3

Methodological Approaches to Data Collection

Through its review of the evidence, the committee identified elements of the methodologies used for data collection that inform the interpretation and affect the comparability of estimates of obesity prevalence and trends in the U.S. population, particularly among children, adolescents, and young adults. This chapter highlights key elements of the data collection process that differ across data sources and published reports. These include the study design, the settings for data collection, the individuals included in the data source, and data collection methodologies. This chapter also serves as the foundation for Chapter 4, in which specific data sources are evaluated.

This chapter contains two terms or phrases that have the potential to be interpreted in multiple ways. This report uses the following definitions (a full glossary can be found in Appendix A):

• “Published reports” specifically describes the articles the committee used as its evidence base (see Appendix C).
• “Estimate of obesity prevalence or trend” or “estimate” describes a statistic about the proportion or number of individuals affected with obesity at one point in time (prevalence) or over time (trend).

STUDY DESIGNS USED IN DATA SOURCES

Published reports presenting estimates of obesity prevalence or trends have been based on analyses of cross-sectional, repeated cross-sectional, and, to a lesser extent, longitudinal data. Data from each of the three study designs have been used to estimate obesity prevalence. However, only repeated cross-sectional and longitudinal data can be used to assess changes and trends in obesity prevalence over time. In repeated cross-sectional studies, different individuals are sampled at each time point. Longitudinal designs allow for examination of obesity status from the life course

perspective and can be used to assess growth trajectories in children. Such information can help contribute to a fuller understanding of intrapersonal variation in obesity within a population. Longitudinal data, however, are not as commonly used as cross-sectional studies due to factors such as the expense of following a single population over a period of many years. Existing longitudinal studies that are population-representative are generally designed to investigate risk factors for major chronic diseases, such as cardiovascular disease (The ARIC Investigators, 1989). Accordingly, many large, nationally representative surveys that provide data on obesity rely on single or repeated cross-sections (see Chapter 4). A summary of the potential advantages and disadvantages associated with each of these study designs is presented in Table 3-1. The validity and reliability of the obesity prevalence estimates obtained through the various study designs depend to a large extent on the sampling design used. The committee acknowledges that data from intervention studies also are pervasive in the published

TABLE 3-1 Potential Advantages and Disadvantages of Using Cross-Sectional, Repeated Cross-Sectional, and Longitudinal Study Designs to Assess Obesity Prevalence and Trends

NOTE: The committee acknowledges that other study designs, such as interventions, exist.

^a The potential advantages and disadvantages are contingent on the population assessed, the methodology employed, the analytic approach, and the end user seeking to apply such information. Population and methodologic considerations are discussed throughout this chapter. The analytic considerations are more fully explored in Chapter 5, while considerations related to end users are discussed in Chapter 6.

^b Not all reports using cross-sectional data evaluate multiple groups.

literature on obesity, but identified limitations in their ability to be used for the purpose of estimating obesity prevalence or trends (see Box 3-1).

SETTINGS OF DATA COLLECTION

Common settings in which data on obesity status are currently being collected on children, adolescents, and young adults include schools, medical facilities and public health settings, and other research and surveillance settings. The first two categories represent discrete physical locations, and their potential advantages and disadvantages are summarized in Table 3-2. “Other research and surveillance settings” is a heterogeneous category that encompasses settings that are often specific to a particular data source. The committee highlights illustrative examples of other research and surveillance settings, but acknowledges that others exist beyond those presented.

TABLE 3-2 Potential Advantages and Disadvantages Associated with Two Data Collection Settings Common in Reports on Obesity Prevalence and Trends

NOTES: The committee acknowledges other settings for data collection exist. Given their heterogeneity and specificity, other research and surveillance settings have been omitted from this table but are discussed in the text. FERPA, Family Educational Rights and Privacy Act; ICD, International Classification of Diseases.

^a The potential advantages and disadvantages are contingent on the population assessed, the methodology employed, the analytic approach, and the end user seeking to apply such information. Population and methodologic considerations are discussed throughout this chapter. The analytic considerations are more fully explored in Chapter 5, while considerations related to end users are discussed in Chapter 6.

^b Tables D-1, D-2, and D-3 in Appendix D present different protocols used to collect height and weight data.

