Do nutrition knowledge and beliefs modify the association of socio-economic factors and diet quality among US adults?
May A. Beydoun,Youfa Wang
DOI: https://doi.org/10.1016/j.ypmed.2007.06.016
IF: 5.1
2008-01-01
Preventive Medicine
Abstract:Results Multivariate analyses indicated that better SES independently improved likelihood of adequate fruits and vegetables intake and overall diet quality. In several cases, nutrition knowledge and beliefs acted as an effect modifier. In particular, education showed no association with diet quality among subjects in the lowest nutrition knowledge and belief tertile, while the association was consistently stronger in the highest tertile (Education × Nutrition knowledge and beliefs interaction term P < 0.10 for Healthy Eating Index and both fruits and vegetables guidelines). A similar interaction was noted for poverty income ratio. Conclusion For improvement in overall diet quality, socio-economic interventions must be coupled with health education programs targeting all segments of the US population. Keywords Diet quality Fruits and vegetables Socio-economic status Nutrition knowledge and beliefs Abbreviations aMED Alternate Mediterranean Diet Score CSFII Continuing Survey of Food Intake among Individuals DHKS Diet and Health Knowledge Survey EDU Education HEI Healthy Eating Index NKB Nutrition knowledge and beliefs PIR Poverty income ratio SES Socio-economic status USDA United States Department of Agriculture V&F Vegetables and fruits. Introduction A growing body of evidence suggests that socio-economic status (SES) and dietary intake are associated, and many previous studies show that the poor adhere less to dietary guidelines ( Ball et al., 2006; Billson et al., 1999; Bodnar and Siega-Riz, 2002; Forshee and Storey, 2006; Friel et al., 2003; Groth et al., 2001; Hulshof et al., 2003; Mishra et al., 2005; Pollard et al., 2001 ). A meta-analysis of 11 studies in seven European countries, concluded a positive association between SES and fruits and vegetables (V&F) consumption. Pooled estimates suggested a difference in intake of fruit by 24.3 g/person/day between men in the highest level of education and those in the lowest. The difference was larger in women (33.6 g/person/day) ( Irala-Estevez et al., 2000 ). Hence, disadvantaged groups exhibit poorer diet quality and have lower intake of V&F, a pattern associated with higher risk of morbidity in the form of overweight and obesity ( Guo et al., 2004; Kennedy et al., 2001 ), coronary heart disease ( Srinath Reddy and Katan, 2004 ), stroke ( Ding and Mozaffarian, 2006 ), hypertension ( Appel, 2000 ), inflammation ( Ford et al., 2005 ) as well as cancer and all-cause mortality ( Kant et al., 2000; Mai et al., 2005 ). Dietary behaviors including food choices are determined by a myriad of environmental and individual factors. Some of the individual factors include SES and psychosocial factors such as knowledge, beliefs and perceptions about nutrition and health. Similar to SES, diet and health-related knowledge, measured by various scales, has been shown to positively affect the quality of dietary intake as well as healthy food purchasing behavior among adults ( Gittelsohn et al., 2006; Havas et al., 1998; Krebs-Smith et al., 1995; Moser et al., 2005; Turrell and Kavanagh, 2006 ). However, no prior studies considered the interaction between SES and nutrition knowledge and beliefs in affecting dietary quality. In the present study, we tested the possibility of nutrition knowledge and belief being an effect modifier in the relationship between SES factors, V&F consumption and diet quality among US adults. We hypothesized that among subjects with healthier nutrition knowledge and beliefs, SES may improve diet quality as well as increase V&F consumption to meet recommended dietary guidelines. In contrast, subjects who have poorer nutrition knowledge and beliefs may be less affected by improvement in SES in terms of overall diet quality and intakes of fruits and vegetables. Thus, we test the possibility of a synergistic effect between SES and nutrition knowledge and beliefs. Methods Database Data from the US Department of Agriculture (USDA) Continuing Survey of Food Intakes among Individuals (CSFII) 1994–96 were used ( US Department of Agriculture ARS, 1994–96 ). Although this data set is 10 years old, it is the only available data set that provides comprehensive nationally representative data on all the needed study variables in this project such as detailed information on participants' nutrition knowledge, beliefs