The quality and monetary value of diets consumed by adults in the United States

Abstract

Background: Food prices are an established determinant of food choice and may affect diet quality. Research on diet cost and diet quality in representative populations has been hindered by lack of data.

Objective: We sought to explore the distribution of diet cost and diet quality among strata of the US population and to examine the association between the 2 variables.

Design: In this cross-sectional study, monetary costs of diets consumed by participants in the 2001–2002 NHANES were estimated with the use of a national food price database. Healthy Eating Index (HEI)–2005 values were estimated with the use of the population ratio method for the calculation of average scores. Mean daily diet costs, energy-adjusted diet costs, and HEI-2005 scores were estimated for subpopulations of interest. Associations between energy-adjusted diet cost, HEI-2005 scores, and HEI-2005 component scores were evaluated.

Results: Higher energy-adjusted diet costs were significantly associated with being older and non-Hispanic white, having a higher income and education, and living in a food-secure household. Higher diet costs were also associated with higher HEI-2005 scores for both men and women. Women in the highest quintile of diet costs had a mean HEI-2005 score of 69.6 compared with 52.5 for women in the lowest-cost quintile. Higher diet cost was strongly associated with consuming more servings of fruit and vegetables and fewer calories from solid fat, alcoholic beverages, and added sugars.

Conclusion: Given the observed association between diet cost and diet quality, helping consumers select affordable yet nutritious diets ought to be a priority for researchers and health professionals.

INTRODUCTION

Food choices and eating habits are driven by concerns about nutrition and health, as well as by food economics (1). The higher cost per calorie of some recommended healthful foods, such as fruit and vegetables, may restrict their use by people with limited resources (2–6). Research on the relation between food prices and diet quality can help explain the observed social disparities in nutritional status across socioeconomic strata (7). Although dietary surveillance in the United States at the individual level has not accounted for food prices historically, a national food price database has now become available (8).

Based on individual nutrient intakes, dietary energy density, fruit and vegetable consumption, or adherence to a healthful diet overall, studies in Europe and Japan have linked higher diet costs with higher-quality diets (9–15). In other observational studies, a higher diet cost was positively associated with biomarkers of nutrient intake and was negatively associated with BMI (11, 14, 16). In 2 US-based studies, persons with higher energy-adjusted diet costs consumed fewer calories and had higher intakes of vitamin C, potassium, and fiber (17, 18). This relation between diet quality indicators and diet cost was more pronounced for women than for men (17). Recent analyses of the Nurses’ Health Study data showed a positive association between the alternative HEI⁴ score and energy-adjusted diet cost, and, notably, that not all components of a healthy diet were cost sensitive (19).

As noted in a recent USDA report (20), analyses of diet quality in relation to diet cost in the United States have been hampered by a lack of appropriate data. On one hand, dietary intake and health databases, all at the individual level, have lacked information on the monetary cost of the foods consumed. On the other hand, databases on food expenditures and the economics of food purchase behavior typically provide food-purchasing data at the household or higher levels of aggregation, rather than at the individual level. Moreover, such data sets provide no details on actual food consumption or health at the individual level.

In the present study, individual-level diet costs for 4744 participants in the nationally representative 2001–2002 NHANES survey were calculated by attachment of the national food price database to dietary intake data. This linkage allowed for a descriptive characterization of the relation between diet cost and diet quality across socioeconomic strata, all at the individual level.

SUBJECTS AND METHODS

NHANES

The source of dietary intake data in NHANES 2001–2002 was a single 24-h dietary recall, in which respondents reported all foods and beverages consumed the previous day, from midnight to midnight. The 24-h recall data set included the amount in grams and a description of each individual food and beverage consumed, based on an 8-digit USDA food code. All recalls identified as reliable by NHANES staff and that met the minimum criteria for completeness were included in this study. The examination protocol, data collection methods, and factors used to evaluate the reliability and quality of each 24-h recall are documented elsewhere (21). Individuals who completed a reliable 24-h recall were included in the present study (n = 4744 of 5027 adults aged ≥20 y).