Schools

Approximately half of states in the United States have legislation related to, or that allows for, school-based screening or surveillance of body mass index (BMI) or weight status (Ruggieri and Bass, 2015). The Centers for Disease Control and Prevention’s (CDC’s) 2014 School Health Policies and Practices Study, which sampled a nationally representative collection of public and private schools throughout the country, reported that slightly more than half of sampled schools obtain and maintain information regarding students’ weight status in the students’ records (overall: 54.1 percent [95 percent confidence interval [CI]: 47.2-60.9 percent]) (CDC, 2015c). As will be highlighted in Chapter 4, the grades assessed, number of students

evaluated, and frequency of evaluations differs across sites. Protocols, who performs the measurements, and level of training provided to relevant staff also differs across assessments (see Appendix D, Tables D1, D-2, and D-3), which can affect the quality of data obtained.

Although schools have the potential to be a centralized setting for collecting data on a wide variety of children and adolescents, and may serve as a source of longitudinal data, the committee acknowledges limitations and potential barriers also exist. Three key considerations are the protection of student privacy, parental consent, and the types of schools that are represented by the data.

Protection of Student Privacy

Access to data placed in a student’s educational record, which can include measurements of height and weight conducted at school or as part of enrollment paperwork, can be limited depending on the state and the school district’s interpretation of the Family Educational Rights and Privacy Act (FERPA; see Box 3-2).

In an effort to protect student privacy, some school-based data collection efforts compile aggregate data as opposed to individual-level data. To arrive at an obesity prevalence estimate for districts throughout New York State (exclusive of New York City), for example, a “School Reporter (Nurse)” uses a tally sheet to summarize the weight status classification of the students; the tally sheet is transmitted to the “School District Reporter,” the sole person responsible for entering data for the entire district, who then submits the aggregate numbers to the state using a secure reporting system portal (New York State Center for School Health, 2015). Although aggregate data protect the identity of the students, they limit investigators’ ability to assess the obesity prevalence relationship with individual-level characteristics.

Parental Consent

Two types of parental consent can be used in the school setting: active (having to take an action to opt in) and passive (participating unless an action is taken to opt out). The use of active parental consent can dramatically decrease student participation rates compared to passive parental consent (CDC, 2014b). Active consent has the potential to bias results of an evaluation if the group that opts in does not adequately represent the population at large. This challenge is not unique to the assessment of obesity status in schools (Chartier et al., 2008), and strategies have been explored for increasing active consent response rates (Pokorny et al., 2001; Wolfenden et al., 2009).

Types of Schools Represented

School-based assessments that seek to describe a geographical region rather than an individual school may be limited in their ability to capture schools outside the public school system. As exemplified by the state and local Youth Risk Behavior Surveys (YRBSs), inclusion of private, alternative, and other school types other than public is not always possible (CDC, 2013a). Considering approximately 10 percent of students in the United States attend private schools (NCES, 2014), exclusion or inability to represent these students and those attending other school types, affects the generalizability of the results.

Clinical and Public Health Settings

Individuals who received services at clinical settings and those who participate in public health programs (e.g., the Special Supplemental Nutrition Program for Women, Infants, and Children [WIC]) routinely have their height and weight directly measured as a standard of practice. Data of this nature, considered administrative data, have been used to assess obesity prevalence and trends at different levels, from a single medical practice (Nader et al., 2014) through large-scale assessments (Arterburn et al., 2010; Hruby et al., 2015).

Published reports based on such data vary on what data are used to classify obesity status. Some reports have used the International Classification of Diseases (ICD) codes to identify those affected with obesity in lieu of using direct measured values of height and weight (George et al., 2011; Joyce et al., 2015; Koebnick et al., 2009). Prevalence estimates based on ICD codes may underestimate the prevalence of obesity in adults and children compared to other data collection methods (Al Kazzi et al., 2015; Walsh et al., 2013).

The committee acknowledges that the use of electronic health records (EHRs) and associated medical record databases to assess obesity prevalence and trends is emerging in published reports. This topic will be further discussed in Chapter 4.

Other Research and Surveillance Settings

Other settings for capturing data related to obesity exist beyond schools, medical facilities, and public health programs. The settings for these evaluations are typically specific to the data source, and encompass both in-person and remote data collection. Four data sources are highlighted below to illustrate differences in such settings. Differences in specific data collection approaches are discussed in greater detail in Chapter 4.

In-Person Data Collection

The specific physical locations in which data are collected in person vary across data sources. The National Health Interview Survey (NHIS), for example, is conducted in participants’ homes as a face-to-face interview but collects only reported data (i.e., does not collect directly measured heights and weights) (NCHS, 2015). In contrast, the National Health and Nutrition Examination Survey (NHANES) uses specially designed mobile units that travel around the country and contain state-of-the-art equipment and laboratory space for physical assessments and biological sample collection and processing (Zipf et al., 2013).