and perceptions, which are not collected in other more recent national surveys such as the National Health and Nutrition Examination Surveys (NHANES). This CSFII nationally representative multi-stage stratified sample of 16,103 non-institutionalized persons aged 0 to 90 years residing in the United States contained information about dietary intake (by one or two nonconsecutive, multiple-pass 24-h recalls that were 3 to 10 days apart); socioeconomic, demographic and health parameters. The 1994–1996 CSFII included the Diet and Health Knowledge Survey (DHKS). The two surveys allow linking individuals' dietary and health knowledge, beliefs and perceptions to their food intakes. The DHKS collected data from individuals aged 20 years or older who provided at least 1 day of dietary intake information (one person per household). The dietary intake interviews and the DHKS interviews were separated by at least 1 week. In our present analyses, only those respondents to the DHKS with 2 complete days of intake data were included. Study population In the CSFII 1994–96, a total of 20,126 individuals was initially selected into the sample. Among those, about 16,103 were respondents with a response rate of 80%. However, each respondent was assigned a sample weight which compensated for the following selection biases: (1) variable probabilities of selection, (2) differential nonresponse rates, and (3) possible deficiencies in the sampling frame. The analysis estimates obtained when sample weights being taken into account are nationally representative. Among 16,103 respondents to CSFII 1994–96 of all ages, 9872 were aged over 20 years with complete data on day 1 of recall. From those, only 5765 subjects (one per household) were sampled to complete the DHKS. We further excluded those over the age of 65 years ( n = 1319) and with only one 24-h dietary recall ( n = 90). As a result, our sample (CSFII/DHKS) consisted of 4356 individuals (2219 men and 2137 women). Variable definitions Socio-economic factors Two main exposures of interest were considered. First, education was measured by years completed and then grouped into 0: “< High school (HS) education”, 1: “High School” (12 years) and 2: “> High school education”. Second, poverty income ratio (PIR: as percentage of poverty line) was categorized as 0–130 (poor: food stamp eligible) ( Siega-Riz et al., 2004; US Department of Agriculture ), 131–299 (near poor), and 300 or more (not poor). Nutrition knowledge belief score One scale from DHKS was selected, consisting of 11 questions that were initiated by the following cue: “To you personally, is it very important (score of 4), somewhat important, not too important, or not at all important (score of 1) to”: a) use salt or sodium only in moderation, b) Choose a diet low in saturated fat, c) Choose a diet with plenty of fruits and vegetables, d) Use sugars only in moderation, e) Choose a diet with adequate fiber, f) Eat a variety of foods, g) Maintain a healthy weight, h) Choose a diet low in fat, i) Choose a diet low in cholesterol, j) Choose a diet with plenty of breads, cereals, rice, and pasta; k) Eat at least two servings of dairy products daily. Using principal components analysis or PCA ( Sharma, 1996 ), we extracted a single component score by imposing an eigenvalue criteria > 1.0, as well as examining the scree plot. The component which we designated as “nutrition knowledge and beliefs” (NKB) explained around 40% of the variance in the 11 items, which loaded almost equally on this component (loadings ranged from 0.23 to 0.35). This continuous score was then transformed into tertiles, ranking individuals according to their nutrition and health-related knowledge and belief, namely poor, moderate and good. We attempted to construct the same component using alternate means, including averaging the z -scores and factor analysis. Given that all three methods yielded highly correlated components ( r > 0.99), we selected the predicted component for NKB based on PCA which is the method of choice for data reduction. Fruit and vegetable consumption Based on responses from the two 24-h recalls elicited in the CSFII, including quantities and portion sizes, nutrient intake was estimated using food composition tables that were designed specifically to be used for this survey. In addition, the survey re-grouped the foods consumed by broader categories. We considered average dietary intakes from the 2-day 24-h recalls. Attention was focused on intake of vegetables and fruits (V&F) with serving estimates made available