Simulating the impact on health of internalising the cost of carbon in food prices combined with a tax on sugar-sweetened beverages

Strengths and limitations

This study has several strengths. It is the first study to investigate the potential conflicts between low carbon diets and adverse health consequences through simulating a tax based on GHGEs alongside a tax on SSBs. The tax scenarios analysed and the reported outcomes are the result of this unique price structure and account for resulting substitution and complementing effects. They are therefore not simply equivalent to adding results of previous analyses of a GHGE tax and SSB tax [7, 36]. Our analysis identifies other possible negative health consequences of a GHGE tax despite incorporating an SSB tax. These negative consequences include an increased consumption of cakes, buns, pastries, and biscuits, and undesirable changes to individual nutrients such as higher salt and lower vitamin A intake. Furthermore, we simulate realistic dietary scenarios with marginal shifts in consumption of different food groups whilst maintaining an adequate nutritional composition of the overall diet. We use contemporary data from routine sources and simulate not only own-price but also cross-price effects of changes in price. The econometric modelling uses methods which address censoring and ensure theoretical consistency.

There is no single data source that contains all the information required to conduct the modelling in this study and as such we use a range of data sources. All datasets are either sampled or weighted to be representative of the UK population, however, their strengths, limitations, and sample sizes vary. For example, UK census and mortality data are almost complete for the entire UK population whereas the LCF used for estimating baseline diets is based on a sample of 5,531 households [37]. A strength of this work is that both consumption data and price elasticities are based on the same dataset, unlike other studies in this area [13, 20]. However, fully linked datasets would have enabled estimation of differential mortality outcomes by age and sex. Furthermore, as the LCF is measured at the household level rather than the individual level it does not allow for estimating baseline diet or price elasticities by different age or sex groups; in reality these may vary.

Due to the amount of data available in the LCF, it was not possible to disaggregate foods into more than the 32 groups presented in Table 2. As greater resolution is introduced, there are more zero observations in the data and the uncertainty in the estimation increases. This is demonstrated by the particularly large uncertainty intervals around estimates of the change in consumption of some drinks categories (Table 2, Fig. 2). It is also likely that within each food group there will be variation in the changes to purchases of different foods which are assumed to be the same.

Elasticities are based on marginal weekly variations in prices and as such may not truly represent household’s responses to large changes in price (so-called price shocks) such as a 20 % tax on SSBs or a £1.79/kg increase in the price of beef (roughly equivalent to a 5 % to 45 % price increase depending on cut and quality of beef). Our use of unconditional elasticities allows expenditure to shift between food and drink budgets, but not outside of food budgets. It may be that people choose to spend less on their non-food budgets to make up for increased food prices thereby reducing the potential impact of the tax scenarios modelled.

Our greenhouse gas estimates are based on work by Audsley et al.; these represent the most complete collection of life cycle analyses for food consumed in the UK, including estimates based on whether the food is imported from Europe or elsewhere in the world [38]. However, GHGE estimates are unavailable for some food groups found in the LCF and on the FAOSTAT database. Therefore, as with Briggs et al. [7], it was necessary to assume that emissions for some foods were the same as related food products, and where no specific estimates were available for imported products, they were assumed to have the same emissions as if produced elsewhere in the world.

The social cost of carbon is estimated using all sources of GHGEs (from agriculture and elsewhere) and therefore to truly internalise the cost of GHGEs it would be appropriate to raise the price of all food products rather than just those with above average emissions. However, food is a necessity rather than a luxury and the intention of any GHGE based price change would be to shift populations to a less GHGE intensive diet. Therefore, as with Briggs et al., we only estimated price increases on those foods with above averages, with and without subsidies on food with lower emissions, as this is likely to be a more politically acceptable and less regressive policy rather than raising prices across all foods [7, 31]. Similarly, we simulated revenue-neutral policies where foods and drinks with GHGEs below average were subsidised. Sales taxes are regressive however this can be mitigated through redistributing the revenue generated through food subsidies or other progressive tax benefits [47]. Such redistribution is popular with the general public (as discussed by Cornelsen and Carreido, 2015 [48]).

As with previous work in this area, we assume that 100 % of the tax is passed on to the consumer and that all food is consumed [7, 13, 36]. In reality the pass-on rate may be higher or lower than 100 %, however French data from their soft drinks tax suggests that 100 % is reasonable [49]. All food purchased is assumed to be consumed; food waste is not accounted for within the LCF and it is possible that following a change in price, levels of food wastage will fall. Due to the likely differential effect of this between different food groups with different price changes, simulating this effect is not attempted here.

As discussed in the study methodology and consistent with previous modelling and empirical research, we assumed that liquid calories were non-satiating [36, 43–45]. The change in energy intake for each scenario shown in Table 3 is derived entirely from changes to liquid calories. These changes will vary for consumers with different baseline consumption levels of different drinks categories.

Credible intervals are based on the uncertainty in the coefficients used to compute price elasticities. This stage in the modelling has the most uncertainty and is therefore reported in preference to other areas of uncertainty (such as uncertainty in the parameters describing the relationship between food consumed and chronic disease). We do not include an estimate of the model’s structural uncertainty. The PRIME model is a cross-sectional model and therefore does not include any time component of how long the changes to diet would take to manifest in terms of changes to non-communicable disease mortality. The model instead reports the number of annual deaths that would be delayed or averted in the UK population were the population to have always been consuming the new diet compared to baseline.