Remote Data Collection

Not all data used in published reports on obesity prevalence and trends are collected in a defined physical location. Some are collected remotely. The California Health Interview Survey, for example, is a phone-based survey that captures height and weight data on children, adolescents, and adults (UCLA Center for Health Policy Research, 2016). The redesign of the National Survey of Children’s Health (NSCH), which is currently being pretested, plans to use both mail- and Web-based data collection, with minimal phone contact (MCHB, 2015).

INDIVIDUALS INCLUDED IN THE DATA SOURCE

The individuals who are included in the data source, and ultimately in the analysis, are the basis of estimates of obesity prevalence and trends. Key considerations when reviewing information about the study population and sample include the source of the data and associated sampling approach, size of the study sample, demographic characteristics of the study sample, as well as the extent to which the study population was stable during the time period when data were collected for trends analyses. The committee identified three key features related to the individuals included in a data source: sampling approach, sample size, and stability of the population over time. These features vary across published reports and, in turn, affect the interpretation of an estimate of obesity prevalence or trend.

Sampling Approach

It is typically not feasible or an efficient use of resources to measure every individual to determine obesity prevalence and trends in a given population. Sampling provides options to maximize generalizability and specificity of a sample, especially when fixed resources limit the number

of individuals who can be assessed in a given period of time.¹ The type of sampling approach used will also affect the ability to generate precise estimates of prevalence and trends in specific subgroups of interest. The committee identified sampling approaches that appear in published reports and considered approaches that may be used in small-scale or local settings, but not be published in the peer-reviewed evidence base. The potential advantages and disadvantages of the identified approaches are summarized in Table 3-3. The committee also identified data sources in which sampling strategies were not used because the vast majority of individuals in the population contributed data. A description of two such examples and the interpretation considerations associated with each are presented in Box 3-3.

When the total population is not assessed, the generalizability of resulting estimates of obesity prevalence and trends can become compromised if the participants in the sample do not reflect the overall target population. Sampling strategies can be carefully employed to prevent and correct for bias to the extent possible. Bias can affect the prevalence and trends estimates, and can be challenging to account for in a study. As previously mentioned, some school-based BMI assessments may require active parental consent. Students who return a signed consent form may or may not represent the overall student population. Calculation of response rates and comparisons of the sampled population to the total target population can provide insight into the representativeness of the data and facilitate adjustment for potential sources of bias.

Intentional Oversampling

Intentional oversampling is a technique used across different sampling approaches. When intentional oversampling is used, a group is sampled at a higher proportion than it exists within the target population, which can provide a more precise and stable estimate of prevalence for the group that is oversampled. NHANES, for example, has oversampled a variety of groups over the course of the survey history (Johnson et al., 2014). Since the 2011 cycle, one such group has been non-Hispanic, non-black Asian individuals. The 2010 Census indicated that this group comprised 4.8 percent of the total U.S. population (U.S. Census Bureau, 2011). However, to ensure enough participants identifying as Asian were evaluated, approximately 14 percent of the 2011-2014 NHANES sample were Asian (Johnson et al., 2014).

___________________

¹ For additional information about sampling approaches, the reader is referred to “Survey Sampling” (Kish, 1965).

TABLE 3-3 Potential Advantages and Disadvantages Associated with Sampling Approaches, for the Purposes of Generating Estimates of Obesity Prevalence in a Population

NOTE: Other sampling approaches exist.

^a The potential advantages and disadvantages are contingent on the population assessed, the methodology employed, the analytic approach, and the end user seeking to apply such information. Population and methodologic considerations are discussed throughout this chapter. The analytic considerations are more fully explored in Chapter 5, while considerations related to end users are discussed in Chapter 6.

^b Multistage sample involves two or more rounds of sampling. The principle is the same for each subsequent round of sampling.

Sample Size

As exemplified by the need for intentional oversampling, the sample size largely determines what statistical procedures and comparisons can be meaningfully conducted. Estimates of prevalence and trend are more stable when they are based on larger samples. Even in very large samples, the representation of a subpopulation of interest may be small, which could lead to highly variable estimates. Variability in estimates is generally expressed in terms of standard error and confidence intervals.

Stability of the Population Over Time

When considering estimates of prevalence over a period of time, the extent to which the underlying population studied remained stable is an

important consideration. Changes in the distribution of exposures and sociodemographic characteristics within subgroups of the population over time can affect the interpretation of obesity prevalence data. These changes should be measured and accounted for during the analysis. Comparisons of contextual variables, such as the birth rate, unemployment rate, poverty rate, income distribution, and the racial and ethnic composition of the study sample at the first time point and at subsequent measurement points, can provide a basis for gauging the stability of the population under consideration. The influx or efflux of population groups, especially those with a disproportionate risk of obesity, can affect the interpretation of the results if not accounted for during the analysis.

