Family Income and Education Were Related with 30-Year Time Trends in Dietary and Meal Behaviors of American Children and Adolescents^,,

Abstract

Recent survey data reveal the persistence of long-acknowledged socioeconomic status (SES) differentials in the prevalence of obesity in U.S. children and adolescents. We examined 30-y changes in the association of dietary and meal behaviors with family income and education to understand the possible contribution of these trends to SES trends in obesity rates in 2- to 19-y-old Americans. We used dietary and SES data for 2- to 19-y olds from the NHANES 1971–1974 to 2003–2008 (n = 39,822). The secular changes in the independent association of family income and education with 24-h dietary behaviors [energy intake (kcal), amount of foods and beverages (g), percent energy from all beverages and from nutritive beverages, and energy density of foods] and 24-h meal behaviors [number of eating occasions, energy from snack episodes (%), and mention of breakfast] were examined using multivariable regression methods. The secular increase in energy intake and food and beverage amount was significant in the lowest family SES categories. The positive association of family income and education with intakes of energy, food amounts, and beverage energy, noted in 1971–1974 or 1976–1980, was not observed in later surveys. There was an age gradient in changes in most diet and SES associations over time, with largest adverse changes in 12- to 19-y olds. Higher education was associated with lower energy from snack episodes, breakfast skipping, and energy density of foods and these associations did not change over time. Overall, these results suggest both income and education differentials in secular increases in food amounts and energy intakes.

Introduction

A recent analysis (1) of 20-y trends in prevalence of obesity in U.S. children and adolescents showed the persistence of long-recognized socioeconomic differentials (2–5). Apart from the immediate social and health consequences of higher body weight in affected children and adolescents, childhood obesity is a strong correlate of adult body weight, which in turn may contribute to increased risk of several chronic diseases and set the stage for lifelong differentials in achievement and income (6, 7).

Obesity reflects a state of positive energy balance resulting from an interaction of biology and environment. Thus, it is reasonable to expect that socioeconomic differentials in dietary behaviors may contribute to socioeconomic differentials in body weight. This hypothesis is supported by a small body of published evidence (8–12). An extension of this hypothesis is the expectation that socioeconomic differentials in diet may track the trajectories observed for body weight. An additional consideration is that the commonly used measures of family socioeconomic status (SES)⁶, family income, and education of family head are known to be correlated but may impact dietary behaviors in several unique and related ways. For instance, low income may be related to access to foods due to a resource constraint and possible residential segregation in adverse food environments, but education may be linked to acquisition, understanding, and implementation of knowledge about desirable dietary behaviors. Arguably, interventions for addressing income or education differentials may differ, yet we found little published evidence of the attempt to understand the independent associations of these measures of SES with body weight or dietary attributes. Moreover, most published studies focus on adolescents; there is relatively little information on the association of SES and dietary behaviors of younger age groups. Lastly, we found few published studies that provide a concurrent assessment of multiple dietary behaviors to allow an understanding of how changes in one behavior relate to other reported dietary behaviors. Therefore, although several published studies have added to our understanding of changes in dietary behaviors of U.S. children (13–40), to our knowledge, few studies have systematically examined time trends in socioeconomic differences in dietary and meal behaviors of U.S. children and adolescents (2, 41, 42).

It is speculated that adverse dietary and meal behaviors, increasing portion sizes, food form (beverages relative to solid foods), energy density, snacking, and breakfast skipping, may promote positive energy balance and thus contribute to increasing obesity in U.S. children and adolescents (43, 44). Therefore, to provide a comprehensive picture of these dietary and meal behaviors, we used nationally representative data to examine 30-y trends in the association of family income and education with dietary and meal behaviors of U.S. children and adolescents.

Methods

We used public domain data from the NHANES conducted by the National Center for Health Statistics (NCHS) from 1971–1974, 1976–1980, 1988–1994, 1999–2000, 2001–2002, 2003–2004, 2005–2006, and 2007–2008 for this study (45). This study was approved by the Queens College institutional review board for protection of human participants with an exempt review. Each NHANES included a nationally representative, stratified, multistage, probability sample of the U.S. population. The survey procedures included a home interview and an examination in the mobile examination center (MEC). The examination component included anthropometry and a dietary interview (45). The unweighted response rate for the 2- to 19-y-old MEC examined sample in each survey was >75% (46–53).

SES information in each survey

We used family poverty income ratio (PIR) and the education level of the household head (education) as indicators of family SES. The PIR is based on a comparison of family income with the poverty threshold determined by the U.S. Bureau of Census. The PIR is specific to each survey cycle and is adjusted for inflation (46–53). Ratios of <1 are considered below the poverty threshold. We operationalized the PIR as <130, 130–349, and ≥350% of the poverty threshold. The lowest category (<130%) is used as a criterion for eligibility for the federal Supplemental Nutrition Assistance Program. The years of formal education information for the household head was categorized as <12 y, 12 y, some college, and completed college. These categories for PIR and education are similar to those used by the NCHS for its reporting of the NHANES SES weight differentials (1).

Race/ethnicity information in each survey

Because race/ethnicity group membership in the US is correlated with family income and level of education (54), race/ethnicity must be accounted in any evaluation of SES. In the NHANES I and II, race was determined by interviewer observation and was categorized as white, black, and other (46, 47). The white category in both of these surveys included Mexicans and other Hispanics. In the NHANES III and subsequent surveys (48–53), respondents self-reported race/ethnicity; for public domain release, the NCHS provides variables that combine race and ethnicity by adjudicating the available information. For all surveys since 1988–1994, the public release data thus provide non-Hispanic white, non-Hispanic black, Mexican American, other Hispanic, and other race/ethnicity categories. To create roughly comparable race/ethnicity categories across surveys, we identified NHANES I and II respondents reporting Mexican or Hispanic ancestry to create non-Hispanic white and non-Hispanic black race/ethnicity groups. Due to small numbers of Mexican Americans and other Hispanics in the NHANES I and II, these categories could not be created for these surveys. Thus, the 3 categories for all surveys were non-Hispanic white, non-Hispanic black, and all other race/ethnicity groups.

