.⁹ This entailed a compromise and in some cases a range of fat contents exists within the group. The clearest example of this is in the milk and creams group which, for example, includes cream, full-fat and semi-skimmed milk. In the interests of brevity we do not give a detailed explanation of the composition of most groups which are, on the whole self-explanatory. Some detail is in order for the following groups, however. The ‘other meat’ category is largely made up of ready-made meat products such as meat pies, sausage rolls and meat-based ready-meals. The ‘potatoes’ category includes fresh, processed and canned potatoes. The ‘one-a-day only’ fruit and vegetables category comprises those fruits and vegetables, such as baked beans and fruit juices, for which multiple portions do not count towards the recommended five-a-day.¹⁰ Lastly, the ‘other fruit and vegetable’ category is largely ready-meals and take-out meals and ‘all fats’ include margarines and oils.

The tax applied to selected food groups is based on saturated fat content, while the subsidy is applied to fruit and vegetables. The choice of saturated fat as the prime target of the tax is justified by evidence from the medical literature: Saturated fats are an important risk factor in the occurrence of coronary heart disease (Hu et al., 1997), higher systolic blood pressure (Esrey, Joseph, Grover, 1996) and higher plasma concentration of cholesterol (Ascherio et al., 1994). Fruit and vegetables, on the other hand, are positively linked to lower risks of various cancers (Ames, Gold, Willett, 1995; Riboli and Norat, 2003), major chronic diseases (Hung et al., 2004) and ischaemic stroke (Joshipura et al., 2001). Specifically, the fiscal scheme simulated increases in the price of each food group by 1 per cent for every per cent total energy comprised of saturated fat within the group. The EFS dataset contains nutrient conversion tables that are used to convert food group items into nutrient content. For example, since milk contains 1.72 per cent of saturated fats, its price increases by 1.72 per cent. A ceiling of 15 per cent is placed on the simulated price increase. To offset this tax burden, and to encourage the consumption of fruit and vegetables, a subsidy on fruit and vegetables is introduced. The subsidy is calculated so that the overall fiscal policy is revenue neutral, so the revenue generated from the tax completely funds the subsidy. This works out to be ∼27 per cent on the selected fruit and vegetable categories. Table 1 presents the tax and subsidy rates applied to the different component food group items.

5. Estimating food demand elasticities

The calculation of the second-order measure of welfare change requires estimates of the own-price elasticities of demand. We obtain these by estimating a linear approximate AIDS model using the same dataset that is used to calculate the welfare changes themselves. The model is estimated at the household level and we use the infrequency of the purchase model (Cragg, 1971; Tiffin and Arnoult, 2010) to address the censoring which occurs because the data are collected in a 2-week window. In addition to the standard variables in the AIDS, we include a set of socio-demographic variables to account for the effects of differing household composition on demand. Table 2 presents the own-price (uncompensated) elasticities for each of the component food groups.¹¹

6. Results

We present results for two alternative settings. In the first, we focus solely on the part of the household budget that is spent on food. Thus, the distributional characteristics are computed using only the food groups in Table 1 and the expenditure weights are computed using expenditure on food for the variable $E$ in equation (14). In the second, we analyse all expenditure by including an additional group to represent consumption of non-food items measured as non-food expenditure. In this case, we use total expenditure to compute the welfare weights. The distributional characteristic, discussed in Section 3.1, is a scale-free measure with higher values indicating that consumption of a particular good is concentrated in households for which the marginal social benefit of increased income is highest. It measures the socially weighted level of consumption relative to a situation in which the social weights are equal across households (see equation (2)). The social weights are not normalised across the two settings in which the distributional characteristic is calculated and therefore the distributional characteristics for each setting are not comparable with each other. Comparison between the food expenditure and total expenditure weighted results is therefore conducted using the position of a particular food in the ranking of foods by distributional characteristics.

The larger the distributional characteristic of a good, the more disproportionate is the burden on poorer households. Likewise, a subsidy on a good with a smaller distributional characteristic will be disproportionately paid to better off households. Table 3 presents the distributional characteristics for the food groups under examination. We report characteristics for a single value of the inequality aversion parameter because the pattern is similar across values. The ranking of the food groups is broadly similar across the two alternative calculations. There are some notable exceptions, however, of which the most marked is lamb which is $28th$ in the food expenditure weighted results and $13th$ in the results where the welfare weights are based on all expenditure. This implies that lamb purchases are more highly concentrated in households that have low incomes than in households that have low food expenditure. The four food groups with the highest ranking (i.e. the highest distributional characteristics), reflecting expenditure being concentrated among households with high marginal utilities of expenditure are milk, bread, eggs, chips and crisps, and rice and pasta. The four food groups with the lowest ranking, reflecting expenditure being concentrated in the higher-income households are alcohol, lamb, water and fish. Table 3 also shows that in many instances the distributional characteristics are very close in value for some goods. For example, tea and coffee, fats, breakfast cereal and confectionery have distributional characteristics that are very close.