Data from the demographic and food-security questionnaires were also used in the present analyses. Primary stratification variables from the demographic questionnaire were age group (20–29 y, 30–44 y, 45–64 y, 65–74 y, or ≥75 y), sex, race-ethnicity (Mexican American/other Hispanic, non-Hispanic white, and non-Hispanic black), family income-to-poverty ratio (<2, 2–3.99, or ≥4), and educational attainment (<high school, high school graduate/equivalent, some college, or college graduate). The Mexican American and “other Hispanic” groups were combined because of small numbers in the “other Hispanic” group. Individuals who indicated that they belonged to another race group or multiracial group were included in all analyses, but their cost and diet quality estimates were not presented because this group was heterogeneous and the point estimates were highly variable because of the small sample size. Family income-to-poverty ratio adjusts for the number of adults and children in each family. Analyses of educational attainment were limited to adults aged ≥25 y, because most adults have completed their education by this age. Household food security was assessed by an 18-item Food Security Survey Module. Individuals were dichotomized as either fully food secure or not fully food secure (with or without hunger).

Food price database

The CNPP price database, released in May 2008, provided the cost per 1 g edible portion of all foods and beverages reported in NHANES dietary recalls, with the exclusion of alcoholic beverages and water (8). The prices were based on retail prices paid by members of the Nielsen Homescan Consumer Panel during the same period of 2001–2002 NHANES data collection and reflected the average prices paid by households across the United States. In creating the database CNPP made the assumption that all foods and beverages were obtained from stores. Accordingly, the database did not permit estimations of actual food expenditures. Rather, merging of the CNPP database with the NHANES provided estimates of the monetary value of foods consumed, which can be used in the assessment of the “economic accessibility” of a healthy diet.

Full documentation of how the food price database was created is available from the CNPP (8). In brief, all foods reported in the diet recall were translated into a purchasable form. Mixed dishes that were prepared from a recipe were decomposed into ingredients in purchasable form. For example, a portion of vegetable lasagna reported in NHANES was broken down into 17 separate ingredients, which included 3 types of fresh cheese; canned tomato paste; raw spinach; and cooked, enriched macaroni. Each food had to be in purchasable form and was adjusted for preparation and cooking losses and gains. In some cases, mixed dishes were deemed to be in ready-to-eat or ready-to-heat form, which means that no ingredients were analyzed because the dish could be obtained either from a deli section or refrigeration case. Next, the list of all foods and beverages in purchasable form was matched to price data for the foods and beverages reported in the Nielsen panel. Finally, prices were computed per gram edible portion based on the average purchase price. The prices for a small number of foods and beverages (n = 32) were missing or were identified as potentially extreme within-group outliers and were substituted with prices for comparable foods (eg, the price per gram of grape nuts was 3 times higher than that of similar cereals, or prices for some berries such as loganberries and mulberries were missing). This substitution was done by the authors.

Additional data from the CPI Average Price Data Series were used to estimate the prices of alcoholic beverages (22) that were not available from the CNPP. Prices were available for beer, wine, whisky and vodka for 2001–2002. Prices in the CPI may be systematically different from those in the CNPP database, because the CPI database is used for retail price monitoring, not to reflect consumer expenditures. In light of this, we first compared the prices of select foods with known prices from the CNPP database with the prices from the same year's CPI, to evaluate the degree by which the CNPP database may under- or overestimate the prices compared with the CPI database. The comparison foods were fresh celery, ground coffee, and white bread. These items were selected based on the availability of price estimates for multiple years in the CPI, the high frequency of consumption of these foods in NHANES, and the variability of food groups represented. We showed that prices per edible portion were 10.4% lower in the CNPP database compared with the CPI data, so the CPI alcohol prices were downwardly adjusted by this amount to account for this systematic difference. The final step was to attach these adjusted prices to the CNPP database, including the creation of recipes for mixed drinks with the use of common drink recipes. For example, the estimated price per 1 g of vodka tonic was 0.33 × price of vodka/g + 0.67 × price of tonic/g multiplied by the portion in grams consumed.