by USDA website http://www.barc.usda.gov/bhnrc/foodsurvey/ . According to recent dietary guidelines, adults are recommended to consume at least two servings of fruits and three servings of vegetables daily for energy intakes between 1200 and 3000 kcal ( US Department of Health and Human Services, and Agriculture, 2005 ). We created three binary outcome variables categorized as: 0: “does not meet the guidelines”, and 1: “meets the guidelines for (a) fruits, (b) vegetables, (c) total of five servings of V&F or more and (d) meets guidelines for V&F simultaneously”; i.e. V ≥ 3 and F ≥ 2. Diet quality indices Moreover, we estimated two widely known diet quality indices, namely the Health Eating Index (HEI) ( Fung et al., 2005; McCullough et al., 2000 ) and the Alternate Mediterranean Diet Score (aMED) ( Fung et al., 2005; Trichopoulou et al., 2003; Trichopoulou et al., 1995 ), as outlined in Appendix A. A HEI and aMED score was calculated for each subject, respectively. Our analysis shows moderate agreement between the two scores, (Spearman correlation r = 0.50, P < 0.05) a kappa = 0.15, P < 0.05 and % agreement of 32.2% (>expected agreement by chance of 20%) for age- and gender-specific quintiles. While HEI is measured on a scale of 0 to 100, aMED may range from 0 to 10. Covariates Covariates considered as potentially confounding variables in our models included: age (in years), gender, race/ethnicity (Non-Hispanic whites, Non-Hispanic blacks, Hispanics, Others), 1990 Census geographic regions (Northeast, Midwest, South, and West) and degree of urbanization (Metropolitan Statistical Area-central city, MSA — suburban, and rural). Statistical analysis First, we described sample characteristics observing design complexity for variance estimation. In particular, we accounted for stratification by geographic location, degree of urbanization, and socioeconomic characteristics. To test association between categorical variables, we used χ 2 test. To assess significance of differences in means across categorical variables, we conducted ANOVA test. We conducted logistic regression as well as ordinary least squares (OLS) regression models (Eqs. (1.1) and (1.2)). Next, we carried out stratified regression models to test whether associations between SES and different outcomes varied across tertiles of NKB score. Effect modification of the association between SES and dietary outcomes by NKB was concluded based on a statistically significant interaction term (NKB (1,2) × SES (1,2) ) tested in separate models, where subscripts 1 and 2 refer to the two dummy variables created to represent the contrast between the upper two levels of each variable in question compared to the lowest level. For logistic models, a statistically significant interaction term is interpreted as a significant difference between two odds ratios (OR) comparing level d = 1,2 of SES to level 0 of SES at stratum n = (1 or 2) of NKB, compared to the corresponding OR at level 0 of NKB (lowest tertile) (Eq. (2.1)). The same approach was used for OLS models, although the interpretation was in terms of regression coefficient differences (Eq. (2.2)). See Appendix B. In all tests, a P -value < 0.05 was considered as statistically significant. However, tests for significance of interaction terms were allowed a type I error of 0.10 due to their lower power compared to main effects models' regression coefficients ( Selvin, 2004 ). These analyses were conducted using survey commands in STATA release 9 ( STATA, 2005 ), which take design complexity into account, produce correct estimates of standard errors and with selection of appropriate weights, yield nationally representative estimates. Results V&F consumption and diet quality by study population characteristics Table 1 summarizes the characteristics of the study population. The highest proportion of subjects were aged 35–54 years (47.9%), 79.2% were non-Hispanic whites, more than a third lived in the South while 44% lived in suburban areas. The prevalence of adherence to recommended guidelines was consistently low in the South of the United States and lowest in rural areas. Age was positively associated with adherence to fruit guidelines only, both fruits and vegetables guidelines, high HEI and aMED scores. Men were more likely than women to adhere to vegetable, both guidelines and consuming either food group with a daily intake of five servings or more. However, they tended to have lower HEI and aMED scores compared to women ( P < 0.05). Diet quality and fruit and vegetable intake were dependent