An examination of the distributional characteristics of the five food groups shows which will be subject to the highest tax rates: cheese; all fats; cookies, cakes and pastry; beef and lamb show that they cover a wide range. Consumption of all fats and cookies is relatively concentrated among the socially deserving while that of the red meats and cheese is concentrated among the socially less deserving. The distributional characteristic of fresh fruit and vegetables shows that consumption is concentrated among the socially less deserving. Considering the results obtained with the welfare weights based on total expenditure, it can be seen that the distributional characteristics of the food groups are all substantially larger than that for non-food expenditure. This supports the widely recognised view that taxation of food will be regressive.

Despite this, the narrow range covered by the distributional characteristics for the food groups (in comparison with those calculated for all goods in Newbery (1995) and Madden (1995a)) suggests that a policy that taxes food groups according to their saturated fat content at this level of aggregation would not be much more regressive than a tax on food as a whole. It is worth emphasising, however, that the distributional characteristic of fresh fruit and vegetables is towards the bottom end of the range, showing that consumption is concentrated among the socially less deserving. Welfare economics would therefore suggest that goods in this category should be taxed for redistributive purposes. The subsidy which is proposed here as a means of improving diets would increase the regressivity of the policy in comparison with a situation where only a tax on high-fat products is imposed. These results show that a tension exists between redistributive and public health policy objectives. While policy guided by public health concerns may desire higher prices in food groups such as fats and lower prices in fresh fruits and vegetables, a policy guided by economic welfare considerations would suggest a subsidy on fats and a tax on fresh fruits and vegetables. It should be noted, however, that as stated earlier drawing conclusions alone from the distributional characteristics should be nuanced since these measures do not account for consumer price response nor do they provide an assessment of the size of the fiscal burden resulting from a fiscal policy. Later the discussion turns to the welfare measures derived above, specifically the second-order welfare approximation to address these issues.

It has been noted that some of the groups which are analysed in Table 3 include a range of saturated fat contents. In Tables 4 and 5, we present the distributional characteristics for selected goods within a highly disaggregated set of 244 goods. The same caveat as above, concerning non-comparability between the situations in which the welfare weights are based on food and total expenditure, respectively, is applicable to these tables. The distributional characteristics for a given setting, be it food expenditure- or total expenditure-based weights can be compared between Tables 3, 4 and 5, however. Tables 4 and 5 show that by focusing attention at a more disaggregated level the differences in diet between rich and poor households are amplified.¹² For example, referring to Table 4, it can be seen that whole milk is among the goods which are most concentrated in socially deserving households, while the lower-fat-skimmed and semi-skimmed alternatives are much less so. From an economic perspective, this effect is important because it shows that a tax which is designed to induce substitution of skimmed and semi-skimmed milk for whole milk will be more regressive than a uniform tax on all milk. Table 5 reveals a similar situation applies to the all fats category. Overall, foods in this group are among the most concentrated in the socially deserving households but those which are lower in saturated fats are less so. A fiscal policy that is designed to induce substitution towards the more healthy products will again be regressive. Comparing the results obtained with food expenditure-based welfare weights with those for total expenditure-based weights, it can be seen that substantial differences in ranking exist for infant milk and for lard, suet and dripping. Both dried and ready-to-drink infant milk is concentrated in households which have relatively low levels of food expenditure and high levels of total expenditure. In contrast, lard, suet and dripping tend to be consumed in households with relatively high food expenditure and low levels of total expenditure.

