Contribution of food prices and diet cost to socioeconomic disparities in diet quality and health: a systematic review and analysis

Literature search strategy

The search, conducted in 2014 using PubMed, Google Scholar, and Web of Science databases, identified publications linking food prices, dietary quality, and socioeconomic status. As shown in Figure 1, the literature search led to 819 abstracts dating to the year 2000 and later. Letters, editorials, and commentaries were excluded on the basis of titles and abstracts, leading to 720 full-text articles. The list was further restricted to 130 articles after 590 citations that did not pertain to nutrition economics or diet cost were excluded after review of the full text. Twenty-one articles dating prior to the year 2000 were included in the review on the basis of expert consultations and bibliographies of papers published by researchers, including the present authors. These processes yielded the 151 studies on diet cost and nutrition economics that formed the basis of the present review.

Data analysis

To confirm conclusions from the literature, the relations between the nutritional quality of food and the price of food were assessed using a French nutrient composition table for 1117 food items consumed by a representative sample of 2624 adults participating in the 2006–2007 national dietary survey INCA2.⁴⁰ The national food price database for France was obtained from the 2006 Kantar Worldpanel, based on annual food expenditures of a representative sample of 12 000 French households. For each food item reported as consumed by INCA2 respondents, its standard price was calculated by dividing the annual expenditures by the amounts purchased for all the food products corresponding to this item in the Kantar Worldpanel and available on the French market. Given that these prices were paid by a representative panel of consumers, the standard prices were weighted by the most frequently purchased forms of each food. After adjusting for preparation and waste, food prices in the database were expressed in euros per 100 g of edible portion. Median and mean costs of energy (i.e., energy cost in €/100 kcal) provided by foods were calculated for each main food group. Nutrient quality of foods was estimated using the SAIN:LIM ([score of nutritional adequacy of individual foods] ÷ [score of nutrient to be limited]) ratio, calculated as described.^41,42 Correlation analyses were conducted to assess the relation between energy density (kcal/100 g) and energy cost and between nutrient quality and energy cost.

Relation between food prices and adoption of healthier diets by lower-income groups

That food prices have a disproportionate impact on lower-income households is a direct consequence of Engel’s Law.⁴³ As incomes rise, the absolute amount of money spent on food increases, but the proportion of income spent on food drops.⁴³ Whereas average food spending in the United States has been estimated at around 11% of income, households in poverty can spend more than 25%.⁵ As a result, the impact of food prices on dietary choices in the United States tends to be underestimated by higher–socioeconomic status groups that are affected the least.^44,45 In France, households in the top decile of incomes spent 22% of disposable income on food in 2006, whereas those in the bottom decile spent as much as 29%.⁴⁶

In surveys on motivations for purchases, consumers indicate that, along with taste and convenience, the price of food is one of the main factors influencing their food choices. This is especially true for lower-income households.^47,48 Multiple socioanthropological studies have shown that food prices weigh more heavily on purchase decisions made by lower-income groups.^49–51

Repeatedly, lower-income respondents have identified economic factors as the main barriers to adopting healthier diets. The high perceived cost of fruit and vegetables was noted,⁵² as was the high price of meat, perceived as an integral part of the core diet.^51,53 Although lower-income groups did not purchase less meat, they bought cheaper and fattier cuts.^46,49,54 Poor women in Australia, when asked what foods they would include in the market basket if given another 25% to spend, selected foods of higher nutritional value.⁵⁵

In focus groups, low-income women in the United Kingdom⁵³ and in Holland⁵⁶ have likewise stressed the key role of food prices. Foods reputed to be unhealthy were considered to be cheaper than healthier foods. They were easier to find and to prepare, and children preferred them because they were tastier. Low-income mothers in St Paul, Minnesota, reported in focus groups that they knew fruit and vegetables were healthy and that they would like to eat them more often and give them to their children, but that they could not afford to do so.⁵⁷ Similarly, participants in a qualitative survey in Paris⁵⁸ recognized vegetables, fruit, dairy products, fish, and fresh produce as “healthy,” and said that they would buy them more often if only they had the means. Perceived as a dietary priority for children, dairy products were viewed as affordable, unlike fish, fresh vegetables, or fruit.⁵⁸ These families' diets were monotonous, which is partly explained by the reluctance to try new foods, given the potential risk of food waste, which is an intolerable situation in poverty.⁵⁸

In the United States, the observed links between socioeconomic status and diet quality were partly explained by the perceived importance of food prices.⁵⁹ Among lower-income groups, a perception of food price as very important was linked to more energy-dense diets.⁶⁰ By contrast, the perceived importance of good nutrition did not differ by socioeconomic status.⁶⁰ Having positive attitudes toward nutrition did not remove the critical influence of socioeconomic status on diet quality.⁶¹

Merging food prices with nutrient composition data

Self-reports by lower-income groups, as reported in the literature, suggest but do not prove that the social gradient in diet quality is partly caused by higher prices of more nutrient-rich foods. Researchers studying the nutrient density of foods in relation to their monetary cost need access to detailed data on food nutrient composition and price. It was not until the late 1990s that data on the nutritional composition of food products were introduced into surveys on food purchases⁶² and – conversely – data on food prices were introduced into surveys on food consumption.⁶³ In 1994, Australian researchers estimated the price of each of the 229 foods on the frequency questionnaire used in a food survey of the general population²⁵ using supermarket food prices. Since that time, retail food prices have been obtained for research purposes from a variety of sources, both public and private.^{26,29,31–33,35,64–67} In general, food prices were incorporated as a vector in the nutrient composition database and were treated as an additional “nutrient.”