Deriving monetary value

The monetary value of the diet was computed from each individual's dietary recall in combination with the price database by multiplication of the price per gram by the portion of each food consumed by the respondent, and then the summing of these values for each participant. Diet cost was estimated for all foods and beverages, including alcohol. Tap and bottled water were excluded from the monetary cost estimation.

Because the monetary value of the diet was highly correlated with the total quantity of food and energy consumed (r = 0.72 for grams and r = 0.78 for kilocalories), the energy-adjusted diet cost was computed and expressed per 2000 kcal with the use of the residual method of energy adjustment (23). The distributions of both energy and diet cost were positively skewed, so energy and cost were log transformed and back transformed into natural units to ensure that the energy-adjusted diet cost variable was uncorrelated with energy. The energy-adjusted cost variable is more meaningful when comparisons are made between subpopulations that may have different energy requirements or intakes. The energy-adjusted diet cost variable was grouped into quintiles for all analyses.

The HEI-2005

The HEI-2005 was the primary outcome variable. As documented elsewhere (24, 25), the HEI-2005 measure of overall diet quality was based on consumption of 2 nutrients (sodium and saturated fat) and 9 food/beverage groups (total fruit, whole fruit, total vegetables, dark green and orange vegetables, milk, total grains, whole grains, meat and beans, and oils) and a measure of empty calories (SoFAAS). Each HEI-2005 component score can contribute 5 points (eg, total fruit or total grains), 10 points (eg, meat and beans, or energy from saturated fat), or 20 points (SoFAAS). Higher HEI-2005 scores indicate better diet quality; the maximal possible score is 100. The overall HEI-2005 score and each subscore are calculated on a per-calorie basis and are therefore adjusted for energy. For saturated fat, sodium, and energy from SoFAAS, higher HEI subscores reflected lower intakes. For the remaining 9 HEI-2005 food groups to encourage, higher subscores reflected higher intakes.

Analytic approach

Age-adjusted diet costs and energy-adjusted diet costs by independent variables of interest were estimated by the fitting of survey-weighted linear regression models with diet cost as the outcome, and the independent variable of interest (eg, race-ethnicity or income-to-poverty ratio) and age group as covariates. The predicted marginal for each level or category was estimated to determine the mean diet cost for each category or level of the independent variable. These values represented the mean at the average age distribution for the population of interest. A comparable approach was used to estimate the age-adjusted mean HEI-2005 scores by demographic and sociodemographic variables of interest. This approach is described in more detail below.

Because a single 24-h dietary recall does not sufficiently capture usual dietary intakes, which are ultimately the target of dietary recommendations and guidance, we used the score of the population ratio method recommended by Freedman et al (26) in analyses of HEI-2005 scores and subscores. This method has been shown to best reflect group-level dietary patterns when only a single 24-h recall is available. The method adapted for use here is described in detail elsewhere (27). A survey-weighted linear regression model was first separately fit for total energy and each subscore (eg, total fruit consumption) that included the independent variable of interest (eg, quintiles of energy-adjusted diet cost) and covariates (eg, age group). The predicted marginals were then estimated for each level of the categorical independent variable. With the use of the formula for each component and the covariate-adjusted calorie estimate, the HEI-2005 subscores were then estimated. This step was repeated for each independent variable of interest. Each subscore was truncated at the specified score (5, 10, or 20), and the total HEI-2005 score was the sum of the truncated subscores. The SEs of the adjusted HEI-2005 and the subscores were estimated with the use of a stratified jackknife variance estimator, which was the approach described by Korn and Graubard (28) and used by Breslow et al (27). The SEs and subsequent hypothesis tests based on this method were notably more conservative than the Taylor-series linearization SE estimates with survey data. However, the reduction in bias of an HEI-2005 score that reflected actual intake greatly outweighed the loss of statistical precision. For hypothesis testing for unordered categorical predictors, a survey-weighted Wald test was used. For ordered categorical variables, a linear trend test was used. P values <0.05 were considered statistically significant. When a significant trend or association was observed, pairwise tests were conducted with the use of a survey-weighted Wald test.