Analysis of the distributional characteristics alone gives an indication of where consumption of particular goods is concentrated and as a result, how the fiscal burden will be distributed. It does not give any measure of the size of the fiscal burden nor of how the size of the burden will be influenced by alternative assumptions regarding society's preferences concerning distributional equity. Tables 6 and 7 therefore report the results of computing the first-order approximation in equation (7) and the second-order approximation in equation (8) for the fiscal policy described in Table 1 over different values of inequality aversion. Table 6 reports the results where total welfare is derived only from food and the welfare weights are based on food expenditure. Table 7 reports the results where total welfare is derived from all consumption and the welfare weights are based on total expenditure. In the aggregate analysis, the tax rates were calculated at the level of the aggregate commodity (24 tax rates) and in the disaggregate analysis they are at the level of individual food items (244 tax rates). We report only the first-order welfare changes for the disaggregated goods because the elasticities necessary to estimate the second-order impacts are not available at this level of aggregation. The first-order effects can be taken to be a lower bound on the welfare change. The welfare changes that we report are a measure of the proportionate change in expenditure that could be applied to everyone which would be equivalent in social welfare terms to the effect of the price changes resulting from the implementation of the fiscal food policy (Newbery, 1995). Thus, the results show that with an inequality aversion coefficient of 1 and considering only the first-order effect, the tax only policy is equivalent to the welfare loss that results from a 1.88 per cent reduction in food expenditure and 1.40 per cent reduction in total expenditure.

The results for the tax only policy show that such a policy is equivalent to the reduction in welfare that would result from between a 1 and 2 per cent reduction in consumer expenditure. The largest reduction in welfare arises when we consider food expenditure only and the policy is applied at a disaggregate level, for example, it distinguishes between skimmed and full-fat milk. The impact of the tax increases as the level of inequality aversion increases, a reflection of the regressivity of the tax. The difference is small, however, when the welfare weights are based on food alone and the focus is on aggregate goods. This is a reflection of the relative similarity of distributional characteristics for goods at an aggregate level, a consequence of limited difference in the tendency for taxed and untaxed goods to be concentrated in socially deserving households. The difference in welfare impact between levels of inequality aversion is larger when we examine disaggregated goods and expenditure across all goods. This reflects the greater degree to which goods that are taxed are concentrated in socially deserving households when considering disaggregated goods and the larger differences in total expenditure and hence welfare weights that exist for total expenditure in comparison with food expenditure. In all cases the first-order effects exceed the second-order effects, the difference reflecting the extent to which consumers can ameliorate the impacts of the tax through substitution.

When the tax is combined with a subsidy it can be seen that the subsidy more than compensates for the impact of a tax in all but one of the cases examined. The combined effect ranges from a 0.05 per cent reduction to a 1.41 per cent equivalent increase in expenditure. When considering the effects only at the level of food expenditure the difference in the welfare change that exists between the lowest and highest levels of inequality aversion is increased as a result of combining the tax with a subsidy. Thus, the combined tax and subsidy is more regressive than the tax alone. This is a consequence of the relatively low distributional characteristics associated with fruit and vegetables, indicating that these goods are relatively highly concentrated in the socially undeserving households. This effect is reversed when looking at the case of all expenditure where combining the tax with a subsidy leads to a smaller difference between the welfare changes associated with the smallest and largest values of inequality aversion. Thus, the policy remains more regressive but is less so than a tax alone. This difference can be explained by the fact that tinned and processed fruit and vegetables are much more concentrated in households that are judged to be socially deserving when this judgement is based on levels of total expenditure than when it is based on levels of food expenditure.

In terms of the existing literature, these results reinforce the finding that a fiscal food policy based on a tax on saturated fats and a subsidy of fruits and vegetables is socially regressive. Similar conclusions are reached in Leicester and Windmeijer (2004), Chouinard et al. (2007) and Allais et al. (2010). It should be noted that while Okrent and Alston (2012) find a negative change in welfare from a fat tax, the change is actually positive once changes in public health expenditure are accounted for in the framework. A key difference between the prior studies and this one, however, is the finding that not only are the poorest households more negatively effected by the policy, but that even a subsidy is regressive because of the distribution of consumption of fruits and vegetables. As other studies (Nordström and Thunström, 2011) have clearly demonstrated, a tax may be effective in reducing average fat intakes and, when combined with a subsidy, increasing average fruit and vegetable intakes. The core objective in changing health patterns, however, may not be met as the marginal change in diet that results from price changes is insufficient to move those who have extremely bad diets to the extent that would be required to remove the substantial risk that eating such a diet brings to one's health (Tiffin and Arnoult, 2011). Furthermore, a tax on food will also have a proportionately larger impact on the poorer segments of society which will worsen as the tax becomes increasingly targeted on the specific elements of the diet which are unhealthy.