Both France and the United States have national food price databases that are available for research purposes. The French database, based on purchase data from the Secodip panel (later called the Nielsen and then the Kantar Worldpanel), was compiled in 1997 to analyze diet costs in the national INCA1 dietary survey.^29,64 An updated Kantar food price database created in 2007 was merged with the INCA2 nutrient composition data to generate the illustrative tables and figures for the present review.

The US Department of Agriculture’s national food price database,⁶⁸ which corresponds to the National Health and Nutrition Examination Survey (NHANES) 2001–2002 data, was released in 2008. Other than one update for 2003–2004 NHANES, no further updates have been released. However, that database was sufficient to link national food prices in the United States with the NHANES food recalls³⁵ and with food frequency questionnaires.^33,35 Other researchers obtained prices from the National Food Survey in the United Kingdom²⁶ or the Department of Trade and Tourism in Spain.^31,32

Food prices were also obtained from supermarkets and the Internet,^37,65,67 or directly from a distributor.⁶⁶ When the prices were from supermarkets or the Internet, researchers developed pricing protocols, selecting multiple items or lower-cost foods in family-sized packaging.^65,67 When the prices were drawn from surveys on food expenditures, the average price paid (mean price weighted by the quantities bought) was calculated.^68,69 For example, the average price of green beans reflected the price of fresh, canned, and frozen green beans of several brands, weighted by the quantities consumed by the population.²⁹ Supermarket data have been used to assess the energy cost and nutrient content of a freely chosen market basket of foods.⁷⁰

In all cases, fresh foods as purchased were then transformed into edible foods as consumed, adjusting for preparation and waste. Since cooked pasta absorbs water, 100 g of cooked pasta costs 3 times less than 100 g of dry pasta. Since banana peel is discarded, 100 g of peeled banana costs 1.3 times more than 100 g of fresh banana with skin. For research purposes, all food prices were converted into values per 100 g of edible portion and were merged with the nutrient composition data.

Estimating the monetary cost of diets using retail prices, whether local or national, does not always take temporal trends in diets or health into account. For example, a decline in food prices has been linked to higher obesity rates in the United States.^71,72 In reality, only the price of energy-dense foods⁷³ and sweetened beverages⁷⁴ has declined, whereas the prices of fresh vegetables and fruit and low-fat dairy continue to climb.^73,75,76 In France, the prices for vegetables have risen steeply, whereas the prices for dietary fats have dropped.⁷⁷ One study used correction factors based on a consumer price index to estimate the differential in food prices between the time of price collection and the time at which the dietary data were collected.⁷⁸

Lower cost of energy-dense foods

Comparative metrics of food cost have been based on energy and nutrient content expressed per 100 g of edible portion.⁶⁹ Table 2 shows the energy density (in kcal/100 g), the price (€/100g), and the energy cost (€/100 kcal) of several foods, as based on the INCA2 nutrient composition food database and 2007 prices of current food items, ranked by decreasing order of energy cost. Potato chips, sweets, and biscuits (cookies) were more expensive in terms of euros per kilogram than were low-energy-density apples, tomatoes, and carrots but were cheaper in terms of euros per 100 kcal. This is because apples, tomatoes, and carrots have a higher water content than do sweets and fats. As noted by Atwater in 1896,⁷⁹ water in foods drives energy density because it provides bulk but no energy or nutrients. Given the economic importance of obtaining 2000 kcal per day at an affordable cost, the preferred method of comparing food prices has been per calorie, as opposed to per serving or per unit weight.^79,80

In general, grains, fats, and sweets were associated with lower per-calorie food costs (energy cost, in €/100 kcal). By contrast, fruits and vegetables were associated with higher per-calorie food costs. As a result, the relation between the energy density of foods and the energy cost was negative.^27,28,66,81 Figure 2 shows the negative relation between energy density and energy cost of foods, based on 2007 food prices in France.

The results confirm previous findings obtained with 1997 food prices,^30,69 with energy-dense foods such as fats and oils, added sugars, and refined grains providing calories at the lowest cost, while low-energy-density lean meats, fish, vegetables, and fruit are the most expensive energy sources. Similar findings have been obtained in the United States,^79,82 Australia,⁸¹ and the Netherlands.⁶⁶ The inverse relation between energy density of foods and their energy cost held both between and within food groups.⁸³

Higher cost of nutrient-rich foods

Studies on the relative cost of energy-dense compared with nutrient-dense foods critically depend on accurate metrics of nutrient density. In the past, the relative healthiness of different foods was based on the most arbitrary of criteria.⁵ More recently, the nutritional value of different foods was established using more objective nutrient profiling techniques.^30,69,83