DATA COLLECTION METHODOLOGIES

Surveillance systems and studies use a variety of protocols to capture data on obesity status, as well as various measures to capture the key demographic characteristics discussed in Chapter 2. This variability presents challenges for data analysis and interpretation.

Collecting Data Related to Obesity Status

BMI is calculated from an individual’s height and weight. This information can be captured either through direct measure or by proxy- or self-report. The potential advantages and disadvantages of these approaches are summarized in Table 3-4.

Directly Measured Heights and Weights

Directly measured heights and weights require a data collector, a scale, and a means for measuring height (typically a stadiometer). As described in Box 3-4, protocols for capturing directly measured height and weight vary in terms of the equipment used, participant procedures, and data collector procedures.

Special Considerations Obtaining a directly measured height or weight may present a challenge in some populations, such as young children and individuals with severe obesity.

Young children The accuracy of directly measured height and weights in preschool-aged children is contingent, in part, on the child’s ability to cooperate and follow directions. Compared to their school-aged and older counterparts, preschool-aged children are more susceptible to obesity status misclassification with relatively small inaccuracies in height or weight (Ogden, 2015) (see Table 3-5).

TABLE 3-4 Potential Advantages and Disadvantages of Directly Measured and Reported Height and Weight Data

^a The potential advantages and disadvantages are contingent on the population assessed, the methodology employed, the analytic approach, and the end user seeking to apply such information. Population and methodologic considerations are discussed throughout this chapter. The analytic considerations are more fully explored in Chapter 5, while considerations related to end users are discussed in Chapter 6.

Individuals with severe obesity Individual who have severe obesity may not be able to follow the standard protocol for height and weight. The anthropometric protocol for NHANES, for example, instructs data collectors to capture height when head, shoulder blades, buttocks, and heels are in contact with the stadiometer backboard, but provides additional instructions for capturing height when obesity prevents such a positioning (CDC, 2013b). An individual who has severe obesity also can have a weight that exceeds the capacity of the scale. NHANES provides additional procedures for capturing weight in such instances (CDC, 2013b). Not all protocols, however, make such specifications (see Appendix D, Tables D-1 and D-2).

Reported Heights and Weights

A variety of factors—such as study design, sample size, participant characteristics, and participant accessibility—can make directly measuring height and weight not feasible. Rather than forego evaluation of obesity status in these instances, investigators ask a participant and/or a proxy for the participant to report weight and height. The convenience and ease of capturing self- and proxy-reported height and weight data make it an attractive option for data collection.

The collection of self- or proxy-reported height and weight data should be accompanied by consideration of the impact on the representative-

ness of the sampled population. Reported weight and height may be captured through a paper questionnaire, interview (phone or in-person), or computer-based survey and can enhance or restrict the sampled population in different ways. For example, a paper-based questionnaire will generally require the participant to have reading comprehension at or above the level in which the question is written. Phone-based interviews, in contrast, require the participant or the household to have an operational phone at the time of data collection. In the same vein, language of delivery also has implications for the sample included in the report. Some protocols have the capacity to ask the question only in English. Other protocols have the capacity to ask the question in English, Spanish, and other languages that are dominant within the population(s) of interest (UCLA Center for Health Policy Research, 2016). Phrasing and clarity of the question also

TABLE 3-5 Illustrative Examples Demonstrating Differences in Height or Weight That Categorically Change Weight Status from Normal to Obese at Two Different Ages

NOTE: All data are hypothetical. Calculations were performed using the CDC BMI Calculator for Child and Teen (CDC, 2015a). Values were selected to correspond to a BMI that would be at the upper threshold of the normal BMI-for-age category, or at the lower bound of the obese category. The CDC calculator allows for height to be entered to the nearest tenth of a centimeter and weight to be entered to the nearest tenth of a kilogram.

^a Approximates the 84th percentile on the 2000 CDC sex-specific BMI-for-age growth charts. This is the upper threshold for what is classified as “normal.” Other terminology that has been used for this category is “healthy.”

^b Approximates the 95th percentile on the 2000 CDC sex-specific BMI-for-age growth charts. This is the lower threshold for what is classified as “obese.”

vary slightly across different data sources. Most protocols and surveys use a similar question base (“How much do you weigh?”/“How tall are you?”), but some will expand on these to include a time frame (“now”), desired units (“in pounds”/“in feet and inches”), and specify “without shoes on” (see Table 3-6). Finally, the person providing the reported weight and height should be considered. Some collect self-reported data only from high-school aged students. Others will collect information about elementary-school aged children from a parent or guardian, but allow adolescents to report for themselves.