Analytic sample

All children and adolescents 2–19 y of age with a reliable dietary recall (as determined by the NCHS) were eligible for inclusion in this study (NHANES I = 6825; NHANES II = 7270; NHANES III = 10,433; NHANES 1999–2002 = 8134; NHANES 2003–2008 = 10,998). From the eligible cohort, we excluded respondents missing information on family income, education level of household head, pregnant and lactating females, and those reporting zero calories in a recall otherwise considered reliable by the NCHS (NHANES I = 382; NHANES II = 331; NHANES III = 1023; NHANES 1999–2002 = 1085; NHANES 2003–2008 = 1017). The 2007–2008 survey years include up to seventeen 12- to 19-y olds whose pregnancy and lactation status information is not available in the public domain data. The final analytic sample included 39,822 children and adolescents: (NHANES I = 6443; NHANES II = 6939; NHANES III = 9410; NHANES 1999–2002 = 7049; NHANES 2003–2008 = 9981).

Dietary information

All surveys collected dietary information using an interviewer-administered 24-h dietary recall. In the NHANES I (1971–1974) and II (1976–1980), the recalls were obtained using paper and pencil methods (45, 46). Beginning with the NHANES III (1988–1994), dietary information was collected using a computer-assisted 24-h recall (47–53). The 1988–1994 survey used a 3-step dietary recall, 1999–2001 surveys used a 4-step recall, and 2003–2008 surveys used a 5-step recall with USDA’s Automated Multiple Pass Method (47–53). The NHANES I and II surveys used USDA handbook 8 and several other sources to code foods and beverages for nutrient content; the NHANES III and 1999–2000 NHANES used the USDA’s Survey Nutrient Data Base (46–49); all subsequent surveys used the Food and Nutrient Database for Dietary Studies (versions 1, 2, 3, and 4.1 for survey years 2001–2002, 2003–2004, 2005–2006, and 2007–2008, respectively) (55). For survey years 2003–2008, a second recall, obtained via telephone, is available in the public domain; our analyses, however, use the first recall for all surveys for consistency. In all surveys, parents or other caregivers provided the recall for 2- to 5-y-old children, recalls of 6- to 11-y olds were self-reported with assistance from parents or other caregivers, and 12- to 19-y olds self-reported their intake. We used the 24-h dietary recall data to create dietary and meal behaviors of interest as described below.

Dietary behaviors

24-h dietary energy and amount of foods and beverages.

The 24-h energy intake (kcal) and amount of all foods and beverages (g) were available in the public domain dataset. Surveys conducted from 2005 to 2008 include the amount of plain tap or bottled water as part of the 24-h recall; this amount was excluded to compute the total amounts for these surveys. This allowed the amount variable to be comparable across surveys.

Percent of 24-h energy intake from all beverages and percent of energy from nutritive beverages.

To determine beverage intake, we categorized all reported foods in each survey into beverage or nonbeverage (food) categories using methods we previously described (56). All types of milk, infant formula, all types of juice, juice drinks, energy/sports drinks, carbonated or noncarbonated beverages, shakes and smoothies, coffee, tea, hot chocolate and cocoa, meal replacement drinks, and alcoholic beverages were considered beverages. From this list, all types of milk, infant formula, 100% juice, meal replacement drinks, and milk shakes or fruit smoothies were considered nutritive beverages. Energy contribution of beverages was considered as consumed. In the NHANES III and all subsequent surveys, coding of reported beverages identified beverage combinations (48–53). For example, sugar and powdered creamer added to coffee or tea or fruit or chocolate flavors added to milk were coded as beverage combinations; we considered all components of such combinations as beverages. Plain tap or bottled water reported as part of the 24-h recall in the NHANES 2005–2006 and 2007–2008 was not considered a beverage. (Plain water information was not collected in the NHANES I and II; in the NHANES III, 1999–2000, 2001–2002, and 2003–2004, this information was collected after the recall.) Energy from all beverages and nutritive beverages was operationalized as a percentage of the 24-h energy intake.

Energy density of foods (only) reported in the 24-h recall.

The dietary energy density (kcal/g) of foods reported in the 24-h recall in each survey was determined using methods we previously reported [energy from all foods was divided by weight (g) of all foods] (57). The food items included in beverage combinations (mentioned above) were considered part of beverages and were not included in this computation. Our decision to report the energy density of foods only reflects current thinking about the relative importance of energy density of foods compared with beverages (58) in regulating food intake.

Meal behaviors

Number of eating occasions in the 24-h recall.

In each survey, respondents were asked to report the clock time for each recalled food, beverage, or their combinations. Using methods previously reported (59, 60), we considered each reported clock time of eating as an eating occasion regardless of the amounts consumed. All foods and beverages reported at one clock time were considered as part of the same eating occasion. Eating occasions where the only reported item was plain tap or bottled water (in NHANES 2005–2006 and 2007–2008) were not considered as eating occasions.

Mention of breakfast in the 24-h recall.

In each recall, respondents were asked to name each eating episode. In the NHANES I and II, the eating episode name options were AM, noon, PM, between meal, and all day (extended consumption); in later surveys, the list of meal names was expanded to include commonly used names and their equivalents in Spanish. Those reporting an episode named AM or breakfast/desayuno in the recall were considered to have consumed breakfast (59, 60).

Percentage of 24-h energy from snack episodes mentioned in the 24-h recall.

Snack episode reporting was determined from respondent mention of eating events named as between meal (NHANES I and II), snack, drink, or extended consumption in the recall and operationalized as percentage of 24-h energy from thus-named episodes.

Analytical methods

We followed the NHANES analytic guidelines that recommend combining data from the recent 2-y cycles (1999 onwards) (61). We combined 1999–2000 with 2001–2002, and 2003–2004 as well as 2005–2006 with 2007–2008. Thus, the trend analyses included 1971–1974, 1976–1980, 1988–1994, 1999–2002, and 2003–2008 survey waves. Our analyses were stratified by the age categories of 2–5, 6–11, and 12–19 y. These age categories are used for reporting NHANES body weight data by the NCHS (1) and also reflect the age differences in respondents for the dietary recall.