Foods can be ranked or rated on the basis of their nutrient content relative to calories. Typically, nutrient profile models represent the ratio of nutrients to calories. One nutrient profiling model, the French SAIN,LIM nutrient profiling system, is based on two scores: SAIN, the score of nutritional adequacy of individual foods, and LIM, the score of nutrient to be limited.⁴¹ Whereas SAIN is based on beneficial nutrients (protein, fiber, vitamins, and minerals) whose intake is to be encouraged, LIM measures the content of nutrients to be limited: saturated fat, added sugar, and sodium. In published studies, foods were divided into four categories on the basis of their relative SAIN and LIM scores.⁴¹ Overall nutritional quality of foods was also estimated by the SAIN:LIM ratio.⁴² Importantly, the concept of nutrient density was clearly distinguished from the energy density of foods.⁸²

The US-based Nutrient Rich Foods Index calculated the nutrient content of foods, expressed as the percent daily value per 100 kcal.^84,85 The Nutrient Rich Foods Index was based on 9 nutrients to encourage – protein; fiber; vitamins A, C, and E; calcium; iron; magnesium; and potassium – and on 3 nutrients to limit – saturated fat, added sugar, and sodium. The final algorithm was the sum of percent daily values for the 9 positive nutrients minus the sum of percent daily values for the 3 nutrients of public health concern. These two nutrient profiling models have served as the basis of several quantitative studies on the relation between the nutritional quality of foods and their cost.^30,42,69,83

Multiple studies based on French food databases have shown that higher nutrient density scores were directly linked to higher per-calorie cost.^30,69 In general, the more nutrient-rich foods were more expensive per kilocalorie than were foods of lower nutritional value.⁸⁶ However, there was a wide variability in nutrient density and cost within food groups.³⁰ For example, not all vegetables were equally expensive. The question of concern was which vegetables, including juices and soups, provided the most nutrients per unit cost.⁸⁷ Nutrient density was measured using the Nutrient Rich Foods Index, and food cost was calculated per 100 g, per 100 kcal, and per serving. Analyses showed that tomato juices and tomato soups, dark green leafy and nonleafy vegetables, and deep yellow vegetables, including sweet potatoes, had the highest Nutrient Rich Foods Index scores overall. Highest Nutrient Rich Foods Index scores per dollar were obtained for potatoes, tomato juices and tomato soups, carrots, and broccoli. Processed vegetables, soups, and juices providing affordable nutrition⁸⁷ are often the choices reported by groups of lower socioeconomic status.³⁹

Estimating monetary cost of individual-level diets

The ability to estimate diet costs is a prerequisite for studies on the relation between socioeconomic status, diet quality, and health.⁸⁸ The collection of such data, however, has been hampered by methodological problems. First, dietary and economic data at the national level tend to be collected by different agencies. In the United States, the NHANES, the basis for many policy decisions, provides no data on food prices or food expenditures.³⁵ Conversely, surveys of food expenditures collected by the US Bureau of Labor Statistics lack data on food consumption or health and are collected at the level of the household.

Studies in nutritional epidemiology that use individual health outcomes require that all dietary variables be at the individual level. Data collected at different levels of aggregation pose problems for any analyses of diets and health. In particular, household food expenditures are difficult to link to individual health outcomes. One approach to studying the relation between diet quality and diet cost, adopted early by the present authors, was to attach standard food prices or individual food expenditures to dietary intake data.⁸⁸

Following the design of classic intermethod validity studies,^89–91 Monsivais et al.⁹² compared three alternative methods of estimating individual diet costs. First, 4-day food diaries were linked with actual food expenditures as documented by store receipts. Second, 4-day food diaries were linked to local supermarket food prices. Third, food frequency questionnaire data were linked to local food prices. Attaching supermarket prices to dietary intake data allowed the calculation of the intrinsic monetary cost of each study participant’s diet.⁹² All three estimates of diet cost were significantly linked to household incomes.⁹² Another study also compared the estimated diet cost with either true expenditures (grocery receipts) or a vector of standard food prices.⁷⁸ The results showed that the mean diet cost estimates for the whole population were similar with the two methods, but that at the individual level, estimates differed by as much as £3.00/d.⁷⁸ Another study also found relatively weak agreement between diet costs estimated with standard prices and with prices actually paid by individuals.⁹³

Although the expenditure-based method might be thought to be the true indication of diet cost, it had no parallel in nutritional epidemiology studies. Dietary surveys do not estimate the nutrient value of each food as consumed by each study participant. Instead, researchers have estimated individual energy and nutrient intakes on the basis of standard nutrient composition tables. For example, the vitamin C content of fresh, raw apples, as consumed by NHANES participants, has been set at 4.6 mg per 100 g of edible portion in the Food and Nutrient Database for Dietary Studies nutrient composition database.⁹⁴ That estimate may or may not accurately reflect the vitamin C content of the apples that the NHANES participants had actually consumed. In the studies described below, a cost vector (i.e., standard food prices) was inserted into the nutrient composition database.

The key to modeling individual diet costs is to treat the monetary cost variable as analogous to a nutrient variable.²⁶ Individual daily diet costs are then estimated by attaching dietary intakes to national or local food prices.^{25–29,31–35,37,64–67,95–100} In epidemiologic studies, an adjustment for energy content is made to test the relation between a given nutrient and the outcome variable of interest.^101,102 Similarly, in most of the studies described below, an adjustment for energy was made to test the relation between diet cost and the outcome variables of interest and to allow for comparisons across population subgroups.