Analyses of the association between energy-adjusted diet cost and HEI-2005 used the same approach, with adjustment for age group, sex, and race-ethnicity. Effect modification by sex was expected, based on previous reports, and a multiplicative interaction term was significant (P < 0.01 in a model with adjustment for age group and race-ethnicity) (18). An exploratory analysis evaluated the association between diet cost and the HEI-2005 subscores, with adjustment for age group and race-ethnicity. The adjusted means for each subscore by energy-adjusted diet cost quintile were estimated for men and women separately with the use of the same score of population means approach described above for each subscore. The HEI-2005 was calculated with the use of SAS software (version 9.1; SAS Institute) and statistical analyses with the use of Stata software (version 11.0; StataCorp).

RESULTS

Diet cost and HEI-2005 by sociodemographic strata

Diet costs and HEI-2005 scores by demographic strata are provided in Table 1. The age-adjusted mean diet cost was $4.81 and the energy-adjusted diet cost was $4.44 per 2000 calories. The average HEI-2005 score was 58.5. Both diet costs per 2000 calories and the HEI generally increased with age. Women had higher HEI-2005 scores than did men.

TABLE 1.

Age-adjusted mean daily diet cost, energy-adjusted daily diet cost, and HEI-2005 score by demographic and sociodemographic strata among US adults¹

		Daily diet cost2
	n	$	$/2000 kcal	HEI-20052
Total	4744	4.81 (4.64, 4.98)	4.44 (4.33, 4.55)	58.5 (57.5, 59.5)
Age group
20–29 y (ref)	916	5.03 (4.78, 5.27)	4.15 (3.92, 4.38)	54.6 (52.2, 56.9)
30–44 y	1256	5.26 (4.98, 5.53)	4.41 (4.28, 4.55)a	56.5 (52.0, 61.0)
45–64 y	1417	4.69 (4.49, 4.90)a	4.59 (4.46, 4.72)b	62.7 (61.0, 64.3)c
65–74 y	584	4.17 (3.95, 4.40)c	4.74 (4.60, 4.88)c	68.4 (64.5, 72.3)c
≥75 y	571	3.51 (3.36, 3.66)c	4.32 (4.14, 4.50)	68.0 (63.1, 72.9)c
P-trend		<0.001	<0.001	<0.001
Sex
Male	2250	5.69 (5.48, 5.90)	4.51 (4.39, 4.63)	56.2 (53.6, 58.7)
Female	2494	3.99 (3.84, 4.14)	4.38 (4.28, 4.48)	61.6 (59.7, 63.4)
P-difference		<0.001	0.032	<0.001
Race-ethnicity
Non-Hispanic white (ref)	2494	5.00 (4.80, 5.20)	4.53 (4.39, 4.67)	58.4 (55.7, 61.1)
Mexican American/other Hispanic	1200	4.25 (4.04, 5.20)c	4.17 (4.08, 4.27)c	60.1 (58.3, 62.0)
Non-Hispanic black	890	4.41 (4.16, 4.65)b	4.09 (4.01, 4.17)c	54.7 (52.2, 57.1)a
P-difference		<0.001	<0.001	0.015
Family income-to-poverty ratio
<2 (ref)	1879	4.13 (4.02, 4.24)	4.08 (3.99, 4.16)	55.3 (53.1, 57.4)
2–3.99	1259	4.94 (4.74, 5.15)c	4.41 (4.26, 4.56)b	57.6 (54.6, 60.6)
≥4.0	1291	5.38 (5.12, 5.63)c	4.78 (4.64, 4.93)c	61.3 (59.3, 63.4)c
P-trend		<0.001	<0.001	<0.001
Educational attainment (age ≥25 y)
<High school/equivalent (ref)	1247	4.36 (4.03, 4.68)	4.24 (4.06, 4.41)	53.6 (49.8, 57.3)
High school/equivalent	949	4.71 (4.49, 4.92)	4.31 (4.21, 4.41)	54.2 (52.1, 56.3)
Some college	1056	4.64 (4.44, 4.83)	4.49 (4.37, 4.60)a	61.1 (57.9, 64.3)b
College graduate	920	5.31 (5.00, 5.61)c	4.83 (4.64, 5.02)b	65.0 (62.2, 67.8)c
P-trend		<0.001	<0.001	<0.001
Household food security
Fully food secure	3529	4.89 (4.71, 5.06)	4.50 (4.39, 4.61)	59.7 (57.7, 61.7)
Any food insecurity	930	4.28 (4.04, 4.53)	4.11 (4.01, 4.22)	54.3 (51.2, 57.3)
P-difference		0.001	<0.001	0.011