on ethnicity, with non-Hispanic blacks having consistently the poorest outcomes. Vegetable and fruit consumption and diet quality by SES SES factors were consistently and positively related to all interest dietary outcome variables ( P < 0.05). In the bivariate analyses, however, education did not exhibit a clear linear dose–response relationship as opposed to PIR. When ORs and regression estimates were adjusted for extraneous factors, namely age, sex, region, race/ethnicity and urbanization, SES remained significantly and positively associated with adherence to guidelines and diet quality, with the exception of the associations between education with two adherence outcomes: V ≥ 3 and V ≥ 3 and F ≥ 2. While education showed a threshold effect among subjects with more than HS, the positive association of income with diet quality and adherence to V&F guidelines was seen among the near poor (PIR = 131–299) as well as the non-poor (PIR > 300) categories, when compared to the poor (PIR = 0–130), with a linear dose–response relationship observed in all outcomes ( Table 2 ). Nutrition knowledge and belief as an effect modifier in the SES-F&V and SES-diet quality associations: stratified multivariate models A series of multivariate logistic regression models controlling for age, sex, race/ethnicity, geographic region and urbanization and stratified by NKB tertiles yielded adjusted OR for the association between SES factors and several indicators of adequate V&F intake. Statistical interactions were tested in these models, by adding interaction terms with NKB and SES dummy variables ( Tables 3 and 4 ). For HEI and aMED, using the same approach, linear regression coefficients were estimated. The association between education and different outcomes varied across NKB tertiles, albeit with inconsistent patterns. In particular, the uppermost tertile differed from the lowest in the case of vegetable, both guidelines, five servings and more, and HEI scores. The pattern indicated a stronger association with education in the uppermost tertile in the case of HEI ( β 12 = 3.6 vs. β 10 = − 1.4; β 22 = 8.6 vs. β 20 = 3.2; P < 0.10 for both interaction terms SES (1) × NKB (2) and SES (2) × NKB (2) ) and both guidelines (ORs of 2.17 vs. 0.47; P < 0.10 for interaction term SES (1) × NKB (2) ). However, the opposite pattern of interaction was observed for adherence to vegetable guidelines (0.88 vs. 1.44; P < 0.10 for interaction term SES (2) × NKB (2) ) and intake of five or more servings of either fruits or vegetables (1.24 vs. 2.11; P < 0.10 for interaction term SES (2) × NKB (2) ), even though these ORs did not reach statistical significance. It is worth noting, however, that none of the estimated ORs for education in the lowest tertile of NKB were statistically significant, indicating that among subjects with unhealthy nutrition knowledge and beliefs, education had no association with adherence to guidelines or high diet quality ( P > 0.05 for β = 0). This pattern was similar for PIR, whereby statistically significant ORs in the expected direction were observed more in the middle and uppermost tertiles of NKB. Two interaction terms – those for HEI and aMED scores – were significant. For HEI, results indicated that while PIR was positively associated with this outcome in all NKB strata, the association was significantly stronger in the upper NKB tertile ( β 22 = 7.0 vs. β 20 = 4.6; P < 0.10 for interaction term SES (2) × NKB (2) ). In the case of aMED, PIR had no significant effect on diet quality in the lowest NKB tertile and the difference between the poor and near poor became evident only in uppermost NKB tertile ( β 12 = 0.49 vs. β 10 = 0.01; P < 0.10 for interaction term SES (2) × NKB (2) ). Fig. 1 shows graphically the results of a multivariate logistic regression model with binary upper quintile outcomes being ‘high’ HEI and aMED. In all cases, subjects in the lowest NKB level did not shown any positive association between SES and diet quality, in contrast to the two upper NKB tertiles. In a separate interaction analysis, there was at least one SES × NKB interaction term that was significant in all SES/diet quality outcome pairs. Discussion This study contributed to the growing literature on the complex association between socio-economic (SES) factors and dietary intakes among adults in the United States. It aimed at assessing whether nutrition knowledge and beliefs modifies the association of SES with adequate intake of V&F as well as selected diet quality indices. Our main findings suggested that meeting