This method of estimating individual diet costs on the basis of prevailing retail prices makes the strong assumption that foods are purchased at retail and prepared and eaten at home. Such an assumption may well underestimate the real variability of food prices and of the costs associated with individual food consumption. Food cost variations that may be attributable to home preparation of certain foods, purchase of away-from-home foods, differences between brand foods and generic or low-cost foods, or differences between out-of-season and in-season foods are overlooked with the use of the average price. However, the exact same caveats apply to the average content of calories, fiber, vitamins, and minerals of foods. Nutrient composition data sets have been based on many of the same assumptions.

In general, researchers have treated food prices per 100 g of edible portion as equivalent to a nutrient composition vector and have integrated food prices into the nutrient composition database. That nutrient composition database was then used to estimate individual-level energy and nutrient intakes and individual-level diet cost on the basis of dietary intake data obtained using different methods. In published studies, standard food prices were linked with dietary intakes obtained using food frequency instruments or food frequency questionnaires.^{26,31,33,34,37,65,96–98,103} However, standard prices were also attached to data from diet records,^29,64 dietary history,^27,28 or 24-hour recalls.^15,66

Do healthier diets cost more?

The earliest studies on the higher cost of healthier diets were conducted in Australia and the United Kingdom.^25,26,104 Although rarely cited, those studies were among the first to model the cost of the recommended diets. Cade and Booth¹⁰⁴ attached supermarket food prices to dietary intake data in the United Kingdom to compare the relative cost of healthy and less-healthy diets. Diets consistent with a set of 6 nutritional guidelines cost 10% more, although the difference was not statistically significant.¹⁰⁴ McAllister et al.²⁵ attached supermarket prices to 229 foods to compare the costs of observed and recommended diets. The recommended diets provided 15% less energy, meaning that the energy-adjusted cost of healthier diets was higher.

The linear relation between diet quality and cost was shown, for the first time, in the 1999 study of diets of more than 15 000 women in the UK Women’s Cohort Study.²⁶ As diet quality, measured by the Healthy Diet Index, increased, so did the estimated daily amount spent on food. This relation persisted after adjusting for energy intakes. The Healthy Diet Index was based on saturated fatty acids, polyunsaturated fatty acids, proteins, carbohydrates, free sugars, and fibers and on two food groups (fruit and vegetables, and pulses, grains, and nuts). The positive relation between diet quality and cost was recently confirmed in the same cohort of women.³⁷ The dietary pattern that was most in accordance with the UK EatWell Plate guidelines cost twice the price of the least-healthy diet (£6.63/d vs £3.29/d) and was consumed by women with higher education level and higher occupation status.³⁷ A recent meta-analysis of food prices collected after year 2000 found a smaller difference of US $1.5 (average for 10 different countries) between the healthiest dietary patterns and the least healthy ones.³⁶

Healthier diets were also found to cost more using food- and nutrient-based measures of diet quality, such as the Mediterranean Diet Score,^99,105 the Healthy Eating Index,^99,100,105 the Dietary Approaches to Stop Hypertension score,¹⁰⁶ and the modified Healthy Eating Index.^33,35 A positive association was also observed between diet cost and dietary diversity,^37,100 and between diet cost and nutrient-based scores, whether nutrients were computed separately^{29,32,34,65,67,107,108} or aggregated into a composite measure such as the Mean Adequacy Ratio.^30,103 In some studies, data-driven dietary patterns (e.g., derived by cluster analysis) were scored according to how well they complied with a specific desirable food pattern.^37,95

Dietary energy density²⁷ has also been used as a measure of diet quality. Attaching 50 food prices to the dietary survey data from the Val de Marne study^27,28 showed that energy-dense diets were associated with lower diet costs.²⁷ By contrast, higher-quality diets were associated with higher energy-adjusted diet costs.

Similar results were obtained using the nationally representative INCA1 dietary survey and the retail prices for over 800 foods in France, based on purchase data from Secodip, a nationally representative panel of consumers.^29,64 Importantly, lower dietary energy density and higher nutrient content were each independently associated with higher energy-adjusted diet costs.⁶⁴ When dietary intakes data from two different populations in Spain were examined, the same direct correlation between diet quality and cost was found.^31,32,95 Similar results were later reported in a representative sample of adults in Greece⁹⁹ and in two cohorts of young adults and of elderly persons in Holland.⁶⁶ In the United States, the same correlations were found using food prices attached to data from diet records^65,96 and from food frequency questionnaires.^{34,67,103,107}

Most of the evidence on the relation between diet quality and diet cost was provided by cross-sectional studies, many based on samples of convenience. At least two sets of studies were based on nationally representative population samples in France^29,64 and in the United States.³⁵ Only one US study estimated the diet costs of a nationally representative sample of adults in the NHANES 2001–2002 study.³⁵ In that study, higher values of the Healthy Eating Index were linked to higher diet costs and higher respondent socioeconomic status.³⁵

In multiple studies, specific food and nutrient intakes were associated with lower energy-adjusted diet costs. First, the consumption of energy-dense foods (kcal/100 g) was associated with lower diet costs per 100 kcal.^27,65,66,96 Second, higher consumption of added sugars and saturated and trans fats was also associated with lower diet costs.^28,35,65,67 Sodium content of the diet was not systematically linked to energy costs.^35,108

The same associations between diet quality and cost held across different countries, across different age groups, and for different indicators of diet quality. Only one study conducted in Japan drew slightly different conclusions. Consistent with other findings, higher energy cost was linked to lower energy density. However, higher diet costs were linked to higher consumption of cholesterol and total and saturated fats,¹⁰⁸ which was not the case in either the United States³⁵ or France.²⁹ The Japanese study¹⁰⁸ was conducted with dietetics students whose fat consumption was far below international norms. The Japanese study also differed because the typical food consumed in Japan was different than that consumed in other countries included in this review.