¹P values for trend were obtained from a survey-weighted linear regression model that treated ordered variables as continuous. P values for difference were obtained from a Wald test from a survey-weighted linear regression and indicate any difference in the outcome. P values for pairwise differences were obtained from a Wald test from a survey-weighted linear regression model that compared the value of interest to a reference group (identified in parentheses). ^a–cSignificance of pairwise differences (compared with the reference group): ^aP < 0.05, ^bP < 0.01, and ^cP < 0.001. HEI, Healthy Eating Index; ref, reference.

²Values are means; 95% CIs in parentheses.

Differences in diet cost and the HEI were also observed by race-ethnicity. The age-adjusted mean diet cost for non-Hispanic white adults was greater than among Hispanic and non-Hispanic black adults. There was some indication that Hispanic respondents had higher HEI-2005 scores than did non-Hispanic whites, but the difference was not statistically significant.

Lower family income-to-poverty ratio and educational attainment were associated with lower diet cost and lower HEI-2005 scores. Living in a food-secure household was associated with higher energy-adjusted diet cost and higher HEI when compared with individuals who lived in a household with any degree of food insecurity.

Modeling the relation between diet cost and HEI-2005

For men and women combined, and for both sexes separately, there was a significant and positive association between higher diet cost and higher HEI-2005 score (Table 2). Persons in the highest-diet-cost quintile had HEI-2005 scores that were 13 points higher than those with the lowest-cost diets. As expected, the association was stronger among women, as shown by the large difference in the values between quintiles of diet cost. HEI-2005 scores for men and women in the highest quintile of diet cost were, respectively, 9.7 and 17.1 points higher than for persons in the lowest quintile. Additional adjustment for educational attainment and income-to-poverty ratio did not alter the observed association between energy-adjusted diet cost and HEI (data not shown). Secondary analyses with the use of a ratio of cost per 2000 calories [(diet cost/calories consumed) × 2000] were conducted and yielded nearly identical results (results available from authors).

TABLE 2.

Adjusted mean HEI-2005 score by energy-adjusted diet cost quintile¹

	Combined (n = 4744)		Men (n = 2250)		Women (n = 2494)
Diet cost	n	HEI-20052	n	HEI-20052	n	HEI-20052
Q1 (≤$3.17) (ref)	949	50.8 (47.8, 53.8)	405	49.6 (46.0, 53.1)	544	52.5 (49.1, 55.8)
Q2 ($3.18–$3.77)	949	55.1 (52.5, 57.7)a	436	54.5 (51.7, 57.4)a	413	56.0 (52.7, 59.3)
Q3 ($3.78–$4.46)	949	56.9 (54.7, 59.0)b	431	53.3 (50.8, 55.8)	518	61.2 (57.8, 64.6)c
Q4 ($4.47–$5.41)	949	62.3 (60.1, 64.5)c	470	58.9 (54.9, 62.8)c	479	66.6 (63.7, 69.5)c
Q5 (≥$5.42)	949	63.8 (62.2, 65.4)c	508	59.3(55.8, 62.9)c	440	69.6 (66.6, 72.5)c
P-trend		<0.001		<0.001		<0.001

