The present CSFII analyses are based on publicly available secondary data that did not provide information that could be used to identify participants. Therefore, the IRB, which is charged with protecting people, does not consider these data "person data" and their study was exempt from review.
This study addressed three research questions: a) Are food group and nutrient intakes in adults and children living in the same household associated?, b) Are children more likely to consume diets that trend toward meeting their age-and-gender specific intake recommendations for food groups and nutrients if the female or male HH also consumes a diet that trends toward meeting her/his intake recommendations?, and c) Are children's odds to be a "big eater" higher when the female/male head of household has been identified as a "big eater" at the same eating occasion (i.e. meal type)?
These research questions were addressed using a sample of children 2-18 years old and the male and/or female head of household (HH). All individuals were participants in the Continuing Survey of Food Intake by Individuals (CSFII) 1994-1996 , a stratified, multistage, area probability sample of nationally representative data. Since it is not possible to determine if the mother or father of the child is in the data set, we used the female and male head of households as proxies, representing the female or male in charge. The HHs were identified in the survey by responses to the question "Who is the female (or male) in the household, who is in charge?" The response to this question was used to identify one female and/or male HH in each household. Unlike other approaches that limited the information of male HH to only those males who co-habitated with a female HH , we included all male HH (male only households: n = 335, female only n = 446, and both n = 940). Because more recent nationally representative data sets only provide information on one household member, they could not be used to investigate the present study's research questions regarding adult-child dyads. Thus, although the CSFII data are 15 years old, if one assumes that the basic relationships between adults and children in the same household have not changed during that time, the associations found in this study are still applicable today.
The present study included households with at least one child between 2 and 18 years old (n = 1,721 households with 3,150 children). Children were divided into three separate groups: preschoolers (2-5 year olds), school-age (6-11 year olds), and teenagers (12-18 year olds). When multiple children within the same age group resided in the same household, one child in that age group was randomly selected. Thus, 2-5-year-old children (n = 708), 6-11 year-old children (n = 836), and 12-18-year-old (n = 836) comprised the total sample (n = 2,380). The children not randomized into the study (n = 770) were excluded. This study design addressed the potential for cluster effects of individuals from the same household. We stratified all analyses by age group, including only the female or male HH and one child in each regression model for all six adult-child dyads.
Dietary intake data was obtained via one in-person 24-hour food recall and a second telephone recall collected three to ten days after the first recall but not on the same day of the week. The interview respondent reported diets of children who were younger than six years old. Children seven years and older provided their own diet information, assisted by an adult if necessary. Tippett and Cypel  provided a detailed explanation of the CSFII data collection procedures for dietary intakes.
For the first two research questions, dietary intake estimates of usual intake were based on the calculated two-day average food group and nutrient intakes in individuals with two reported days of intake (n = 2,254) and one day for those with only one day of intake (n = 126). The individuals with only one reported day of intake were not significantly different from those with two days of intake with respect to gender, ethnic, or income group.
Since the third research question was based on eating occasions, only data from the first 24-hour recall was used (n = 2,380). This approach was necessary because it is not possible to ascertain "usual intake" estimates for food group or nutrient intakes specific to an eating occasion. For instance, if an individual ate breakfast on the first day but skipped breakfast on the second day, only the first day would be included in the analysis (the arithmetic mean calculated from the intake of one day and the absence of intake on the second day would not be reflective of usual intake, i.e. (1 egg on day 1+ 0 eggs on day 2)/2 = 1/2 egg each day - which does not represent the person's eating pattern).
Eating occasions were recorded based on the respondent's description. Similar to the manner with which other researchers have used this variable , intakes were categorized by the survey respondent as breakfast, brunch, lunch, dinner, supper, food/beverage, infant feeding, extended eating, and other. Because the present study only included children who were between the ages of 2 and 18, the category "infant feeding" (n = 57) was not included. Since both, "brunch" and "lunch" were reported between 11 am and 2 pm, we combined those two categories. Similarly, we combined "supper" and "dinner". We also combined the infrequent categories "food or beverage" and "other" eating occasions. Following the consolidation, the final categories for "eating occasion" and their corresponding frequencies for children and adults were breakfast (n = 36,763)), brunch/lunch (n = 36,809), supper/dinner (n = 50,586), and "other, food/beverage" (n = 1,980).