Finally, all our estimates for changes to health are based on non-communicable disease mortality attributable to diet. We do not include any estimates of changes to morbidity, nor do we estimate the impact on health of reduced global GHGEs.

Comparisons with other studies

Using 2011 LCF and FAOSTAT data [37, 39], we estimate the total GHGEs related to food consumed in the UK including land use change (up to the retail distribution centre) to be 220,897 ktCO2e, and each simulated tax scenario reduces these emissions by a similar amount (between 7.4 % and 8.5 %). There is variation in the reductions to emissions between different scenarios (for example, scenarios B and D where foods low in GHGEs are subsidised lead to lower reductions than scenarios A and C), although these differences are not statistically significant. Total GHGE reductions are comparable to those previously reported by Briggs et al., who estimated that a similar tax structure to scenario A could reduce emissions from food in 2010 by 18,683 ktCO2e, 7.5 % of 2010 estimates of total emissions related to agriculture [7].

Briggs et al. allowed food as well as liquid calories to change resulting in very different results in terms of the impact on health. They estimated that a scenario equivalent to scenario A would lead to 7,768 deaths delayed or averted (with a 28 kcal reduction), and a scenario equivalent to scenario B would lead to 2,685 more deaths (with an increase in 21 kcal). The difference between the results for scenarios A and B published by Briggs et al. and those reported here is in part driven by changes to energy intake; when Briggs et al. kept energy intake constant, both scenarios A and B led to population health benefits with 1,207 and 2,536 deaths delayed and averted respectively. These are both larger than the equivalent results we report in this study (171 and 1,545 deaths delayed or averted in scenarios A and B respectively when energy remains constant). This due both to different diets in 2010 and 2011 and different tax structures. Data from LCF and FAOSTAT in 2011 suggest that diets are on average lower in GHGEs than in 2010, with the average emissions per kg food being 3.6 kgCO2e in 2011 compared to 4.1 kgCO2e in 2010. This is due to a combination of updated FAOSTAT datasets, changes to UK food import/export patterns, and changes to UK diets. The lower average emissions per kg mean that the simulated tax structure differs between this study and Briggs et al.’s study. Furthermore, price elasticities differ between the studies due to different LCF datasets being used. The fact that more items are taxed in this study than in Briggs et al. (beer is taxed in this study but is not in Briggs et al.) and the levels of taxation are greater (for example, beef is further from the mean and taxed at £1.79/kg compared to £1.76/kg in Briggs et al.) means that revenue is greater. We estimate that scenario A would generate £3.0billion, compared to £2.0billion in the equivalent scenario in Briggs et al. [7]. Given the differences found between this study and Briggs et al., further work should explore the robustness of modelled GHGE tax scenarios.

In this study we take things one step further than in Briggs et al. and identify that a combined food based GHGE tax along with a 20 % tax on SSBs would lead to significant additional health benefits and generate approximately £400million extra in revenue. The effect on change to GHGEs would be minimal compared to a GHGE tax alone. The tax rates simulated in the tax-neutral scenario B are based on the pre-tax baseline diet. Following the tax it is estimated that there would be an overall net loss in revenue of £540million as people move away from taxed products to those that are subsidised. We estimate that this loss would be largely offset by a 20 % SSB tax, resulting in net loss of just £120million in scenario D.

In a study estimating the effects of an SSB tax using similar methodology, Briggs et al. found that a 20 % tax in the UK could reduce calorie intake in adults by around 4 kcal/day [36]. This is in line with our estimates of a 5 kcal/day reduction in energy intake in scenario C. In scenario D, we estimate that there will be a non-significant 1 kcal/day reduction in energy intake, less than in scenario C due to subsidies in scenario D on fruit juice and milk, which have significant increases in consumption of 5.7 % and 16.5 % respectively.

The scale of the deaths delayed or averted is less than other studies in this area. However, many of the counterfactual scenarios simulated by other studies are much further from current dietary patterns than those used here. For example, Scarborough et al. simulated three sustainable dietary scenarios based on those proposed by the UK Committee on Climate Change Fourth Carbon Budget [12]. These include replacing 50 % of meat with fruit, vegetables and cereals. Friel et al. also simulated large reductions in meat consumption but without considering what calories may be replaced with [50].

Our results are consistent with work by Biesbroek et al. who used the EPIC-NL cohort to estimate GHGEs and mortality outcomes of 40,011 diets [19]. The authors found no significant increase in hazard ratios for all-cause mortality for those eating diets in the highest versus lowest quintiles of GHGEs. However, Biesbroek et al. went on to simulate the effect of a 35 g reduction in meat consumption (approximately a third of total meat consumption) on health and GHGEs. They estimate that, depending on the substitute, this could lead to between a 0 % (if substituted with potatoes) to 19 % (if substituted with fish) reduction in all-cause mortality, with the effects of substituting to other food groups, such as fruit and vegetables, all falling in between.

Finally, Edjabou and Smed simulated the effect of a GHGE tax on food in Denmark. Similar to our results, they identified that revenue neutral scenarios lead to less of a reduction in emissions than non-revenue neutral scenarios, and that saturated fat consumption decreases. They estimated that a tax rate of £26.90 per tCO2e/kg (2011 prices) would reduce emissions by 4 % to 7.9 % (depending on underlying consumption data used), which is comparable to our results of a 7.5 % to 8.5 % reduction following a £28.60 per tCO2e/kg tax.