Socioeconomic inequalities in the healthiness of food choices: Exploring the contributions of food expenditures

1A body of evidence shows that purchasing and consumption of unhealthy diets, in particular, eating fewer fruits and vegetables, is strongly patterned by socioeconomic status (SES). One likely contributor to the socioeconomic patterning in healthy diets is the cost of food: less nutritious, energy-dense foods are often cheaper sources of calories, and higher diet quality has been associated with higher diet cost. Moreover, given most research to date has estimated diet cost by linking dietary intake data to prevailing food price data, the socioeconomic inequalities in expenditure may have been underestimated by assuming a constant price for particular foods (i.e. only accounting for differences between types of foods purchased and not variation between brands). Even so, dietary cost explains some of the relationship between SES and nutrient density of consumed foods, and estimated diet cost has also been shown to mediate the pathway between socioeconomic status (income) and diet quality in a US sample.

2A potentially related avenue of research has suggested that consumers who patronize low-priced supermarkets are more likely to have lower-quality diets and higher BMI. Yet even within the same store, more educated households have been found to make healthier purchases. One contributing factor may be the prioritization of low cost, which may lead to preferences for certain supermarkets and also limit food choices within store. As such, concerns about cost may be driving some of the association between supermarket price tier and healthiness of diet. Conversely, if individuals choose to patronize a particular supermarket for reasons other than price (for example, believing a store to offer a wider range of healthy products), this choice may still contribute to subsequent diet cost, given expenditure is a consequence of customers' product choices. Both choice of products within store and choice of the store itself are likely to depend in part on individuals' motivations (e.g., price, health, convenience), which may vary by SES. As such, the relationship between diet cost and diet quality may be bi-directional. Although the extent of this bi-directionality cannot be determined in cross-sectional analyses, reverse mediation analyses offer an initial exploration of the potential contributions of each factor to socioeconomic inequalities.

3Of particular interest is the extent to which financial motivations mediate socioeconomic differences in healthiness of choices, given that this would likely indicate the potential effects of changing food prices on healthiness of diet. However, financial motivations are often not possible to reliably measure directly and instead food expenditure has often been used as a proxy. Food expenditure may be influenced by supermarket choice, which may in turn have been influenced by financial motivations, but may also have been determined by other motivations (e.g. convenience). As such, in the current study we will investigate whether food expenditure mediates the pathway between socioeconomic status and healthiness of choices, with and without controlling for supermarket choice. By supermarket choice, we potentially control for the likely self-selection to a given type of supermarket. On the other hand, without controlling for supermarket choice, we run the risk of ignoring that selection of supermarket is not necessarily a free choice (e.g. it may be limited by geographic access, which could be associated with SES). So the above mediation models may reflect upper and lower bounds. As such, this study aims to explore a range of values for the possible mediation of expenditure on the socioeconomic differences in healthiness of choices.

4This study extends explorations of the role of food costs as a mediator of socioeconomic inequalities in healthiness of choices; firstly, by looking at actual expenditure (rather than estimated diet costs) in a large UK sample. Secondly, reverse mediation between food expenditure and healthiness of choices will be explored. Thirdly, it will explore the associations between SES, food expenditure and healthiness of choices alongside the contribution of supermarket choice.

5Data were obtained from the Kantar WorldPanel (KWP) UK household survey from 2010 (as this involved analyzing de-identified existing data, ethical approval was not required). The sample consists of an ongoing panel, originally recruited via post or email to be representative of the UK in terms of age group, household size and region of residence. Households must meet minimum volume and spending criteria based on household size for inclusion, based on 4-week purchasing blocks. Further details of sample recruitment and quality control have been described elsewhere.

6Participating households (n = 24,879) recorded all food and beverage purchases brought home (i.e. excluding purchases that were consumed away from home), including volume purchased, spend, nutritional content, and the retail chain from which products were purchased. Sociodemographic data including number of adults and children in each household, ages and genders of household members, and socioeconomic indicators were also collected.

7Head-of-household occupation using the UK Registrar General's social class classification was categorized into three groups: Higher Managerial and Professional (‘Higher’: n = 5332); White Collar and Skilled Manual (‘Middle’: n = 13,621); and Semi-skilled and Unskilled Manual (‘Lower’: n = 5926).

8Expenditure was calculated from the households' total spend (£) on take-home food and beverages over the 52 week period, divided by the total number of calories those purchases for the same period, multiplied by 2000 to give an energy-adjusted food expenditure variable (£ per 2000 kcal).