¹The combined model was adjusted for sex. A model adjusted for income-to-poverty ratio and educational attainment in addition to the adjustments in this model yielded similar results. P values for trend were obtained from a survey-weighted linear regression model that treated ordered variables as continuous. P values for pairwise differences were obtained from a Wald test from a survey-weighted linear regression model that compared the value of interest to a reference group (identified in parentheses). ^a–cSignificance of pairwise differences (compared with the reference group): ^a P< 0.05, ^b P< 0.01, and ^cP < 0.001. HEI, Healthy Eating Index; Q, quintile; ref, reference.

²Values are means adjusted for age group and race-ethnicity; 95% CIs in parentheses.

Modeling the relation between diet cost and HEI-2005 component scores

Higher-cost diets were associated with a significantly higher consumption of fruit and vegetables per 1000 calories, as shown in Table 3. The trend was apparent for both sexes but was stronger for women. Among men, only the highest-diet-cost quintile failed to earn the maximal HEI-2005 subscore for total grains. Among men, a weak negative association was observed for diet cost and the subscore for milk, whereas a nonsignificant positive association was observed for women. For neither sex did we observe a trend between diet cost and HEI-2005 subscores for whole grains. Higher-cost diets contained less saturated fat and less energy from SoFAAS as a percentage of total calories. The negative relation between diet cost and SoFAAS was markedly stronger among women. Higher-cost diets consumed by women included more meat and beans than did lower-cost diets. Of dietary components to limit, only consumption of sodium increased with more costly diets.

TABLE 3.

HEI-2005 component scores by energy-adjusted diet cost quintile¹

	Adjusted mean HEI-2005 component scores for men				Adjusted mean HEI-2005 component scores for women
	Q1 (n = 405)	Q3 (n = 431)	Q5 (n = 508)	P-trend2	Q1 (n = 544)	Q3 (n = 518)	Q5 (n = 440)	P-trend2
Total fruit [5]	1.5 ± 0.20	2.7 ± 0.21	3.5 ± 0.29	<0.001	1.9 ± 0.18	3.7 ± 0.33	4.9 ± 0.28	<0.001
Whole fruit [5]	1.5 ± 0.25	2.8 ± 0.26	4.6 ± 0.46	<0.001	2.4 ± 0.25	4.1 ± 0.43	5.0 ± 0.42	<0.001
Total vegetables [5]	2.3 ± 0.16	3.0 ± 0.20	3.4 ± 0.14	<0.001	2.3 ± 0.14	3.5 ± 0.16	4.8 ± 0.25	<0.001
Dark-green and orange vegetables and legumes [5]	0.7 ± 0.11	0.8 ± 0.12	1.1 ± 0.13	0.009	0.6 ± 0.15	1.4 ± 0.23	2.5 ± 0.20	<0.001
Total grains [5]	5.0 ± 0.19	5.0 ± 0.20	3.9 ± 0.17	<0.001	5.0 ± 0.20	5.0 ± 0.23	4.9 ± 0.16	0.58
Whole grains [5]	0.7 ± 0.08	1.1 ± 0.12	0.8 ± 0.13	0.40	1.2 ± 0.16	1.2 ± 0.11	1.4 ± 0.14	0.19
Milk [10]	5.2 ± 0.43	6.0 ± 0.22	4.4 ± 0.37	0.044	5.1 ± 0.27	6.6 ± 0.47	5.8 ± 0.39	0.052
Meat and beans [10]	8.6 ± 0.38	10.0 ± 0.32	10.0 ± 0.59	0.058	8.1 ± 0.22	9.5 ± 0.35	10.0 ± 0.69	0.009
Oils [10]	7.4 ± 0.45	6.1 ± 0.33	6.5 ± 0.45	0.054	7.6 ± 0.38	7.1 ± 0.42	6.9 ± 0.48	0.52
Saturated fat [10]3	6.0 ± 0.53	4.7 ± 0.35	8.9 ± 0.52	<0.001	6.8 ± 0.38	6.3 ± 0.35	8.8 ± 0.45	<0.001
Sodium [10]3	5.0 ± 0.27	4.2 ± 0.29	4.2 ± 0.31	0.003	4.7 ± 0.24	3.4 ± 0.29	2.7 ± 0.33	<0.001
Calories from SoFAAS [20]3	5.8 ± 0.79	6.9 ± 0.63	8.1 ± 0.42	<0.001	6.8 ± 0.88	9.4 ± 0.58	11.8 ± 0.57	<0.001