Because the characteristics of survey respondents changed over the 30-y span examined, we used multivariable linear and logistic regression analyses for continuous and dichotomous dependent variables, respectively, to examine the independent association of each dietary and meal behavior with PIR or education across surveys. These analyses included each dietary/meal behavior as a continuous or dichotomous outcome and the covariates potentially related to family PIR or education and diet as independent variables. All regression models included gender, age, PIR, education, race/ethnicity, day of week of dietary intake, season of MEC exam, household size, BMI for gender-age percentile (62), and dummy variables indicating survey(s) as independent variables. The inclusion of BMI as a covariate in regression models allows some adjustment for possible differences in dietary reporting in association with BMI. In this analysis, the combined surveys were included together as independent samples. We tested for homogeneity among the PIR or education categories in time trends in dietary and meal behaviors by testing the interaction of PIR or education (as scored trend variables) with survey year using models in which midpoint of the survey time interval was a scored trend variable. When significant PIR or education by survey interactions were not present, we tested the main effects of PIR or education level using the above regression models without the PIR or education by survey interaction term but adjusted for all other covariates (63, 64). We present covariate-adjusted means of various outcomes (also called predicted margins) ± SE for each SES category and survey in the results section. These estimates were obtained from regression models where survey and SES variables were categorical. (The results of hypothesis testing in this paper are from regression models where survey and SES exposures were scored trend variables.) The reported P values for tests of significance are based on Wald’s F statistic and are not adjusted for multiple comparisons; we provide the actual P values as a guide to identify dietary/SES relationships of potential interest. A 2-sided P value < 0.05 was considered significant. Similar methods were used when BMI for gender-age percentile was the dependent variable and all tables and figures also include associations with prevalence of high BMI for context.

We followed NCHS recommendations for sample weights applicable to analyses from combined surveys (61). These analyses were carried out using SAS version 9.2 (RTI) and SAS-callable SUDAAN (65) to adjust for the complex sample design of the NHANESs (64).

Results

The percentage of respondents with a PIR of ≥350% and education of at least some college increased from 1971–1974 to 2003–2008 (Table 1). The percentage of non-Hispanic whites decreased, while the percentage of respondents with “other” race/ethnicity increased across survey years.

TABLE 1.

Characteristics of Americans 2–19 y of age by survey (NHANES 1971–74 to 2003–2008)¹

	n	All surveys (n = 39,822)	NHANES I971–1974 (n = 6443)	NHANES 1976–1980 (n = 6939)	NHANES 1988–1994 (n = 9410)	NHANES 1999–2002 (n = 7049)	NHANES 2003–2008 (n = 9981)
	%
Males	20,084	51.1 ± 0.4	50.7 ± 0.8	50.9 ± 0.7	50.9 ± 0.9	50.7 ± 1.04	52.0 ± 0.9
Age group, y
2–5	12,403	20.9 ± 0.3	20.0 ± 0.6	18.6 ± 0.5	22.8 ± 0.8	21.4 ± 0.8	21.7 ± 0.7
6–11	11,069	33.7 ± 0.4	33.7 ± 0.7	32.4 ± 0.6	34.4 ± 0.9	34.7 ± 1.3	33.1 ± 0.8
12–19	16,350	45.4 ± 0.5	46.3 ± 0.8	49.0 ± 0.8	42.8 ± 1.2	43.9 ± 1.0	45.2 ± 1.2
Race/ethnicity
Non-Hispanic white	17,149	69.2 ± 0.8	79.9 ± 1.4	76.7 ± 1.8	66.8 ± 1.6	61.5 ± 2.0	61.7 ± 2.3
Non-Hispanic black	10,843	14.1 ± 0.6	13.2 ± 1.1	14.0 ± 1.5	15.2 ± 1.1	13.7 ± 1.7	14.6 ± 1.4
Others	11,830	16.7 ± 0.7	6.9 ± 1.5	9.3 ± 1.1	18.0 ± 1.3	24.8 ± 2.3	23.7 ± 1.7
Family PIR, %
<130	16,768	30.9 ± 0.6	27.7 ± 1.3	28.7 ± 1.3	31.6 ± 1.4	34.5 ± 1.4	32.0 ± 1.5
130–349	17,089	47.2 ± 0.6	56.3 ± 1.1	59.8 ± 1.1	47.4 ± 1.5	37.5 ± 1.3	36.2 ± 1.3
≥350	5965	21.9 ± 0.6	16.0 ± 1.0	11.5 ± 0.8	21.0 ± 1.3	28.0 ± 1.6	31.8 ± 1.9
Education of household head
<12 y	14,821	27.40 ± 0.6	37.3 ± 1.4	33.0 ± 1.1	23.8 ± 1.1	24.0 ± 1.3	19.4 ± 1.1
12 y	11,796	31.18 ± 0.5	37.3 ± 0.9	33.0 ± 0.9	34.8 ± 1.2	25.9 ± 1.2	25.7 ± 1.2
Some college	7889	22.21 ± 0.5	13.0 ± 0.7	18.5 ± 1.1	20.2 ± 1.1	25.1 ± 1.3	33.3 ± 1.0
Completed college	5316	19.21 ± 0.5	12.4 ± 0.8	15.5 ± 0.9	21.2 ± 1.2	25.0 ± 1.7	21.6 ± 1.4
Weekday of recalled intake
Monday-Thursday	20,032	61.5 ± 0.6	71.9 ± 0.8	73.2 ± 1.0	49.5 ± 1.7	56.6 ± 1.4	56.2 ± 1.3
Month of MEC exam
November-April	18,793	45.7 ± 1.8	44.5 ± 4.2	48.6 ± 3.8	58.9 ± 4.0	37.9 ± 4.9	39.9 ± 3.3
May-October	21,029	54.3 ± 1.8	55.5 ± 4.3	51.3 ± 3.8	41.1 ± 4.0	62.1 ± 4.9	60.1 ± 3.3
Household size
≤2	2096	5.5 ± 0.2	4.4 ± 0.4	6.2 ± 0.5	5.5 ± 0.5	5.4 ± 0.6	6.1 ± 0.5
3–4	16,868	45.6 ± 0.5	32.6 ± 0.9	41.6 ± 1.0	50.3 ± 1.1	50.7 ± 1.3	52.6 ± 1.1
≥5	20,858	48.9 ± 0.5	63.3 ± 1.0	52.3 ± 1.1	44.1 ± 1.2	43.8 ± 1.6	41.3 ± 1.1
BMI percentile for sex and age
<85th	29,268	76.0 ± 0.4	84.6 ± 0.7	84.8 ± 0.5	75.9 ± 1.0	68.3 ± 1.1	67.4 ± 1.3
85th to <95th	5061	12.4 ± 0.3	10.2 ± 0.6	9.4 ± 0.4	12.8 ± 0.8	14.0 ± 0.6	15.4 ± 0.7
≥95th	4865	10.3 ± 0.3	5.3 ± 0.3	5.5 ± 0.4	9.9 ± 0.5	14.7 ± 0.7	15.7 ± 0.8
Unknown	626	1.3 ± 0.1	0	0.3 ± 0.04	1.4 ± 0.2	3.1 ± 0.3	1.5 ± 0.2

¹Values are weighted percents ± SEs. MEC, mobile examination center; PIR, poverty income ratio.