guidelines for V&F and diet quality were positively associated with both education and income in the total study population. Further, we found that the positive association of SES with diet quality indicators and indices varied by nutrition knowledge and belief. In our multivariate logistic models, an interaction was found between NKB tertiles and educational level or income whereby this positive association was stronger among people in the two upper NKB tertiles. Importantly, among those in the upper tertile of NKB (healthiest nutrition knowledge and belief), education had a strong and a significant positive association with HEI score, while among those in the lowest tertile (unhealthy nutrition knowledge and belief), the relationship was non-significant. This highlights the importance of health/nutrition education, as education alone cannot help improve diet quality. Previous research suggested that consumption of V&F among other healthful food choices was positively related to SES. Findings from a study of 7000 British households indicated that compared to the highest income group (A), the lowest income groups (D and E2) consumed more milk (but less semi-skimmed milk), meat and meat products (of which more is higher fat meat products), fats, sugar and preserves, potatoes, and cereals. They consumed fewer fresh vegetables, fruits and high-fiber products such as brown and whole-wheat bread. The gap between these socioeconomic groups have widened between 1980 and 1995 ( James et al., 1997 ). Other studies conducted on various European and US populations came to the same conclusion regarding the positive association between SES and diet quality among adults ( Ball et al., 2006; Billson et al., 1999; Bodnar and Siega-Riz, 2002; Friel et al., 2003; Groth et al., 2001; Hulshof et al., 2003; Mishra et al., 2005; Pollard et al., 2001 ). To uncover costs of a healthy diet as a means to explain SES differences in dietary quality other studies have been conducted. One cross-sectional study which was carried out among 15,191 UK women concluded that to achieve a particularly healthy diet, independent predictive factors included spending more money, being a vegetarian, having a higher energy intake, having a lower body mass index (BMI) and being older ( Cade et al., 1999 ). The same higher cost pattern of healthy diets was later on confirmed by others ( Drewnowski, 2004; Drewnowski and Darmon, 2005a,b ). Further, healthy nutrition knowledge and belief as measured using various scales was shown to positively affect intake of V&F as well as overall diet quality. For instance, a recent study carried out among 1003 Australian adults suggested that SES differences in dietary knowledge represented part of the pathway through which education exerts an influence on diet; and food purchasing differences by household income were related to diet in part via food-cost concern ( Turrell and Kavanagh, 2006 ). Another recent study conducted among African-American men suggested that fruit consumption is motivated by perceived benefits and standards set by important people in their lives while vegetable consumption is a function of extrinsic rewards and preferences for high-calorie, fatty foods ( Moser et al., 2005 ). A third study conducted among American Indian populations showed that food acquisition and use behaviors were predicted by food use intentions while food intention scores were predicted by food self-efficacy and the latter by food knowledge. These findings supported the use of food knowledge, self-efficacy, and intentions in understanding food-related behavior in this setting ( Gittelsohn et al., 2006 ). Others came to similar conclusions regarding the importance of psychosocial factors ( Havas et al., 1998; Krebs-Smith et al., 1995 ). Our study has several additional interesting findings. For instance, there were ethnic and geographic differences in consumption of V&F as well as overall dietary quality scores, whereby adequate consumption was consistently least prevalent among non-Hispanic blacks and among subjects living in the South or in rural areas. We suspect that these could be part of the underlying causes of the ethnic and geographic disparities in the prevalence of obesity and overweight reported based on nationally representative data, which show that the prevalence in most southern states were higher than the other parts of the country and some minority groups such as non-Hispanic blacks were at higher risk than non-Hispanic white ( Ogden et al., 2006; Wang and Beydoun, 2007 ). Our study