¹Values are means ± SEs adjusted for age and race-ethnicity. Numbers in brackets indicate the maximal score for each subscore. HEI, Healthy Eating Index; Q, quintile; SoFAAS, solid fats, alcoholic beverages, and added sugars.

²P values for trend were obtained from a survey-weighted linear regression model that treated diet cost quintile as continuous.

³Higher scores represent lower consumption of saturated fat, sodium, and calories from SoFAAS.

DISCUSSION

To our knowledge, this is the first report to characterize differences in diet cost by sociodemographic groups with the use of individual-level data from a nationally representative sample of the US population. A significant association between diet cost and HEI-2005 score was observed for the entire US population and for men alone, and it was especially strong for women. This association held after adjustment for additional sociodemographic factors (income and educational attainment). The difference in HEI-2005 scores between women in the highest- and lowest-diet-cost quintiles was greater than the difference between younger and older adults, which suggests that the observed differences are not only statistically significant, but also meaningful within a broader public health context. Within the United States, diet costs varied by population subgroups as characterized by race-ethnicity, education, and income. Non-Hispanic whites had higher diet costs than did non-Hispanic blacks and Mexican American/other Hispanics. However, despite lower diet costs, there was some suggestion that Hispanics had higher HEI-2005 scores than did non-Hispanic whites. We also identified some groups that consumed lower-cost, yet higher-quality diets; these groups included older adults, women, and Mexican Americans/other Hispanics (see Table 1). Studies of the relation between diet quality and diet cost at different levels of income and educational attainment can add to our understanding of the mechanisms behind the observed disparities in nutrition and health.

Recent dietary guidelines have emphasized an increase in the consumption of fruit, vegetables, low-fat dairy products, and whole grains, whereas energy from solid fat and added sugars is reduced (29, 30). An examination of each component score of the HEI-2005 separately in relation to diet cost showed that some component scores were associated with higher diet costs, whereas other component scores were not. Higher consumption of total vegetables, dark-green/orange vegetables and legumes, and whole fruit among both men and women was associated with higher costs. By contrast, total grains, whole grains, and milk and milk products were largely cost neutral. Food pattern modeling studies are required to determine whether the implementation of the 2010 Dietary Guidelines on a population basis would entail an additional diet cost.

It is important to emphasize that the present analyses were based on the intrinsic monetary value of the diet and not on actual food expenditures. For example, the present estimate of mean diet cost for a NHANES participant was $4.81/d. By contrast, the 2002 CEX estimated individual expenditure on food and alcoholic beverages to be $6.30/d, based on yearly purchases (31). Several factors may account for this 31% difference. First, the estimates based on NHANES data included only foods actually consumed, whereas CEX data were based on expenditures for all foods purchased but not necessarily consumed. Second, the assumption made by the USDA in creating the CNPP food price database was that all foods were purchased in retail settings and prepared at home. Whereas foods and beverages consumed in restaurants and other eating establishments are included in the current estimates, the CNPP food price database does not account for the added cost for service and preparation (8). In contrast, the CEX did include all food expenditures from retail stores and restaurants. Although the approach used here did not permit estimations of food expenditures, it provided estimates of the monetary value of the foods consumed. The latter measure might be useful in future research to evaluate the affordability of a healthy diet with current eating patterns.