Given our primary study objective of trying to understand if the association of SES and diet changed over time, the following section of results is organized so that the focus of the narrative is significant interactions of survey year with education (Table 2) and PIR (Table 3).

TABLE 2.

Dietary and meal behaviors with significant interaction of survey year with family heads' education in children and adolescents 2–19 y of age (NHANES 1971–1974 to 2003–2008)¹

Family head education level (y)	NHANES 1971–1974	NHANES 1976–1980	NHANES 1988–1994	NHANES 1999–2002	NHANES 2003–2008	Survey effect P-trend
BMI2 ≥95th percentile for age and gender, %
12–19 y	Survey by education,3 P-interaction = 0.01
<12 y	8.9 ± 0.9	6.9 ± 0.8	14.2 ± 1.9	22.4 ± 2.0	19.8 ± 2.1	<0.0001 ↑4
12 y	5.0 ± 1.1	4.8 ± 0.8	12.4 ± 1.7	18.7 ± 1.8	18.1 ± 1.5	<0.0001 ↑
Some college	2.8 ± 0.6	2.8 ± 0.9	7.1 ± 1.7	13.0 ± 1.6	17.9 ± 1.7	<0.0001 ↑
Completed college	3.7 ± 1.5	2.6 ± 0.9	5.0 ± 2.1	11.3 ± 2.2	15.0 ± 3.6	<0.0001 ↑
Education effect, P-trend	0.0002 ↓	0.0007 ↓	0.002 ↓	<0.0001 ↓	0.2
24-h energy intake, kcal/d
2–5 y	Survey by education, P-interaction = 0.002
<12 y	1530 ± 29	1520 ± 26	1520 ± 31	1640 ± 56	1600 ± 40	0.03 ↑
12 y	1660 ± 26	1540 ± 19	1590 ± 26	1630 ± 68	1640 ± 35	0.9
Some college	1670 ± 39	1510 ± 34	1540 ± 29	1660 ± 34	1620 ± 43	0.3
Completed college	1630 ± 34	1550 ± 40	1530 ± 36	1540 ± 34	1490 ± 31	0.01 ↓
Education effect, P-trend	0.005 ↑	0.8	0.8	0.1	0.1
12–19 y	Survey by education, P-interaction = 0.01
<12 y	2250 ± 33	2190 ± 45	2380 ± 48	2280 ± 50	2280 ± 48	0.1
12 y	2300 ± 39	2240 ± 60	2480 ± 61	2420 ± 38	2240 ± 41	0.3
Some college	2390 ± 78	2290 ± 69	2400 ± 67	2260 ± 61	2330 ± 41	0.5
Completed college	2320 ± 67	2340 ± 59	2400 ± 63	2320 ± 64	2240 ± 58	0.3
Education effect, P-trend	0.1	0.03 ↑	0.9	0.7	1.0
Amount of all foods and beverages in the 24-h recall, g/d
2–5 y	Survey by education, P-interaction = 0.0002
<12 y	1310 ± 23	1280 ± 21	1344 ± 27	1470 ± 54	1440 ± 41	<0.0001 ↑
12 y	1460 ± 23	1350 ± 21	1450 ± 25	1500 ± 69	1450 ± 32	0.3
Some college	1520 ± 37	1320 ± 29	1380 ± 31	1550 ± 44	1510 ± 48	0.02 ↑
Completed college	1470 ± 33	1370 ± 34	1430 ± 25	1390 ± 36	1370 ± 34	0.1
Education effect, P-trend	<0.0001 ↑	0.1	0.1	0.2	0.4
6–11 y	Survey by education, P-interaction = 0.01
<12 y	1630 ± 29	1580 ± 31	1660 ± 39	1670 ± 84	1610 ± 44	0.6
12 y	1730 ± 31	1680 ± 39	1620 ± 32	1620 ± 53	1660 ± 32	0.04 ↓
Some college	1780 ± 44	1710 ± 44	1600 ± 37	1670 ± 49	1650 ± 38	0.06
Completed college	1810 ± 68	1640 ± 32	1710 ± 49	1660 ± 46	1560 ± 42	0.01 ↓
Education effect, P-trend	0.001 ↑	0.05 ↑	0.4	0.9	0.4
12–19 y	Survey by education, P-interaction = 0.0005
<12 y	1960 ± 40	1900 ± 44	2130 ± 40	2110 ± 51	2080 ± 52	0.0005 ↑
12 y	1960 ± 36	1940 ± 42	2180 ± 65	2220 ± 66	2070 ± 47	0.0004 ↑
Some college	2080 ± 67	2050 ± 65	2120 ± 63	2060 ± 57	2080 ± 55	1.0
Completed college	2040 ± 53	2050 ± 54	2140 ± 58	2080 ± 77	1960 ± 61	0.5
Education effect, P-trend	0.1	0.01 ↑	0.7	0.3	0.2
24-h energy intake from all beverages, %
6–11 y	Survey by education, P-interaction = 0.0004
<12 y	24 ± 0.4	26 ± 0.5	22 ± 0.8	22 ± 1.1	20 ± 0.8	<0.0001 ↓
12 y	25 ± 0.5	24 ± 0.6	21 ± 0.7	22 ± 1.1	20 ± 0.5	<0.0001 ↓
Some college	25 ± 0.7	26 ± 0.7	20 ± 0.8	22 ± 1.0	19 ± 0.7	<0.0001 ↓
Completed college	28 ± 0.8	25 ± 0.4	20 ± 0.9	21 ± 1.3	18 ± 0.8	<0.0001 ↓
Education effect, P-trend	<0.0001 ↑	0.8	0.1	0.3	0.1
12–19 y	Survey by education, P-interaction = < 0.0001
<12 y	24 ± 0.5	24 ± 0.6	24 ± 0.6	26 ± 0.8	23 ± 0.6	0.4
12 y	24 ± 0.4	24 ± 0.7	21 ± 0.6	25 ± 1.0	21 ± 0.5	0.1
Some college	24 ± 0.8	25 ± 0.7	21 ± 0.8	23 ± 0.9	21 ± 0.5	0.0002 ↓
Completed college	24 ± 0.8	24 ± 0.8	22 ± 1.0	21 ± 1.1	19 ± 0.8	<0.0001 ↓
Education effect, P-trend	0.5	0.8	0.03	<0.0001 ↓	0.0001 ↓
24-h energy from nutritive beverages, %
6–11 y	Survey by education, P-interaction = 0.001
<12 y	18.8 ± 0.5	19.3 ± 0.5	14.0 ± 0.5	12.6 ± 1.1	12.1 ± 0.7	<0.0001 ↓
12 y	20.4 ± 0.6	17.8 ± 0.7	13.5 ± 0.8	12.7 ± 1.0	11.9 ± 0.5	<0.0001 ↓
Some college	21.0 ± 0.7	20.1 ± 0.7	12.7 ± 0.9	12.6 ± 0.8	11.4 ± 0.7	<0.0001 ↓
Completed college	23.5 ± 0.8	19.5 ± 0.4	13.1 ± 0.9	11.8 ± 0.7	11.8 ± 0.7	<0.0001 ↓
Education effect, P-trend	<0.0001 ↑	0.1	0.4	0.5	0.6
12–19 y	Survey by education, P-interaction = 0.01
<12 y	14.7 ± 0.4	13.7 ± 0.5	9.3 ± 0.5	9.3 ± 0.5	8.9 ± 0.4	<0.0001 ↓
12 y	16.1 ± 0.4	14.6 ± 0.5	8.1 ± 0.4	9.8 ± 0.7	8.4 ± 0.5	<0.0001 ↓
Some college	17.2 ± 0.9	14.7 ± 0.8	8.6 ± 0.7	9.1 ± 0.7	8.3 ± 0.4	<0.0001 ↓
Completed college	17.3 ± 0.8	15.5 ± 0.8	9.8 ± 0.7	8.7 ± 0.5	8.8 ± 0.6	<0.0001 ↓
Education effect, P-trend	0.0003 ↑	0.04 ↑	0.4	0.3	0.8
Energy density of foods only, kcal/g
6–11 y	Survey by education, P-interaction = 0.02
<12 y	2.09 ± 0.03	2.13 ± 0.02	2.11 ± 0.04	2.30 ± 0.07	2.22 ± 0.04	0.001 ↑
12 y	2.07 ± 0.02	2.13 ± 0.02	2.15 ± 0.03	2.30 ± 0.07	2.33 ± 0.03	<0.0001 ↑
Some college	2.06 ± 0.04	2.13 ± 0.03	2.14 ± 0.06	2.22 ± 0.04	2.23 ± 0.03	0.0006 ↑
Completed college	2.09 ± 0.04	2.12 ± 0.04	2.09 ± 0.05	2.14 ± 0.06	2.13 ± 0.05	0.5
Education effect, P-trend	0.9	0.9	0.6	0.1	0.01 ↓