has several strengths. First, it added to the growing literature on SES disparities in dietary intake and patterns among adults in the United States and other developed countries ( Bodnar and Siega-Riz, 2002; Friel et al., 2003; Groth et al., 2001; Hulshof et al., 2003 ). Second, its main contribution relates to the effect modification of the association between SES factors and diet quality by nutrition knowledge and belief, a construct that is rarely measured in large-scale dietary surveys. Third, we made use of a large nationally representative dataset with a wealth of social, demographic, psychosocial and nutritional information. Fourth, we took into account the complexity of the sampling design by using survey-related commands and provided nationally representative population estimates. Finally, we used HEI and aMED to measure each individual’s overall diet quality. Despite these strengths, our study had its limitations. First, it is based on cross-sectional data, which does not allow for ascertainment of temporality. However, SES position among adults is often determined in early adulthood and unlikely affected by people's present dietary intake. In addition, attitudinal variables are often antecedents to dietary behaviors and intakes. Second, dietary intake assessment was conducted using average values from only two 24-h recalls, an approach that may be insufficient to reflect individual usual intake over a long period of time. We suggest that the observed relations would be stronger if measurement errors could be smaller. Third, the dataset is over 10 years of age. However, there are to date, no other surveys able to provide all the necessary data to answer our questions while conferring national estimates. We do not expect large changes in the relationships between our study variables over the past decade. Finally, our analysis with interaction terms was conducted both for linear and logistic models. However, we did not intend to compare results between these two models in terms of effect modification since one assesses superadditivity while the other assessed supermultiplicativity. Our findings underscore the complexity of SES and nutrition knowledge and beliefs' associations with dietary intake. Given the higher cost incurred by healthier diets compared to unhealthy eating patterns, as demonstrated by previous research ( Drewnowski, 2004; Drewnowski and Darmon, 2005a,b ), the positive association of SES with dietary quality is theoretically plausible. Other possible mechanisms may include unmeasured factors such as the ability to defer gratification and greater social capital which may be important confounders affecting both SES and diet quality. However, the alternate mechanisms and their contribution to this association are still unproven and require further research. Our finding regarding interactions with nutrition knowledge/beliefs and SES indicate that SES factors may have an influence on dietary choice only for those who have the desirable knowledge and beliefs about nutrition. In particular, an overall better dietary quality is achieved with increased SES among subjects with healthier nutrition knowledge beliefs. Hence, for overall diet quality to reach desired levels in the population, SES interventions must be coupled with health education programs targeting all segments of the US population. Acknowledgments The study was supported in part by the US Department of Agriculture (2005-35215-15372), the NIDDK/NIH (R01 DK63383), and the Johns Hopkins Center for a Livable Future. Appendix A 1. Healthy Eating Index (HEI) Components Criteria 1 Score < 50 years ≥ 50 years Grains 9.1 servings/day 7.4 servings/day 10; 1 point less for each 10% less than intake required for full score Range: 0–10 Vegetables 4.2 servings/day 3.5 servings/day Same as above Fruit 3.2 servings/day 2.5 servings/day Same as above Milk 2.0 servings/day 2.0 servings/day Same as above Meat 2.4 servings/day 2.2 servings/day Same as above Total fat ≤ 30% energy ≤ 30% energy 10 31–44% of energy 31–44% of energy 5 ≥ 45% of energy ≥ 45% of energy 0 Saturated fat ≤ 10% of energy ≤ 10% of energy 10 11–14% of energy 11–14% of energy 5 ≥ 15% of energy ≥ 15% of nergy 0 Cholesterol < 300 mg < 300 mg 10 301–449 mg 301–449 mg 5 ≥ 450 mg ≥ 450 mg 0 Sodium ≤ 2,400 mg ≤ 2,400 mg 10; 1 point less for each 10% less intake required for full score Variety Top 10% intake of sum of unique foods Top 10% intake of sum of unique foods Same as above 1 Based on 2200 kcal for the < 50 years category and 1900 kcal for the ≥ 51 years category. Source: Fung et al. (2005) and McCullough et al. (2000) . 