9We defined supermarket choice as in: firstly, supermarkets were categorized as high-, medium- or low-cost based on market definitions. Households were then classified according whether or not they ever patronized high- or low-cost supermarkets (in addition to medium-cost supermarkets, which were almost universally patronized), giving four groups: Used low-cost supermarkets exclusively or low- and medium-cost supermarkets (‘Low-cost’); Used medium-cost supermarkets only (‘Medium-cost’); Used high-cost supermarkets exclusively or medium- and high-cost supermarkets (‘High-cost’); Used all three tiers of supermarkets (‘All-types’).

10Two outcome variables assessed healthiness of food and beverage choices, comprising less-healthy and healthier indices:Percentage of food energy purchased from less-healthy foods and non-alcoholic beverages, as classified by FSA Nutrient Profile scores for individual products (Scores are calculated from the energy, saturated fat, sugar, sodium, fiber, protein, and fruit, vegetable and nut content, per 100 g; foods scoring 4 or more, and beverages 1 or more, are categorized as less-healthy).Percentage of food energy purchased from fruit and vegetables — this included fresh, canned, frozen and dried fruit, vegetables and legumes, but excluded juice, potatoes, and fruit and vegetables present in processed products.

11Percentage of food energy purchased from less-healthy foods and non-alcoholic beverages, as classified by FSA Nutrient Profile scores for individual products (Scores are calculated from the energy, saturated fat, sugar, sodium, fiber, protein, and fruit, vegetable and nut content, per 100 g; foods scoring 4 or more, and beverages 1 or more, are categorized as less-healthy).

12Percentage of food energy purchased from fruit and vegetables — this included fresh, canned, frozen and dried fruit, vegetables and legumes, but excluded juice, potatoes, and fruit and vegetables present in processed products.

13Firstly, multiple regression analyses were conducted to explore the pathways linking ‘Socioeconomic status’ → ‘Food expenditure’ → ‘Healthiness of choices’ in this dataset, estimating:Food expenditure by SES (using dummy variables) (pathway ‘a1/a2’ in Fig. 1)The percentage of energy purchased from (a) less-healthy foods/beverages and (b) fruit and vegetables by:expenditure (pathway ‘b’); and/orSES (pathway ‘c’)

14Food expenditure by SES (using dummy variables) (pathway ‘a1/a2’ in Fig. 1)

15The percentage of energy purchased from (a) less-healthy foods/beverages and (b) fruit and vegetables by:expenditure (pathway ‘b’); and/orSES (pathway ‘c’)

16expenditure (pathway ‘b’); and/or

17SES (pathway ‘c’)

18Mediation analyses (conducted using the product of coefficients method with bootstrapped standard errors) then examined the role of expenditure as a potential mediator of socioeconomic inequalities in healthiness of choices (i.e. ‘Socioeconomic status’ → ‘Food expenditure’ → ‘Healthiness of choices’: pathways ‘a1/a2’, ‘b’ and ‘c′’ in Fig. 1), as well as the reverse pathway (the role of healthiness of choices as a potential mediator of socioeconomic inequalities in expenditure, i.e. ‘Socioeconomic status’ → ‘Healthiness of choices’ → ‘Food expenditure’ in Fig. 1 (‘a1/a2’, ‘c′’ and ‘d’)). Socioeconomic status was indicated by three ordinal levels of occupational social class, modelled using dummy variables, running separate analyses for Higher vs. Middle occupational social class and Higher vs. Lower occupational social class (with Higher occupational social class as the reference group in both analyses). These estimates were then aggregated to give the total indirect and direct effects of expenditure as a mediator of occupational social class in healthiness of choices. Model estimates reflect the difference in purchase of each food group associated with decreasing occupational social class. To examine the impact of supermarket choice on these relationships, analyses were conducted with and without controlling for supermarket choice (i.e. comparing ‘a1’ and ‘a2’).

19Analyses (using Stata MP version 13 (StataCorp, College Station, USA)) used robust standard errors, given evidence of heteroscedasticity, and the percentage of energy from fruit and vegetables and expenditure were log-transformed to address positively skewed distributions. Reported significance levels were adjusted for multiple testing using Bonferroni's correction. Analyses are reported in terms of a 20% increase in expenditure, which equates to an approximately £0.65 increase in spend per 2000 cal at the median value of expenditure (£3.24 per 2000 cal), and would move a household at the median value within each expenditure quintile into the quintile above.

20Regressions controlled for a number of potential confounders including age, gender, and ethnic group (white/non-white) of main shopper; number of adults in household, number of children in household, and for region of residence (Midlands, North East, Yorkshire, Lancashire, South, Scotland, Anglia, Wales & West, South West and London). Sensitivity analyses using hierarchical models by region, to account for possible clustering effects, produced very similar results, which we present in the Supplementary Materials.