¹Values are adjusted means or percentages ± SEs from multivariable linear or logistic regression models that included each dietary or BMI variable as a continuous or dichotomous dependent. Separate models were conducted for each age group and included age (continuous), gender, race-ethnicity (non-Hispanic white, non-Hispanic black, all others), survey (NHANES I-1971–1974, NHANES II-1976–1980, NHANES III-1988–1994, NHANES 1999–2002, and NHANES 2003–2008), month of MEC exam (November to April, May to October), weekday of recalled intake (Monday through Thursday, Friday through Sunday), family PIR (<130, 130–185, ≥350%), education of household head (<12 y, 12 y, some college, completed college), household size (≤2, 3–4, ≥5), and BMI-sex-age-percentile (<85th, 85 to <95th, ≥95th), and survey by education interaction. n for models with complete covariate information: 2–5 y (11,955); 6–11 y (11,016); 12–19 y (16,223). PIR, poverty income ratio.

²Models for BMI excluded month of MEC exam, day of recalled intake, and household size.

³Models with interaction terms included survey year as a scored trend variable and education categories as a trend variable.

⁴The direction of the arrow suggests an upward (↑) or a downward (↓) trend.

TABLE 3.

Dietary and meal behaviors with a significant interaction of survey year with family PIR in children and adolescents 2–19 y of age (NHANES 1971–1974 to 2003–2008)¹