2. Alternate Mediterranean Diet Score (aMED) 1 Food groups Criteria for 1 point Vegetables Greater than median intake (servings/day) Legumes Greater than median intake (servings/day) Fruit Greater than median intake (servings/day) Nuts Greater than median intake (servings/day) Whole grains Greater than median intake (servings/day) Red and processed meat Less than median intake (servings/day) Fish Greater than median intake (servings/day) Ratio of monounsaturated to saturated fatty acids Greater than median intake (servings/day) Ethanol 5–25 g/day Sources: Fung et al. (2005) , Trichopoulou et al. (2003) , and Trichopoulou et al. (1995) . 1 Score may range between 0 and 10. Appendix B (1.1) Logit ( Y i j = 1 ) = β 0 j + ∑ d = 1 2 β 1 k d SES i k d + ∑ l = 1 2 β 2 1 Z 1 (1.2) DQI i j = β 0 j + ∑ d = 1 2 β 1 k d SES i k d + ∑ l = 1 m β 2 l Z l + ε i j k (2.1) Logit ( Y i j = 1 ) = β 0 j + ∑ d = 1 2 β 1 k d SES i k d + ∑ l = 1 m β 2 l Z l + ∑ n = 1 2 β 3 n NKB i n + ∑ p = 1 4 y p SES i k d × NKB i n (2.2) DQI i j = β 0 j + ∑ d = 1 2 β 1 k d SES i k d + ∑ l = 1 m β 2 1 Z l + ∑ n = 1 2 β 3 n NKB i n + ∑ p = 1 4 γ p SES i k d × NKB i n + ε i j k Notations : Y (1 or 0) is the binary outcome for V and F consumption variables; DQI are the diet quality indices (HEI and aMED); Subscript j is for each outcome variable; subscript k is for each SES variables (EDU vs. PIR); subscript d is for each dummy within each SES variable; subscript l is for confounding factors; subscript m is the total number of confounding variables entered into the model; subscript n is for NKB dummies; subscript p is for interaction term; subscript i is for individual subjects. References Appel, 2000 L.J. Appel The role of diet in the prevention and treatment of hypertension Curr. Atheroscler. Rep. 2 6 2000 521 528 Ball et al., 2006 K. Ball D. Crawford G. Mishra Socio-economic inequalities in women's fruit and vegetable intakes: a multilevel study of individual, social and environmental mediators Public Health Nutr. 9 5 2006 623 630 Billson et al., 1999 H. Billson J.A. Pryer R. Nichols Variation in fruit and vegetable consumption among adults in Britain. An analysis from the dietary and nutritional survey of British adults Eur. J. Clin. Nutr. 53 12 1999 946 952 Bodnar and Siega-Riz, 2002 L.M. Bodnar A.M. Siega-Riz A Diet Quality Index for Pregnancy detects variation in diet and differences by sociodemographic factors Public Health Nutr. 5 6 2002 801 809 Cade et al., 1999 J. Cade H. Upmeier C. Calvert D. Greenwood Costs of a healthy diet: analysis from the UK Women's Cohort Study Public Health Nutr. 2 4 1999 505 512 Ding and Mozaffarian, 2006 E.L. Ding D. Mozaffarian Optimal dietary habits for the prevention of stroke Semin. Neurol. 26 1 2006 11 23 Drewnowski, 2004 A. Drewnowski Obesity and the food environment: dietary energy density and diet costs Am. J. Prev. Med. 27 3 Suppl 2004 154 162 Drewnowski and Darmon, 2005a A. Drewnowski N. Darmon The economics of obesity: dietary energy density and energy cost Am. J. Clin. Nutr. 82 1 Suppl 2005 265S 273S Drewnowski and Darmon, 2005b A. Drewnowski N. Darmon Food choices and diet costs: an economic analysis J. Nutr. 135 4 2005 900 904 Ford et al., 2005 E.S. Ford A.H. Mokdad S. Liu Healthy Eating Index and C-reactive protein concentration: findings from the National Health and Nutrition Examination Survey III, 1988–1994 Eur. J. Clin. Nutr. 59 2 2005 278 283 Forshee and Storey, 2006 R.A. Forshee M.L. Storey Demographics, not beverage consumption, is associated with diet quality Int. J. Food Sci. Nutr. 57 7 2006 494 511 Friel et al., 2003 S. Friel C.C. Kelleher G. Nolan J. Harrington Social diversity of Irish adults nutritional intake Eur. J. Clin. Nutr. 57 7 2003 865 875 Fung et al., 2005 T.T. Fung M.L. McCullough P.K. Newby Diet-quality scores and plasma concentrations of markers of inflammation and endothelial dysfunction Am. J. Clin. Nutr. 82 1 2005 163 173 Gittelsohn et al., 2006 J. Gittelsohn J.A. Anliker S. Sharma A.E. Vastine B. Caballero B. Ethelbah Psychosocial determinants of food purchasing and preparation in American Indian households J. Nutr. Educ. Behav. 38 3 2006 163 168 Groth et al., 2001 M.V. Groth S. Fagt L. Brondsted Social determinants of dietary habits in Denmark Eur. J. Clin. Nutr. 55 11 2001 959 966 Guo et al., 2004 X. Guo B.A. Warden S. Paeratakul G.A. Bray Healthy Eating Index and obesity Eur. J. Clin. Nutr. 58 12 2004 1580 1586 Havas et al., 1998 S. Havas K. Treiman P. Langenberg Factors associated with fruit and vegetable consumption among women participating in WIC J. Am. Diet. Assoc. 