21The sample characteristics, stratified by quintile of food expinditure, are presented in Table 1. Higher food expenditure was associated with higher mean age of the main food shopper and fewer children in the household. Highest-spending households were also most likely to identify as white and reside in the London region. Table 1 also revealed socioeconomic inequalities in expenditure, with those from lower occupational social classes and with lower incomes tending to fall into lower quintiles of expenditure. Supermarket choice appeared to be associated with expenditure, with 59% of those in the lowest quintile of expenditure shopping at low-/medium-cost stores compared to 16% in the highest expenditure quintile. Conversely, 3% of the lowest expenditure quintile vs. 29% of the highest quintile shopped at high-/medium-cost stores.

22Table 1 also suggested a trend with those in the higher quintiles of expenditure purchasing higher percent energy from fruit and vegetables and lower percent energy from less-healthy foods and beverages than those in lower quintiles of expenditure.

23Fig. 2 shows the results of a regression analysis estimating expenditure among occupational social class and supermarket choice groups, with both higher SES and higher-cost supermarkets being significantly associated with greater expenditure. The variation in expenditure between supermarket choice groups (approximately £0.90–£1 difference between groups) was greater than the variation between SES groups (around £0.50–£0.60). This variation by supermarket choice groups was consistent across SES groups, with heterogeneity in food expenditure even in the lower SES group.

24Table 2 shows the results of the three models of expenditure as a predictor of the healthiness of choices. In terms of expenditure, in the basic model (Model 1: expenditure and demographic variables only) a 20% increase (equating to a household at the median value within each expenditure quintile moving into the quintile above) was associated with a 0.7 percentage point decrease in the percent energy from less-healthy foods and beverages, whereas a 20% increase in expenditure was associated with a 7.2% increase in percent energy from fruit and vegetables. There was little change to the coefficients for expenditure between the different models (when adding occupational social class (Model 2) and then supermarket choice (Model 3)), although the increase in percent energy from fruit and vegetables was slightly reduced, to 6.4%.

25Fig. 3 shows socioeconomic inequalities in healthiness of choices, with and without controlling for expenditure, suggesting that socioeconomic inequalities in healthiness of choices was reduced when expenditure was included in the models for both outcomes.

26Table 3 shows the results of mediation analyses examining whether expenditure mediates the relationship between SES and healthiness of choices, with and without controlling for supermarket choice. For purchase of less-healthy foods and beverages 63% of the association was mediated without controlling for supermarket choice, whilst the equivalent figure for purchase of fruit and vegetables was 36%.

27For both outcomes, controlling for supermarket choice reduced the indirect effect (and the proportion mediated: to 53% and 31% respectively).

28The reverse mediation analyses were also conducted to examine whether healthiness of choices mediates the pathway between SES and expenditure. For percent energy from less-healthy foods and beverages, 11% of the total effect was mediated (indirect effect: − 0.03; Bonferroni-corrected 95% CIs: − 0.04, − 0.02) without controlling for supermarket choice group, and 13% when controlling for supermarket choice. Similar results were obtained for percent energy from fruit and vegetables (without supermarket choice group: 11%; indirect effect: − 0.03; Bonferroni-corrected 95% CIs: − 0.04, − 0.03; with supermarket choice group: 12%).

29Analyses from hierarchical regression models that clustered households within regions (rather than adjusting for region) showed similar results. Supplementary Table S1 shows percent energy purchased from less-healthy foods and from fruits and vegetables estimated from multivariable regression models (like Table 2) and again from hierarchical models. Additionally, estimated food expenditure across socioeconomic groups was similar whether based on hierarchical models or models that adjusted for region (Supplementary Table S2).

30This analysis of a large UK dataset, employing detailed scanner data, suggests that food costs may be an important contributor to socioeconomic inequalities in healthiness of food and beverage choices. Our analyses (using actual rather than estimated food expenditure) supported the findings of previous studies in that higher SES households were found to have significantly higher food spending and also had somewhat healthier patterns of food purchasing. Independent of SES, food expenditure had a small positive association with healthier patterns of food and beverage purchasing. In order to explore these relationships in more detail, we investigated the role of expenditure as a mediator of socioeconomic inequalities in healthiness of choices, going beyond previous research by analysing: (1) reverse mediation and (2) the likely range of values for these mediation effects in this dataset by comparing analyses with and without controlling for supermarket choice.