Relation between different food groups and diet cost

Vegetables and fruit, recognized as the core components of a healthy diet, also account for a large part of diet cost. In the pioneering Cade et al.²⁶ study, fruits and vegetables drove the diet quality score and also accounted for the biggest share of the total diet cost. The observation that diets high in fruits and vegetables had highest per-calorie diet costs was subsequently confirmed in Spain,³¹ Greece,⁹⁹ the United States,³⁵ and Sweden.¹⁰⁰ In Japan, the highest proportion of diet cost was attributable to fish and vegetables.⁹⁸ Among low income women in California, greater vegetable variety was associated with higher Healthy Eating Index scores and lower dietary energy density, but also with higher diet costs.¹⁰⁹

In the average French diet, the highest proportion of diet cost was attributable to meat, fish, vegetables, and fruit.^27,28 These food groups account for a far smaller proportion of energy intakes. By contrast, starches, added fats, and fatty, sweet, and/or salty products provided a large proportion of energy, while contributing relatively little to the total cost of food.³⁰ The consumption of milk and milk products was cost neutral.²⁸ On the basis of the French INCA1 survey, the food group of meat/eggs/fish contributed only 18% of daily calories, while accounting for 35% of diet cost. Fruits and vegetables provided 8% of calories, while accounting for 17% of cost. Starches and added fats contributed 23% and 10% of the total energy intake but only 9% and 2% of the total cost, respectively. The energy contribution of dairy products was equivalent to their cost, around 11%.³⁰

Figure 3, based on the French INCA2 nutrient composition food database and 2007 food prices, indicates the mean and median cost of energy provided by foods in each of the major food groups (€/100 kcal). These results, consistent with those from the INCA1 food survey,³⁰ showed that the fruits and vegetables food group and the meat/eggs/fish group were the most expensive sources of dietary energy. The median cost was €0.82/100 kcal for fruits and vegetables and €0.64/100 kcal for meat/eggs/fish. Dairy products had an intermediate energy cost (€0.32/100 kcal). Fats and sweets only cost €0.22/100 kcal, and refined starches and added fats provided energy at the lowest cost (€0.14/100 kcal and €0.06/100 kcal, respectively).

These illustrative analyses confirmed the major points in the literature. Higher consumption of fruit, vegetables, meat, and fish was associated with higher diet costs. By contrast, higher consumption of fats and oils, added sugars, and refined grains was associated with sharply lower diet costs.²⁸ These analyses essentially confirm the classic 1936 studies of food, health, and incomes in the United Kingdom.²

Modeling dietary choices using linear programming

The question of what combination of foods provides optimal nutrition at the lowest monetary cost can be answered using a diet optimization technique known as linear programming. Initially proposed by Stigler in 1945,¹¹⁰ linear programming was used by Dantzig¹¹¹ to model optimal food patterns under a variety of constraints.

Linear programming modeling has become the method of choice when it comes to generating food patterns subject to a variety of nutritional, behavioral, and economic constraints.^{22,23,112,113} Linear programming models have the advantage of being able to isolate economic constraints and examine their impact on food selection. By contrast, empirical data on food choices by the poor can be complicated by a myriad of social, economic, and psychological factors. Linear programming models generally show that budget restrictions have a negative impact on food choices, orienting food patterns toward foods that are both energy dense²³ and nutrient poor.²² The US Department of Agriculture’s Thrifty Food Plan, designed to meet dietary guidelines at the lowest possible cost, is a related non–linear programming model.¹¹⁴

In general, linear programming models have tried not to depart too much from the current population diet by introducing new foods that are consumed rarely or not at all.¹¹² The British sociologist Liz Dowler¹¹⁵ has suggested that poor people sought to have the same eating habits as the people around them and not be hungry, rather than meeting nutritional needs at the lowest cost.

Modeling the culturally acceptable choices that people would make without going hungry on an increasingly constrained food budget was the topic of one study.²² The linear programming model removed all the nutritional constraints (except energy) and replaced them with social acceptability constraints, while progressively reducing cost. The resulting food patterns were lower in fruit, vegetables, meat, and fish but sharply higher in fats, sweets, and refined grains. The net result was a rise in energy density and a corresponding decline in nutrient quality of modeled food patterns.^22,23 Meat, the lone exception, was not reduced in food patterns of lower-income groups even when it was reduced by the linear programming model.³⁹ Social norms and the symbolic role of meat may ensure that it retains its central role, despite its high cost.⁵¹