PIR (%)	NHANES 1971–1974	NHANES 1976–1980	NHANES 1988–1994	NHANES 1999–2001	NHANES 2003–2008	Survey effect P-trend
24-h energy intake, kcal/d
2–5 y	Survey by PIR,2 P-interaction = 0.002
PIR <130	1600 ± 34	1560 ± 28	1590 ± 30	1650 ± 42	1660 ± 23	0.02 ↑3
PIR 130–349	1610 ± 22	1510 ± 15	1540 ± 24	1610 ± 37	1560 ± 30	0.8
PIR ≥350	1660 ± 36	1540 ± 36	1500 ± 40	1570 ± 45	1540 ± 32	0.1
PIR effect, P-trend	0.3	0.3	0.1	0.2	0.002 ↓
12–19 y	Survey by PIR, P-interaction = 0.0002
PIR <130	2110 ± 58	2060 ± 45	2420 ± 58	2330 ± 49	2280 ± 43	<0.0001 ↑
PIR 130–349	2360 ± 34	2320 ± 50	2360 ± 37	2270 ± 46	2240 ± 47	0.05
PIR ≥350	2380 ± 63	2270 ± 68	2540 ± 84	2370 ± 56	2320 ± 48	0.6
PIR effect, P-trend	0.0006 ↑	0.0006 ↑	0.3	0.7	0.5
Amount of foods and beverages in the 24-h recall, g/d
2–5 y	Survey by PIR, P-interaction = 0.0001
PIR <130	1410 ± 33	1340 ± 29	1430 ± 28	1530 ± 57	1540 ± 25	<0.0001 ↑
PIR 130–349	1430 ± 17	1320 ± 17	1390 ± 20	1480 ± 37	1400 ± 32	0.2
PIR ≥350	1430 ± 32	1350 ± 32	1410 ± 37	1390 ± 54	1380 ± 34	0.5
PIR effect, P-trend	0.5	0.9	0.5	0.1	0.0001 ↓
6–11 y	Survey by PIR, P-interaction = 0.001
PIR <130	1670 ± 38	1630 ± 35	1690 ± 24	1680 ± 59	1710 ± 42	0.3
PIR 130–349	1720 ± 26	1650 ± 23	1630 ± 34	1640 ± 49	1560 ± 31	0.003 ↓
PIR ≥350	1800 ± 44	1690 ± 59	1620 ± 61	1650 ± 44	1600 ± 37	0.004 ↓
PIR effect, P-trend	0.03 ↑	0.4	0.3	0.6	0.01 ↓
12–19 y	Survey by PIR, P-interaction = 0.0002
PIR <130	1860 ± 51	1800 ± 41	2080 ± 72	2140 ± 58	2080 ± 39	<0.0001 ↑
PIR 130–349	2020 ± 33	2020 ± 38	2150 ± 35	2080 ± 59	2000 ± 57	0.6
PIR ≥350	2080 ± 56	2020 ± 66	2230 ± 70	2140 ± 61	2080 ± 43	0.6
PIR effect, P-trend	0.002 ↑	<0.0001 ↑	0.1	1.0	0.8
24-h energy intake from all beverages, %
2–5 y	Survey by PIR, P-interaction = 0.01
PIR <130	27 ± 0.7	26 ± 0.5	24 ± 0.7	26 ± 0.9	26 ± 0.7	0.3
PIR 130–349	28 ± 0.5	27 ± 0.4	24 ± 0.6	26 ± 0.7	25 ± 0.6	<0.0001 ↓
PIR ≥350	28 ± 0.8	28 ± 1.1	26 ± 0.6	25 ± 1.2	24 ± 0.7	<0.0001 ↓
PIR effect, P-trend	0.3	0.1	0.2	0.6	0.1
6–11 y	Survey by PIR, P-interaction = <0.0001
PIR <130	23 ± 0.5	25 ± 0.4	22 ± 0.7	23 ± 1.3	22 ± 0.7	0.01 ↓
PIR 130–349	25 ± 0.4	25 ± 0.4	21 ± 0.6	22 ± 0.9	19 ± 0.7	<0.0001 ↓
PIR ≥350	27 ± 0.8	24 ± 0.7	20 ± 1.0	19 ± 1.0	18 ± 0.7	<0.0001 ↓
PIR effect, P-trend	0.0003 ↑	0.1	0.2	0.01 ↓	0.001 ↓
12–19 y	Survey by PIR, P-interaction = 0.01
PIR <130	23 ± 0.6	24 ± 0.7	21 ± 0.7	24 ± 0.9	22 ± 0.6	0.3
PIR 130–349	24 ± 0.4	24 ± 0.4	22 ± 0.5	25 ± 0.9	21 ± 0.6	0.04 ↓
PIR ≥350	24 ± 0.7	24 ± 1.2	22 ± 0.9	23 ± 0.7	20 ± 0.6	<0.0001 ↓
PIR effect, P-trend	0.2	0.6	0.6	0.3	0.04 ↓
24-h energy intake from nutritive beverages, %
6–11 y	Survey by PIR, P-interaction = <0.0001
PIR <130	18.9 ± 0.6	19.3 ± 0.5	13.9 ± 0.6	13.3 ± 1.2	13.2 ± 0.5	<0.0001 ↓
PIR 130–349	20.7 ± 0.5	19.0 ± 0.4	13.6 ± 0.6	12.5 ± 0.7	11.2 ± 0.5	<0.0001 ↓
PIR ≥350	22.5 ± 0.8	18.5 ± 0.8	12.1 ± 1.0	10.9 ± 0.8	10.6 ± 0.6	<0.0001 ↓
PIR effect, P-trend	0.0002 ↑	0.2	0.1	0.1	0.0004 ↓
12–19 y	Survey by PIR, P-interaction = 0.01
PIR <130	15.3 ± 0.6	13.5 ± 0.6	9.1 ± 0.6	9.1 ± 0.8	9.1 ± 0.3	<0.0001 ↓
PIR 130–349	16.0 ± 0.4	14.5 ± 0.4	8.5 ± 0.4	10.0 ± 0.5	7.8 ± 0.3	<0.0001 ↓
PIR ≥350	16.5 ± 0.8	16.2 ± 1.0	9.0 ± 0.8	8.6 ± 0.5	8.6 ± 0.5	<0.0001 ↓
PIR effect, P-trend	0.1	0.02 ↑	0.8	0.6	0.4
Eating occasions in a 24-h recall, n/d
12–19 y	Survey by PIR, P-interaction = 0.01
PIR <130	4.47 ± 0.1	4.27 ± 0.1	4.30 ± 0.1	4.59 ± 0.1	4.70 ± 0.1	0.0001 ↑
PIR 130–349	4.77 ± 0.1	4.64 ± 0.1	4.43 ± 0.1	4.74 ± 0.1	4.52 ± 0.1	0.04 ↓
PIR ≥350	4.88 ± 0.1	4.60 ± 0.1	4.66 ± 0.1	4.89 ± 0.1	4.79 ± 0.1	0.6
PIR effect, P-trend	0.0002 ↑	0.0002 ↑	0.003 ↑	0.01 ↑	0.2

¹Values are adjusted means or percentages ± SEs from multivariable linear or logistic regression models that included each dietary variable as a continuous or dichotomous dependent. Separate models were conducted for each age group and included age (continuous), gender, race-ethnicity (non-Hispanic white, non-Hispanic black, all others), survey (NHANES I-1971–1974, NHANES II-1976–1980, NHANES III-1988–1994, NHANES 1999–2002, and NHANES 2003–2008), month of MEC exam (November to April, May to October), weekday of recalled intake (Monday through Thursday, Friday through Sunday), family PIR (<130, 130–185, ≥350%), education of household head (<12, 12, some college, completed college), household size (≤2, 3–4, ≥5), and BMI-sex-age-percentile (<85th, 85 to <95th, ≥95th), and survey by education interaction. n for models with complete covariate information: 2–5 y (11,955); 6–11 y (11,016); 12–19 y (16,223). PIR, poverty income ratio.