7. Conclusion

Diet-related disease arising from poor dietary choices is of substantial concern to societies throughout the developed world. For example, being obese increases the risks of a range of chronic health problems including heart disease, type 2 diabetes and high blood pressure. Additionally, it has been shown that increased levels of fruit and vegetable consumption will contribute to a reduction in the incidence of some cancers. As a result, there is an increase in interest in public health policies that are designed to reduce the impacts of diet-related disease. One such policy is a fiscal intervention designed to reduce the consumption of calorie and fat dense food and to encourage the consumption of fruit and vegetables. The literature analysing the impacts of this type of policy has primarily been concerned with measuring the benefits in terms of the numbers of lives saved. The degree to which the costs, as represented by the fiscal burden, have been analysed is limited. Moreover, recognising that such policies are criticised on the grounds that taxing food is likely to be regressive, it is important that the analysis of the fiscal burden addresses the question of distributional equity. It is important that population-based measures, such as a fiscal food policy, reflect potential inequalities that may result from their undifferentiated approach. There have been some attempts at measuring the extent of the regressivity, but none have adopted a formal welfare economics-based analysis that allows for distributional concerns. A significant failing of the existing studies is therefore that they fail to fully reflect differences in the interpersonal impact of alternative policies and assumptions regarding societal preferences for distributional equity. The results in this paper emphasise this point, as findings suggest the regressivity of a fiscal food policy increases as the tax becomes focused on products with high saturated fat contents and is, in fact, more regressive when coupled with a subsidy on fruit and vegetables because consumption of these foods tends to be concentrated in households at the upper end of the distribution.

This paper focuses on an evaluation of the costs associated with a fiscal intervention designed to improve diets. It uses a framework drawn from the literature of marginal tax reform which explicitly incorporates a social welfare function to allow inter-household differences in marginal social utility. Distributional characteristics identify the extent to which consumption is concentrated in households at different points in the spectrum of social marginal utilities. A progressive redistributive policy would have a negative correlation between the tax rate and the distributional characteristic showing that goods which tend to be consumed in the most socially deserving households (those with the highest marginal social utility) are subject to the lowest tax rates. We find that the distributional characteristic does not differ much between aggregate food groups and there is no evidence of a negative relationship between saturated fat content (and thus tax rate) and the distributional characteristic. When looking at more disaggregated goods some important differences arise. For example, consumption of full-fat milk is found to be much more highly concentrated in socially deserving households than reduced fat alternatives. Likewise, margarine consumption is concentrated in socially deserving households while reduced fat alternatives tend to be consumed in the less socially deserving. Finally, we find that the distributional characteristic of fresh fruit and vegetables tends to be high, implying that a policy which includes a subsidy to try and induce an increase in consumption of these goods will be more regressive than one which is based on a tax on saturated fat content alone. These results show that a policy that is designed to induce a movement towards a healthier diet will become more regressive as the policy becomes more targeted.

In exploring the welfare changes that result firstly when only a tax is applied to foods that are high in saturated fat, we find that the loss in welfare ranges from 1.21 to 2.06 per cent. The loss increases with the level of inequality aversion and when it is applied at a more disaggregate level. This reflects the concentration of goods that are subject to high levels of taxation in households that are socially deserving and the increased tendency for this to be the case as the tax becomes more precisely focused on products that are high in saturated fat. This result highlights an obvious tension inherent in the use of a tax to improve the quality of diets: poor diets tend to be concentrated in low-income households and therefore the more effective a policy is in targeting poor diets, the greater the distributional inequity in the fiscal burden. From a social perspective therefore the desirability of this type of policy demands a balancing of our desire to see improvements in the quality of poor people's diets, and the effectiveness of the policy in achieving this, with our desire for a more equitable distribution of wealth. Since healthy food tends to be consumed in wealthier households, coupling the tax with a subsidy on healthy food exacerbates the situation. Thus, while our particular subsidy (of 27 per cent) on fruit and vegetables is sufficient to more than compensate for the tax such that welfare increases, the increase in welfare decreases as the level of inequality aversion increases.

Overall our results support that contention that, when considering the costs of a fiscal food policy, the impacts of the intervention will be regressive. It should be recognised, however, that the benefits of the intervention are also likely to be unequally spread in society and if, as is likely, poor diets tend to occur in poor households the benefits may be progressive. Note, however, that this is dependent not only on the distribution of consumption, but also on the price responsiveness of different income cohorts. The potential for tension to exist between the distribution of cost and benefit is, however, highlighted by our paper: the more targeted the policy is at achieving the desirable health outcome the more damaging it is in terms of economic welfare. Achieving a balance between these competing ends is therefore a paramount consideration in designing such a policy.

References