The food patterns generated by linear programming models were striking similar to those of low–socioeconomic status groups.³⁹ In linear programming models, the selection of food patterns was guided by the necessity to reduce the cost of daily calories and, therefore, by the need to include foods that were inexpensive sources of energy. The observation that the modeled food patterns were of higher energy density suggests a potential link to obesity and weight gain, as hypothesized by the present authors.²³ Clearly, one conclusion based on the literature is that many unhealthy food choices associated with poverty may be a direct result of economic constraints. Adopting the recommended diets of lower energy density and higher nutrient content is likely to be associated with higher per-calorie diet costs.^116,117

Deliberately selecting an energy-dense diet with no limit on cost,²³ however, led to a relatively small decline in diet cost. It would appear that limited food budgets force low-income consumers to select low-cost, low-quality diets, whereas the higher–socioeconomic status groups have the choice of eating well or eating poorly.²³ This suggestion is consistent with observations that low-quality diets are not restricted to poor people but are nevertheless more frequent in lower– as opposed to higher–socioeconomic status groups.^24,39

Inadequacy of food budgets in poverty

The relation between the cost of healthy food and the household's food budget is of crucial importance. Several studies have compared the price of a balanced market basket of foods to the food budget of poor people in the United Kingdom,^118–120 Canada,^121,122 Ireland,¹²³ and Australia.¹²⁴ Based on different populations and different methods, the overall conclusion of those studies was that poor people could not afford a balanced healthy diet.¹²⁵

In the United Kingdom, Nelson et al.¹²⁰ concluded that poor people’s food expenditures were below the threshold for a low-cost-but-acceptable market basket of healthy foods. In Ireland, the share of the budget devoted to a balanced diet of “no-brand” foods was estimated at between 40% (for an elderly person living alone) and 80% (for a single-parent household with 1 child) of disposable incomes.¹²³ In Australia, adhering to the national food guidelines would take 40% of income of the poor, defined as those receiving a living allowance from the state, even with a market basket of cheapest or private-label products.¹²⁴ A 25% budget share was considered acceptable,¹²⁴ with fruit and vegetables accounting for the largest share of the food budget.^124,126 However, in many studies, the definition of a balanced diet was based on experts' guidelines, and the methodology for defining a basket was not described clearly enough to be reproduced. This constitutes a significant limitation.

In France, linear programming modeling was used to generate low-cost nutritionally adequate diets.¹¹³ Starting with a dietary intakes database, nutrient adequacy standards, and a list of foods of known nutritional composition and prices, linear programming modeling created food patterns that met nutritional guidelines as well as social acceptability constraints, all at the lowest possible price. The linear programming model was first applied to the Val de Marne data, using 50 food items,¹²⁷ and later to the INCA1 database, using 614 food items.¹¹³ The same theoretical minimal price for a nutritionally adequate diet was obtained from both sets of data, i.e., €3.20/d for women and €3.40/d for men. This amount was about 2 times lower than the average daily food expenditure in France (€6.5/d) but was close to the food budget of people living in poverty.⁴⁶

Based on the above analyses, the minimum food budget for a nutritionally adequate diet was estimated at €3.5/d. However, that amount imposed significant practical limitations. Even on €3.5/d, the theoretical access to an adequate diet meant zero food waste, drinking water from the tap, knowing how to cook, and not eating out at all. Below that cost level, the linear programming model failed to come up with a mathematical solution. In other words, having a nutritious diet at less than €3.5/d was impossible, not only in practice but also in theory. Yet the average food budget of people receiving food assistance in France was estimated at €2.5/d in 2005,¹²⁸ well below the theoretical minimum for a nutritionally adequate diet.

In fact, when all consumption and cultural acceptability constraints were removed, it was technically possible to cover an adult’s energy and nutritional needs for under €1.5/d (i.e., 4 times lower than the average expenditure for food in France). However, the food patterns generated were hardly realistic, containing only oil, pasta, potatoes, a little wheat germ, carrots and radishes, milk, liver, and herring.¹¹³ Food is not only a matter of nutrition; it is also a marker of cultural and social identity. To be fully effective, dietary recommendations need to be culturally acceptable.

The US Department of Agriculture (USDA) Thrifty Food Plan (TFP) was designed to find foods and combinations of foods that met dietary recommendations, respected the current eating habits of lower-income Americans, and minimized cost.¹²⁹ The price of TFP market baskets has long served to define the amount of institutional food aid in the United States. The calculated cost of the TFP was between $5.00 and $6.00 per day for adults aged 19–50 years ($37.50 and $42.10 per week for women and men in November 2013). The Economic Research Service of the USDA has estimated mean current food expenditures per capita at $14.00/d.¹³⁰ Therefore, in line with similar comparisons in France, the minimum cost of a modeled balanced diet and the average expenditure on food varied by more than a factor of 2.

Several authors in the United States have described the TFP amount as insufficient. First, as noted by Wilde and Llobrera,¹³¹ the cost of each new revised TFP was not recalculated de novo by taking into account trends in foods habits and food prices; it was simply subtracted from the preceding TFP in order to take inflation into account. Further, prices of food on the TFP menu were underestimated by comparison with real prices.¹³² The TFP was called unrealistic, not allowing for the growing use of processed, precooked, and away-from-home foods¹³³ or taking into account the time and effort associated with preparing TFP recipes.¹³⁴

What are affordable acceptable, nutrient-rich foods?