²Models with interaction terms included survey year as a scored trend variable and PIR categories as a trend variable.

³The direction of the arrow suggests an upward (↑) or downward (↓) trend.

Did the independent association of education of household head with dietary and meal behaviors change over time?

Significant interactions of survey year and education were noted in one or more age groups for prevalence of obesity, energy intake, amount of foods and beverages, percent energy from all beverages and from nutritive beverages, and energy density of foods (Table 2). For all other examined outcomes (number of eating occasions, percent energy from snack episodes, and mention of breakfast), significant interactions were not present in any age group and these results are not shown in Table 2.

2- to 5-y olds.

Both the 24-h energy and amount of reported foods and beverages increased with increasing education in 1971–1974 (P ≤ 0.005). A secular upward trend in energy intake and amount of foods and beverages was noted in the lowest education category (P ≤ 0.03). The reported amount of foods and beverages also increased over time in the some college category (P = 0.02). In the highest education category, the reported energy intake declined over time (P = 0.01).

6- to 11-y olds.

In 1971–1974, higher education was associated with higher reported amount of foods and beverages (P = 0.001) and percent of energy from all or nutritive beverages (P < 0.0001). The secular decline in reported amount of foods and beverages was significant in education categories of 12 y (P = 0.04) and completed college (P = 0.04); however, the percent of energy from all and nutritive beverages declined over time in all education categories (P < 0.0001). In all education categories except completed college, energy density of foods increased over time (P ≤ 0.001); in the 2003–2008 survey cycle, education related inversely with energy density (P = 0.01).

12- to 19-y olds.

Over time, the prevalence of high BMI increased in all categories of education (P < 0.0001); in all surveys except the 2003–2008 survey, education was an inverse correlate of prevalence of high BMI (P ≤ 0.002). Secular increases in the reported amounts of foods and beverage were present only in the 2 lowest education categories (P ≤ 0.0005). In 1976–1980, education level was a positive correlate of reported energy intakes and amounts of foods and beverages (P ≤ 0.03). A secular decline in the percent of energy from all beverages was significant (P < 0.0001) in education categories of some college or higher; the inverse association of percent energy from all beverages and education was significant, beginning with the 1988–1994 survey and all subsequent surveys (P ≤ 0.03). The positive association of percent of energy from nutritive beverages with education, noted in 1971–1974 (P = 0.0003) and 1976–1980 (P = 0.04), was not seen in later surveys; however, a strong secular decline in percent energy from nutritive beverages was noted in all categories of education (P < 0.0001).

Did the independent association of family PIR with dietary and meal behaviors change over time?

Significant interactions of PIR and survey year were noted for energy intake, amount of foods and beverages, percent energy from all beverages and nutritive beverages, and number of eating occasions in one or more of the age groups examined (Table 3).

2- to 5-y olds.

The reported energy intake (P ≤ 0.02) and amount of foods and beverages (P < 0.0001) increased over time in the <130% PIR category only; in 2003–2008, both energy intake and amount of foods and beverages decreased with increasing PIR (P ≤ 0.002). The percent energy from all beverages declined over time in the 2 higher PIR categories (P < 0.0001).

6- to 11-y olds.

The reported amount of foods and beverages decreased over time in PIR categories of ≥130% (P < 0.004); the positive association of PIR and reported amounts was noted in 1971–1974 (P = 0.03); in 2003–2008, the association became inverse (P = 0.01). The percent energy from all beverages or nutritive beverages declined over time in all categories of PIR (P ≤ 0.01) and the positive associations of these outcomes with PIR in 1971–1974 (P ≤ 0.0003) became inverse in later surveys (P ≤ 0.001).

12- to 19-y olds.

The reported 24-h energy intake and amount of foods and beverages increased over time (P < 0.0001) in the lowest PIR category; the positive association (P ≤ 0.002) of these dietary variables with PIR was limited to the 1971–1974 and 1976–1980 surveys. A secular decline in percent energy from all beverages was noted in the 2 higher PIR categories (P ≤ 0.04) and an inverse PIR and beverage energy association was noted only in 2003–2008 (P = 0.04). The percent energy from nutritive beverages declined over time in all PIR categories (P < 0.0001). A secular increase in the number of eating occasions was noted in the <130% PIR category (P < 0.0001); the trend was inverse in the 130–349% category (P = 0.04). In all surveys except the 2003–2008 surveys, PIR was a positive correlate of number of eating occasions (P ≤ 0.01).

Main effects of survey year, PIR, and education for dietary and meal behaviors

The main effects of survey year and family education or PIR were not the study objectives, but we present this information in Supplemental Table 1 and Supplemental Figures 1 and 2 for context. The following narrative of these results of main effects is limited to those variables where interactions (described in earlier sections) were not present.

Energy density of reported foods increased across surveys in 12- to 19-y olds (Supplemental Table 1) and higher education predicted lower energy density in all age groups (Supplemental Fig. 1); however, PIR was unrelated (Supplemental Fig. 2). The number of eating occasions increased across surveys in 2- to 5-y olds and 12- to 19-y olds (Supplemental Table 1); education was unrelated in all age groups (Supplemental Fig. 1), while higher PIR predicted higher number of eating occasions in 6- to 11-y olds (Supplemental Fig. 2). Percentage of 24-h energy intake from snack episodes increased across surveys in 2- to 5-y and 6- to 11-y olds (Supplemental Table 1); higher education predicted lower energy from snack episodes at all ages (Supplemental Fig. 1) and PIR was unrelated (Supplemental Fig. 2). Breakfast reporting declined across survey years (Supplemental Table 1) but increased across education categories at all ages (Supplemental Fig. 1); in 12- to 19-y olds, a higher PIR predicted a higher percent reporting breakfast (Supplemental Fig. 2).