31Further exploration of the role of expenditure suggested this may be a significant mediator of socioeconomic inequalities in healthiness of food and beverage choices, as has been found in US studies, using different indicators of SES and diet quality. In addition, while the association between expenditure and healthiness of choices was larger for fruit and vegetables than for less-healthy items, the extent to which expenditure mediated socioeconomic inequalities was greater for less-healthy foods and beverages (63%) than for fruit and vegetables (35%). This may tie in with previous findings that less mediation was seen when looking at the mean adequacy ratio (representing micronutrients in the diet, and perhaps healthier purchases) than when using energy density, which is likely to reflect the ratio of healthier to less-healthy foods and beverages.

32However, the reverse mediation pathway (healthiness of choices as a mediator of socioeconomic inequalities in expenditure) was also significant, albeit appearing smaller. This highlights the need for caution when interpreting mediation results from cross-sectional analyses. The prioritisation of either price or health could lead to socioeconomic patterning of both food expenditure and healthiness of food choices: prioritising price may constrain healthiness of choices, while prioritising health may necessitate higher expenditure. Moreover, prioritisation of price vs. health may vary by socioeconomic group, with previous studies suggesting that lower SES groups are more likely than higher SES groups to prioritise price, and vice versa for prioritising health. One question of interest to public health researchers is whether changing food costs could influence healthiness of choices and/or on socioeconomic inequalities in choices. If the reverse mediation does indicate a bi-directional relationship, one consequence may be that the possible range of effects of price changes on socioeconomic differences in diet healthiness are prone to over-estimation in analyses where causality is not determined.

33Supermarket choice was systematically associated with food expenditure (see Table 1). To explore the role of supermarket choice in these associations between SES, food expenditures and healthiness of choices, we additionally controlled for this variable in the analyses, finding this had no or limited effects on the coefficients for expenditure on healthiness of choices. When controlling for supermarket choice in mediation analyses, the proportion of socioeconomic inequalities in healthiness of choices that was mediated by expenditure was reduced for both of the outcome variables, though even controlling for supermarket tier, we observed substantial mediation of socioeconomic inequalities by expenditure. Individuals' choice of supermarket may be determined in part by factors directly linked to expenditure, e.g. financial constraints, but are also likely to take account of other motivations (such as convenience, perceptions of quality or preferences for particular product ranges). Supermarket choice may influence expenditure for patrons to some extent regardless of their initial motivations. By comparing the mediation effects with and without supermarket choice, this study allows us to examine a likely range of values for the mediation effects of food expenditure in this dataset. As such, the influence of supermarket choice in these mediation analyses may in part represent other motivations that may influence healthiness of food choices and vary by SES, for example, health attitudes or knowledge.

34By exploring the role of expenditure across these different analyses, this study considers the potential contribution of food costs to socioeconomic differences in the healthiness of food choices. While each of the analyses in this study suggested that food costs were likely to be playing a role in socioeconomic differences in food purchasing choices, this paper highlights uncertainties in determining the size of this contribution. In order to explore this more fully, future studies investigating the pathways illustrated in Fig. 1, and in particular, integrating households' motivations in choosing stores and products, would be beneficial. Such analyses might help establish the extent to which these different pathways may reflect different households' behaviour (including the extent to which supermarket choice may result from financial constraints or other motivations), and to what extent this varies by SES. In particular, integrating these results alongside those of experimental or intervention studies could help to disentangle possible bi-directionality in the pathways between expenditure and healthiness of choices. Indeed, the findings are consistent with intervention studies that have provided financial incentives for improving diet; for example, cash-back and subsidy schemes are providing evidence of positive effects of changing food prices on healthiness of choices in low- and middle-income households. Unpicking the pathway between socioeconomic status, expenditure and healthiness of food choices in detail could help inform more effective programs and policies to promote healthier food choices.

35Several limitations need to be borne in mind, however; not least that the data were cross-sectional. Moreover, these results reflect purchasing, and as such may not translate directly to diet. It should be noted that the overall low volumes of food and beverages recorded in this dataset suggest underreporting (with households reporting on average approximately three quarters of the in-home calories, excluding alcohol, reported in a representative household survey of food spending from the UK in 2010). However, the underreporting does not seem to vary systematically by SES. In addition, the mean expenditure per calorie is very similar between these two datasets (Living Costs and Food Survey: 0.172 pence/kcal; study dataset: 0.171 pence/kcal).

36In summary, this study involved actual food expenditure data and allowed a more nuanced exploration of the potential role of this variable in the socioeconomic inequalities of healthiness of food choices than has previously been reported. These findings suggest cost is still likely to be a significant contributor to healthiness of food choices, and that a sizeable proportion of the socioeconomic inequalities in healthiness of choices may be mediated by expenditure in a large UK sample. This suggests actual and/or perceived cost of healthy diets may be key factors in tackling socioeconomic disparities in food purchasing choices.

37The authors declare that there are no conflicts of interest.