The modeling of French eating habits has shown that low-cost yet nutritious food patterns are theoretically possible, provided that the budget exceeds €3.5/d. However, such optimized food patterns may include foods that are rarely eaten by the general population.^113,131 Officially, healthier diets involve increasing the consumption of fruit, vegetables, fish, lean meats, and whole grains, as well as decreasing the intake of animal fats and sweets¹¹⁶; such a pattern was obtained in diets designed to fulfill nutrient recommendations.¹¹² However, cost constraints drove food patterns away from the official dietary guidelines. Although the amount of meat in food patterns was reduced, organ meats and eggs were increased. Furthermore, fresh fruit and dark green vegetables in modeled food patterns were replaced by root vegetables (potatoes, carrots), pulses, nuts, and fruit juices.¹¹²

In order to have a balanced diet on a small budget, it is necessary to select particular foods with a very good ratio of nutritional quality to price. These foods have been identified by linear programming models^42,112 or on the basis of favorable nutritional profiles.^30,42,83 In general, pulses, nuts, oils, and whole-grain cereals were foods with very good ratios of nutritional quality to price. Among animal products, milk, eggs, poultry, organ meats, and canned sardines are also cited as foods with a good ratio of nutritional quality to price.^42,83

Figure 4 shows the relation between nutrient quality of foods (estimated by the SAIN:LIM ratio) and energy cost (in €/100 kcal). Although the relation was positive, there was a wide dispersion around the regression line. Whereas certain foods (such as sweets) had a mediocre nutrition-to-price ratio, other foods had very favorable nutrition affordability metrics. Milk and carrots in particular offered optimal low-cost nutrition. Those foods were preferentially selected by linear programming models to construct nutritionally adequate diets at the lowest possible cost.⁴²

Do healthier diets have to cost more?

Although healthier diets generally cost more per calorie than do less-healthy diets, it is possible to eat better for less?²⁵ For any level of diet quality, there will be a range of diet costs. Similarly, food patterns characterized by the same per-calorie diet cost can vary in quality. In general, creating food patterns that are both healthy and affordable may involve some compromises on taste, cost, convenience, and cultural acceptance. The USDA TFP¹²⁹ was criticized for monotony and unappealing menus.^133,134

Studies on the economic impact of improved diets have reported mixed results. In early studies, lower-fat diets were associated with higher food expenditures.¹³⁵ Lower-income families in the United Kingdom¹³⁶ and Denmark¹³⁷ found some dietary interventions more difficult to follow because of their higher cost. However, other studies based on self-reports and small convenience samples claimed that better diets led to unchanged^138,139 or even lower food expenditures.^140,141 Results for the Mediterranean-style diets were equally mixed. Studies conducted in Spain showed that people who spontaneously adopted a Mediterranean diet spent more on food.^31,95 By contrast, women who were encouraged to adopt a Mediterranean style diet in Canada did not report spending more.¹³⁹

Although healthy food patterns generally costs more than less-healthy ones, there will be some that do not. A healthy Mediterranean-style diet can include seafood, salads, and scampi and be expensive, or rice, lentils, and beans and be cheap.¹⁴² Linear programming models aimed at generating nutritionally adequate food patterns at the lowest possible cost generally impose nutrition standards as well as cost and consumption constraints.^112,113,131 Whereas healthy food patterns can be created relatively easily, the progressive introduction of cost constraints causes the optimized food patterns to depart sharply from existing ones. In other words, the modeled lower-cost food patterns can be nutritionally adequate, but at the cost of departing from social norms.¹¹²

Role of food substitution in diet quality and diet cost

One key question is whether substitution between similar food products allows consumers to obtain diets of comparable nutritional quality but at lower cost. People who need to reduce diet costs while maintaining existing food patterns begin by making substitutions within and eventually between food groups. In economic terms, foods that replace each other given a change in conditions are known as substitute goods. Thus, an increase in the price of whole fruit may drive consumers in the direction of fruit juices, with no consequences for diet quality or nutritional status. Indeed, lower–socioeconomic status groups consume more fruit juices, whereas higher–socioeconomic status groups consume more fruit.³⁹

Study results on substitution effects have been inconsistent. McAllister et al.²⁵ found that substitution of certain foods by their equivalents of better nutritional quality led to increased diet costs and no improvement in diet quality. A UK study of 5 food items (canned tomatoes, orange juice, potatoes, sausages, and sliced bread)¹⁴³ suggests that substituting national brands with cheaper private labels would reduce costs without impacting nutritional quality because the prices of brand products could be up to 4 times higher than the cheaper products with a similar nutritional content.¹⁴³

A study of 220 food products in France¹⁴⁴ examined items in 17 product categories, such as cold meats, cooked dishes (such as cassoulet, ravioli, couscous, lentils, and sausages), desserts (such as chocolate cream, vanilla ice cream), etc. A nutrient quality score was derived on the basis of the list of ingredients and the calorie, protein, and lipid contents. Nutrient quality was significantly higher for the national brands, which were also considerably more expensive (by a factor of 2.5). Additional studies in France conducted by the Observatoire de la Qualité de l'Alimentation used food labels and price data to examine differences in nutrient composition by sale price.¹⁴⁵ They confirmed that there was no systematic association between sale price and the fat, salt, or sugar content of food products of a particular type (e.g., corn flakes). Such price variability within the same narrow food category may explain why weak agreement was found between diets cost estimated with standard prices and those estimated with actual purchase prices.^78,92,93.