Discussion

The most popular dietary hypothesis to explain the secular increase in prevalence of obesity in U.S. children includes a food environment contributing to increases in food amounts, frequency of eating, snacking, breakfast skipping, beverage consumption, and energy density of foods (43, 44, 66–68). Our results suggest that trends in adoption of diet and meal behaviors in response to environmental imperatives may differ by family SES. First, the expected upward secular trend in the amount of foods and beverages (and energy intake) was noted only in the lowest family PIR and education categories (<130% of PIR or <12 y); in higher PIR and education categories, the secular trends were either not present or were downward. Second, the direction of associations of family income and education with dietary behaviors changed over time. For example, the independent associations of PIR and education with energy intake, amount of foods and beverages, percent energy from beverages, and number of eating occasions in the 1971–1974 and 1976–1980 surveys were positive rather than inverse. In later surveys, these associations were either absent or inverse. Collectively, these results suggest SES differentials in susceptibility to the changing food environment. We are not aware of other studies of a similar nature for corroborative evidence.

Our results suggest that both PIR and education were independent correlates of the amount of foods (energy intake and amounts of foods and beverages) and form of foods (percent energy from all beverages and nutritive beverages). However, income and education differed in their associations with meal behaviors. In 6- to 11- and 12- to 19-y olds, higher income was associated with a higher number of eating occasions. Education, however, was unrelated with frequency of eating in any age group but predicted lower energy from snack episodes and higher likelihood of reporting breakfast; moreover, these associations did not change over time. These findings suggest that education countered the possibly adverse societal trends of increased snacking and breakfast skipping.

We noted fewer secular changes in the SES diet association in 2- to 5-y olds relative to older children, especially 12- to 19-y olds. With increasing autonomy in food selection and consumption, adolescents may readily adopt dietary behaviors in response to the changing food environment. However, in 2- to 5-y olds, care providers may temper the environmental effect by control of when, what, and how much food is made available.

Recent reports have suggested that an income constraint may be associated with selection of energy-dense foods with few protective nutrients (69–71). In the present study, family PIR was not independently associated with energy density (kcal/g) of foods and this association did not change over time. Family education and energy density, however, were inversely associated, which suggests that the quality of foods selected on the basis of their energy content relative to weight (kcal/g) may be related to knowledge rather than income. These results suggest that while need-based supplemental food assistance programs can decrease dietary differentials related to resource constraints, education-related knowledge gaps in food quality remain and present as intervention targets.

To our knowledge, few published studies have examined the question asked in this study using comparable methods. Miech et al. (2) reported that increases in breakfast skipping and energy from sweetened beverages from 1988–1994 to 1999–2002 were more likely in adolescents with family PIR below the poverty threshold. Notably, however, these analyses did not adjust for family education. The frequency of breakfast reporting in another nationally representative sample of 8th, 10th, and 12th graders from 1993 to 2003 was in accord with our findings of education differentials that did not change over time (42).

To provide context for our results, we also examined trends in the prevalence of high BMI (BMI percentile of ≥95th) and SES associations. Expectedly, a strong secular trend in increasing prevalence of high BMI was noted in all age groups. Unlike an earlier report of complex associations of family PIR and prevalence of obesity in U.S. children from 1971–1974 to 1999–2002 (3), in our analysis, with family education in the models, there was no independent association of family PIR in predicting high BMI. In 2- to 5-y olds, neither PIR nor education was related to prevalence of high BMI. Family education was a strong inverse correlate of high BMI in 6- to 11-y olds regardless of survey; in 12- to 19-y olds, prevalence of high BMI increased in all education categories over time and the inverse association of education and high BMI was noted in all survey years except 2003–2008. The observed inverse associations between family education and prevalence of obesity are in accord with findings of a recent systematic review (4). Overall, we found family education to be an independent correlate of both dietary and meal behaviors and prevalence of high BMI.

Our study has the following limitations. The dietary recall methodology used in the various NHANESs changed over time. New recall methods implemented in 2002 are thought to have improved the completeness of the recall (72); however, to our knowledge, the validation studies of the USDA’s Automated Multiple Pass Method have been limited to adults. As noted in the methods section, the coding of eating episodes and beverages also differed across surveys. We note, however, that within each survey, the same methods were used for all respondents. Thus, while we urge caution in interpretation of diet trends across surveys, our methods allow valid conclusions about changes in associations of dietary behaviors with SES measures over time. Our caution in interpreting survey trends differs from the approach taken by others. In these reports (e.g., 16, 19, 21, 23, 27), time trends analysis combines different types of surveys (e.g., the Nationwide Food Consumption Surveys, the Continuing Survey of Food Intakes by Individuals, and the NHANESs) that differ in design and coding, have different response rates, mix recalls and records for dietary data collection, and have inconsistent adjustment for differences in socio-demographic characteristics across surveys. By using the NHANES, which has its limitation as noted above, we restricted our multiple variable-adjusted trends analysis to surveys with approximately same design and response rates and at least the same retrospective method (24-h recall) of dietary assessment.

Biased reporting (mostly energy underreporting) is an acknowledged problem in self-reported dietary data (73, 74). Due to heightened awareness of diet and body weight concerns, it is possible that the prevalence of biased reporting may have increased over time. If biased reporting has increased, we may expect a secular decline in energy intake. However, such secular trends in energy intake were not seen in any age group. Moreover, the average energy intakes in these surveys were well within the reported range (from sedentary to high physical activity level) of estimated energy requirements for children in the age groups examined in our study (75). The usual approach to account for low energy reporting is to exclude the so-called “low energy reporters” based on comparison of reported energy intakes with calculated energy requirements. Such an approach is not without limitations and the sample remaining after exclusions may not be a population representative sample. Essentially, all published reports examine secular trends without consideration for misreporting. Nevertheless, in the absence of published evidence to suggest that reporting bias associated with low income and education differed across surveys, we urge cautious interpretation of our results.

Lastly, the dietary estimates from a single recall are not the usual long-term intakes of individuals. Therefore, we present regression-adjusted means and proportions that are considered unbiased estimates of usual intakes of groups (76). Also, in our analysis, dietary variables are dependent variables; therefore, their random measurement error will be accounted for in the error terms of the regression models.

In conclusion, our results suggest both income and education differentials in susceptibility to secular increases in reported energy intake and amount of foods and beverages. However, meal behaviors, energy density, and BMI were related mostly with education regardless of survey. Further work on the possibility of differential SES associated bias in dietary reporting across surveys is indicated. Overall, our results suggest the vital importance of educational efforts, among adolescents in particular, to promote desirable dietary practices to close the knowledge gaps associated with low level of education.