The substitution of prices from restaurants and other food service establishments with store-bought foods for prices in the CNPP database represents one of the limitations of this study. Evidence from the CEX suggests that the purchase of away-from-home foods accounts for a greater proportion of food spending for individuals who live in more affluent households, which suggests that the differences by socioeconomic status observed here are likely to be even more profound. Another limitation is that the food price database was computed from prices paid by a national panel that lacked geographic specificity, which do not represent the prices actually paid by respondents in the sample. Variation may occur at the regional level (ie, differences in prices between the northeast and southwest regions of the United States) or local level (ie, difference in food prices between types or location of retailers within a city or region) (32). For these reasons, the observed differences of diet cost between sociodemographic subgroups were most likely conservative, especially if more affluent individuals were more likely to purchase foods from more expensive grocery stores or purchase more costly versions of a nutritionally comparable food (eg, organic free-range chicken compared with nonorganic chicken).

Moreover, the diet costs reported here are based on the 24-h recall dietary assessment, which can underestimate portion sizes or fail to include foods and beverages that are either forgotten or consumed infrequently, yet these limitations are inherent in all dietary assessment methods (8, 23). Together, these factors produce a downward bias on the diet cost estimate and contribute to the difference between the estimates observed here and the US consumer food expenditures in the CEX survey. In addition, because only a single 24-h recall was used to estimate diet cost, the distribution observed here is wider than would be observed if multiple recalls were available (23). The data used here are the only data for individual diet cost, diet quality, and health measures in a nationally representative sample.

However, the results of the present analyses, based on 24-h recalls and retail food prices, are consistent with past results. In past studies, dietary intake data obtained from food-frequency questionnaires (13, 14, 17–20), diet history questionnaires (11, 12, 33), food diaries/records (15, 34), and 24-h recalls (10) and were merged with local or national food store prices, as was done in this study. To our knowledge, no study has measured actual food expenditures at the individual level in relation to diet quality and health. The strengths and weaknesses of dietary intake data have been well documented (23). Many of the same issues relate to the estimation of diet cost (ie, error from misestimation of portion size or omitted foods or in the nutrient or price database may result in error in the diet cost estimate). Despite the diverse assessment methods and diet quality measures, the results of investigations into the association between cost and diet quality have been consistent.

In this study we observed an association between diet quality and diet cost: higher-cost diets tended to be higher in quality. The association between diet cost and diet quality observed here may be one mechanism that contributes to diet-related health disparities (35). Identification of features of low-cost yet nutritious diets is an urgent priority, given that under current consumption patterns, more-costly diets were associated with higher diet quality. Previous work has shown that eggs, beans, whole grains, citrus fruit, potatoes, and carrots were among affordable yet nutrient-rich foods (2, 20).

The goal of this observational study was to describe the association between diet quality and costs in the diets consumed by US adults. Whether high-quality diets would still be associated with higher diet costs if population eating habits were to change is a different question, best addressed by diet optimization modeling (36). Modeling studies, based on linear programming, can take nutritional and consumption data and cost constraints into account to create food-intake patterns that are nutritionally improved but cost neutral (37). One important avenue of research is the identification of individuals and groups that consume lower-cost yet higher-quality diets. In this study, these groups included older adults, women, and Mexican Americans/other Hispanics. Further examination of the diets of these groups is warranted and may help inform policies and guidance to promote food patterns that are both healthful and affordable. The combination of NHANES and the CNPP food price database remains a practical resource, useful in the evaluation of diet cost and diet quality in the US population.