Total energy in kcal, the number of servings from six Food Guide Pyramid  food groups (fruits, vegetables, total grain, whole grain, milk/dairy, meat), and consumption of macronutrients and micronutrients in grams and micrograms, as appropriate, for total fat, saturated fat, unsaturated fatty acids (linoleic and linolenic acid), protein, carbohydrates, cholesterol, dietary fiber, added sugar, iron, calcium, vitamins A, B12, and C, and folate were calculated. Dietary recalls that were coded in the data set as "reliable" but showed average energy consumptions with implausible values (< 500 kilocalories (kcal) of total energy intake reported per day) were excluded . To update the Food Guide Pyramid serving sizes to the corresponding MyPlate portion sizes, consumption was converted to represent the serving sizes cups and ounces of the MyPlate food groups. Means and standard errors for food group and nutrient intakes were calculated and expressed in energy-adjusted terms (servings of food groups or grams/micrograms of nutrients per 1,000 kcal of total energy consumed). Binary indicator variables were created to identify the individuals whose dietary intake was below the recommended consumption level (= 0) or those who trended toward meeting the recommended level (= 1) for total energy intake (Estimate Energy Requirements (EER)), macronutrients (Acceptable Macronutrient Distribution Range (AMDR)) or micronutrients (Estimated Average Requirement (EAR)) and Adequate Intake (AI) for dietary fiber and calcium of the Dietary Reference Intakes (DRI). Total food group density (fruit, vegetables, total grain, whole grain, milk/dairy, meat) per eating occasion (breakfast, brunch/lunch, dinner/supper, other food/beverage) was calculated to represent the energy adjusted amount of intake from each food group consumed at each eating occasion. Food group densities were ranked for each of the five population groups (female HH, male HH, children ages 2-5, 6-11, 12-18 year olds). To examine if "big eaters" clustered within families, we estimated the odds that the child's intake fell in the highest quintile of food group consumption per eating occasion if the female or male HH's intake fell in the highest quintile. This measure of analysis was chosen because grams of food consumed or calories consumed are variables that do not account for the person's intake of other foods and body size, whereas number of servings per 1,000 kcal consumed (food group density) accounts for both of these cofactors of intake. Thus, we defined "big eaters" as those individuals who consumed within the highest 20% of total food group density consumption (highest quintile) within each population group (preschoolers, school-age children, adolescents, female HH, or male HH) and eating occasion (breakfast, brunch/lunch, supper/dinner, other food/beverages).
Descriptive statistics (proportions, means and standard errors) were obtained. Linear and logistic regression models for complex sample survey data were fit to explore a) the association between the food group and nutrient densities reported for the female and/or male HH and for the children in the same household, b) the odds that the child consumed a diet that trended toward meeting his/her age- and gender-specific dietary intake recommendations for food groups and nutrients when the female/male HH consumed a diet that trended toward meeting her/his corresponding intake recommendation and, c) the odds that the child's intake fell in the highest quintile for food group density of the eating occasions if the female or male HH's intake fell in the highest quintile of the same food group and eating occasion.
To control for socio-demographic variables, age, gender, race, country of origin, years of education, employment status, daycare/school attendance, and total household size and income were examined. Education (years of schooling) was categorized as less than high school (< 12 years of school), high school (12 years of school), and more than high school (> 12 years of school). Employment status was categorized as employed or not employed. Total household income was used as a continuous variable. Its skew of the distribution was addressed by log transformation. To describe the sample, income was also categorized as: less than or equal to 1.3 times the poverty income ratio (PIR) (food stamp eligible), 1.31 to 3.5 times the PIR (medium income), and more than 3.5 times the PIR (high income). To capture cultural differences, a study participant's race (black, white and other) and Hispanic origin (from the Mexican, Puerto Rican, Cuban, other Spanish subgroups, or not) variables were re-coded into Non-Hispanic white, Non-Hispanic black, Non-Hispanic Asian, Non-Hispanic other, Mexican-American, and other Hispanic. Binary dummy variables were subsequently generated for ethnicity (not Hispanic, Hispanic) and HH's educational level (less than high school, high school graduate, more than high school graduate).
Twenty-one separate linear regressions in the six adult-child dyads were modeled for the six food groups and 15 nutrients under study. Linear regression models were fit to calculate the associations between food group and nutrient densities in the children's and adult head of household's diets; logistic regression models for complex sample survey data were used to obtain adjusted odds ratios for the odds of children consuming diets that trended toward meeting intake recommendations for food groups and nutrients if the female/male HH consumed a diet that trended toward her/him meeting her/his; and logistic regression models were used to obtain adjusted odds ratios for the odds of children to be in the highest food group density intake quintile for an eating occasion if the female/male HH was in the highest quintile. Regression models were not fitted if one of the four cells in the 2 × 2 table calculating the odds ratios had a sample size of < 10 subjects. Models controlled for household size, age, education level, ethnicity, and income. All statistical methods were executed in Stata Statistical Software: Release 9.0 (Stata Corporation, College Station, TX, USA) using complex sample survey routines that account for the CSFII's unequal selection probabilities, stratification, and clustering, and maintain the nationally representative character of the data.