By contrast, the prices of recommended foods in Australia were reported to be substantially higher than those of less-healthy options most markedly (30% difference) for legumes, canned fruit, beef, canned fish and margarine.¹⁴⁶ In Canada, margarines rich in saturated and trans-fatty acids cost less than the others,^147,148 and breads with whole-grain claims are less likely to be low in price than breads without such claims.¹⁴⁹

Whether discount food should be recommended to lower-income consumers is a problematic issue. On one hand, processed and frozen vegetables and fruit provide comparable nutrition at lowest cost.^87,150 On the other hand, some processed foods within the same food group may prove to be of lower nutritional quality than fresh foods. For instance, fresh fruit tends to provide greater nutrient content relative to canned options.¹⁵¹ It is very difficult to provide a clear-cut final answer to this question. The subject has been examined twice by the French National Food Council, in 2002^{1 52} and in 2010.¹⁵³ Both times, the conclusion was that existing data do not allow differences in the nutritional value of foods to be deduced from their differences in prices.

The high variability of food prices, both between and within food groups and between similar food items, may explain why wide variations in diet quality were observed at any level of cost.^33,106 Overall, the studies showed that, although it was more difficult to obtain good nutrition at a moderate cost, two conditions made it theoretically feasible. The first condition was to have the food budget exceed the strict minimum required for fulfilling the nutritional recommendations.¹¹² The second condition was to select foods with the highest nutrient-to-price ratio.⁴² Those population subgroups whose higher-quality diets were not associated with higher diet costs^33,106 may well have made those selections. For example, Hispanic adults in the United States had higher-quality but lower-cost diets as compared with non-Hispanic adults.^35,106 This was likely due to higher consumption of foods with a good ratio of nutritional quality to price.^42,83,154

Linking food prices and diet quality with socioeconomic status

The literature shows that more-expensive and higher-quality diets were consumed by higher–socioeconomic status groups with better health outcomes. In the United States, food-insecure households had lower estimated diet costs, expressed in dollars per day and in dollars per 2000 kcal, based on analyses of the 2001–2002 NHANES database.³⁵ In Sweden, the cheapest and most unhealthy diets were consumed by those children whose parents were the least educated and had manual and low-skill occupations.¹⁰⁰ By contrast, higher–socioeconomic status groups in the United States had higher diet quality and higher estimated energy-adjusted diet costs.³⁵ In other studies also, higher socioeconomic status was uniformly related to higher household food expenditures and to higher estimated diet costs.^15,67,96,155 A study on supermarket baskets in Phoenix, Arizona, showed that shoppers of low socioeconomic status purchase calories in an inexpensive form that is higher in fat and less nutrient dense.⁷⁰ Cross-sectional positive associations between diet quality and cost were also found within different socioeconomic status strata.^65,96

The Seattle Obesity Study found that the relation between socioeconomic status and diet quality was attenuated once diet costs were included in the model.¹⁰⁷ Socioeconomic status measures included education and incomes, whereas measures of diet quality were based on the mean adequacy ratio. First, there were significant positive correlations between socioeconomic status and diet quality, socioeconomic status and diet cost, and diet quality and diet cost, adjusting for numerous demographic covariates. The inclusion of the diet cost variable in the regression model attenuated the correlation between socioeconomic status and diet quality, suggesting that the impact of socioeconomic status variables on diet quality may be mediated by diet cost. This study¹⁰³ was the first to provide empirical confirmation of hypotheses^1,5,39 and models^22,23 by showing that the cost of food contributed to explaining social inequalities with regard to nutrition. However, given the cross-sectional nature of the data, it was not possible to establish a causal link between the three factors studied: socioeconomic status, diet quality, and energy-adjusted diet cost.

Dietary guidelines for the prevention of noncommunicable diseases have emphasized the consumption of nutrient-rich foods and the selection of high-quality diets.¹¹⁶ Although this dietary advice has been aimed at all strata of society, diet quality also depends on socioeconomic status. Lower–socioeconomic status groups have lower-quality diets³⁹ as well as higher rates of obesity, type 2 diabetes, and cardiovascular disease.^156–158 By contrast, higher–socioeconomic status groups have better diets and better health.^20,159 One explanation for the observed social gradient in diets and health is provided by disparate food prices and diet costs.^1,5,67 The literature on the observed links between socioeconomic status, diet quality, and diet cost suggests some potential consequences for weights and health.¹¹

Very few studies have related diet cost to health indicators. Studies based on cohorts of university students have found a negative correlation between energy cost, body mass index, and waist measurement in Japan¹⁰⁸ and a positive association between low energy cost and 5-year weight gain in Spain.⁹⁵ Finally, in a Greek cohort, no correlation was observed between food cost (in euros per week) and the 5-year incidence of cardiovascular disease.⁹⁹ Longitudinal data have been scarce. By some reports, the lack of money at the end of the month led to a drop in fruit and vegetable consumption.¹⁶⁰ Elsewhere, a drop in income¹⁶¹ or a loss of job security¹⁶² was found to be associated with weight gain.