98 10 1998 1141 1148 Hulshof et al., 2003 K.F. Hulshof J.H. Brussaard A.G. Kruizinga J. Telman M.R. Lowik Socio-economic status, dietary intake and 10 y trends: the Dutch National Food Consumption Survey Eur. J. Clin. Nutr. 57 1 2003 128 137 Irala-Estevez et al., 2000 J.D. Irala-Estevez M. Groth L. Johansson U. Oltersdorf R. Prattala M.A. Martinez-Gonzalez A systematic review of socio-economic differences in food habits in Europe: consumption of fruit and vegetables Eur. J. Clin. Nutr. 54 9 2000 706 714 James et al., 1997 W.P. James M. Nelson A. Ralph S. Leather Socioeconomic determinants of health. The contribution of nutrition to inequalities in health BMJ 314 7093 1997 1545 1549 Kant et al., 2000 A.K. Kant A. Schatzkin B.I. Graubard C. Schairer A prospective study of diet quality and mortality in women JAMA 283 16 2000 2109 2115 Kennedy et al., 2001 E.T. Kennedy S.A. Bowman J.T. Spence M. Freedman J. King Popular diets: correlation to health, nutrition, and obesity J. Am. Diet. Assoc. 101 4 2001 411 420 Krebs-Smith et al., 1995 S.M. Krebs-Smith J. Heimendinger B.H. Patterson A.F. Subar R. Kessler E. Pivonka Psychosocial factors associated with fruit and vegetable consumption Am. J. Health Promot. 10 2 1995 98 104 Mai et al., 2005 V. Mai A.K. Kant A. Flood J.V. Lacey Jr. C. Schairer A. Schatzkin Diet quality and subsequent cancer incidence and mortality in a prospective cohort of women Int. J. Epidemiol. 34 1 2005 54 60 McCullough et al., 2000 M.L. McCullough D. Feskanich E.B. Rimm Adherence to the Dietary Guidelines for Americans and risk of major chronic disease in men Am. J. Clin. Nutr. 72 5 2000 1223 1231 Mishra et al., 2005 G. Mishra K. Ball A. Patterson W. Brown A. Hodge A. Dobson Socio-demographic inequalities in the diets of mid-aged Australian women Eur. J. Clin. Nutr. 59 2 2005 185 195 Moser et al., 2005 R.P. Moser V. Green D. Weber C. Doyle Psychosocial correlates of fruit and vegetable consumption among African American men J. Nutr. Educ. Behav. 37 6 2005 306 314 Ogden et al., 2006 C.L. Ogden M.D. Carroll L.R. Curtin M.A. McDowell C.J. Tabak K.M. Flegal Prevalence of overweight and obesity in the United States, 1999–2004 JAMA 295 13 2006 1549 1555 Pollard et al., 2001 J. Pollard D. Greenwood S. Kirk J. Cade Lifestyle factors affecting fruit and vegetable consumption in the UK Women's Cohort Study Appetite 37 1 2001 71 79 Selvin, 2004 S. Selvin Statistical Analysis of Epidemiologic Data 3rd ed. 2004 Oxford University Press Sharma, 1996 S. Sharma Applied Multivariate Techniques 1996 Wiley USA Siega-Riz et al., 2004 A.M. Siega-Riz S. Kranz D. Blanchette P.S. Haines D.K. Guilkey B.M. Popkin The effect of participation in the WIC program on preschoolers' diets J. Pediatr. 144 2 2004 229 234 Srinath Reddy and Katan, 2004 K. Srinath Reddy M.B. Katan Diet, nutrition and the prevention of hypertension and cardiovascular diseases Public Health Nutr. 7 1A 2004 167 186 STATA, 2005 STATA Statistics/Data Analysis: Release 9(Version 9.0) 2005 Stata Corporation Texas Trichopoulou et al., 2003 A. Trichopoulou T. Costacou C. Bamia D. Trichopoulos Adherence to a Mediterranean diet and survival in a Greek population N. Engl. J. Med. 348 26 2003 2599 2608 Trichopoulou et al., 1995 A. Trichopoulou A. Kouris-Blazos M.L. Wahlqvist Diet and overall survival in elderly people BMJ 311 7018 1995 1457 1460 Turrell and Kavanagh, 2006 G. Turrell A.M. Kavanagh Socio-economic pathways to diet: modelling the association between socio-economic position and food purchasing behaviour Public Health Nutr. 9 3 2006 375 383 US Department of Agriculture ARS, 1994–96 US Department of Agriculture ARS, F. S. R. G. (1994–96). Data (CD-ROM) and documentation for the 1994–96 Continuing Survey of Food Intakes by Individuals (CSFII) — Diet and Health Knowledge Survey. No. PB98-500457. US Department of Agriculture US Department of Agriculture, F. a. N. S. Food stamps program: Applicants and Recipients http://www.fns.usda.gov/fsp/applicant_recipients/default.htm . Retrieved Nov. 22nd, 2006. US Department of Health and Human Services, and Agriculture, 2005 US Department of Health and Human Services, and Agriculture, U. D. o. (2005). Dietary Guidelines for Americans 2005. Washington, DC: USDHHS and USDA. Wang and Beydoun, 2007 Y. Wang M.A. Beydoun The obesity epidemic in the United States—gender, age, socioeconomic, racial/ethnic, and geographic characteristics: a systematic review and meta-regression analysis Epidemiol. Rev. 29 2007 6 28