TABLE 3-6 Variation in Questions Asking for Reported Weight and Height

Measured Versus Reported Weights and Heights

The discussion below highlights the differences between proxy-reported heights and weights and self-reported heights and weights, in relation to directly measured values.

Proxy-Reported Heights and Weights Depending on the data source, the proxy reporting the child’s height and weight can be a parent, guardian, or adult in the household who is most knowledgeable about a child. In young children, relatively small differences between proxy-reported and directly measured values can cause a significant shift in the child’s weight status classification (see Table 3-5), with estimates of obesity prevalence notably affected by errors in proxy-reported heights (Akinbami and Ogden, 2009; Weden et al., 2013). In general, proxy-reported weights are lower than directly measured values, but the error may vary by factors such as the child’s age, sex, and weight status (Lundahl et al., 2014; O’Connor and Gugenheim, 2011).

In a study comparing different nationally representative surveys, investigators found mean height from the NHIS proxy-reported data was 3 to 6 cm less than the mean height from NHANES directly measured data among children ages 2 to 11 years (data 1999-2004; P<0.001) (Akinbami and Ogden, 2009). Akinbami and Ogden (2009) concluded that discrepancies between proxy-reported and measured values lead to misclassification of weight status and BMI in preschool- and elementary-aged children, and therefore recommended that proxy-reported measures not be used to estimate obesity prevalence for these ages (see Table 3-7).

Although the differences between proxy-reported and directly measured height and weight values have been assessed in cross-sectional study designs, the committee did not identify reports comparing secular or longitudinal trends in the two data collection approaches.

Self-Reported Heights and Weights In general, self-reported heights and weights tend to underestimate BMI in both adolescents and adults, although reporting error can vary by factors such as age, sex, race and ethnicity, and weight status (Gillum and Sempos, 2005; Jayawardene et al., 2014; Mozumdar and Liguori, 2016). Elementary school-aged children generally do not accurately self-report their heights and weights and ability to self-report reasonable values appears to be better among older children than younger children (Beck et al., 2012). In a convenience sample of adolescents, self-reported height were overreported and weight underreported, but were found to be reliably reported when tested and re-tested over a 2-week period (Brener et al., 2003). Some reports have noted that female adolescents have a tendency to underreport their weight more than

TABLE 3-7 Differences in Mean Heights and Weights from Nationally Representative Directly Measured (NHANES) and Proxy-Reported (NHIS) Data and Associated Effects on Obesity Prevalence Estimates

NOTES: Absolute difference values were obtained from mean height and weight data from two nationally representative surveys, NHANES (directly measured) and NHIS (proxy-reported), 1999-2004. Data presented do not express changes in bias over the life course, but rather estimates at a specific time.

^a Obesity defined as BMI ≥95th percentile according to the 2000 CDC sex-specific BMI-for-age growth charts.

^b Statistical significance based on t-test P<0.05.

^c Trends suggested an overestimation of obesity prevalence for this age group.

^d Trends suggested a similar estimate of obesity prevalence for this age group.

^e Trends suggested an underestimation of obesity prevalence for this age group.

SOURCE: Akinbami and Ogden, 2009. Data adapted and reprinted with permission.

their male counterparts (Sherry et al., 2007). When evaluated by weight status, data self-reported by adolescents with elevated weight statuses tend to lead to an underestimation of BMI, while those reported by adolescents with underweight tend to overestimate BMI (Jayawardene et al., 2014; Morrissey et al., 2006). Because bias in self-reported height and weight affects the resulting BMI and can thereby affect obesity prevalence (see Table 3-8), some investigators have suggested that such data have limited utility for estimating obesity prevalence in adolescent populations or may be used cautiously when measured data are not available (Morrissey et al.,

TABLE 3-8 Bias in Self-Reported Heights and Weights Compared to Directly Measured Data and Associated Effect on Obesity Prevalence Among Children and Adolescents

NOTES: Only studies evaluating U.S. child and adolescent populations are included. This table describes the overall bias, and not the magnitude, of under- or overestimation of height, weight and obesity prevalence. Reporting error can vary by age, sex, race and ethnicity, weight status, and other variables, not described by this table. Data presented do not express changes in bias over the life course, but rather estimates at a specific time. NR, not reported.

^a Shows overall direction of bias; obesity defined as BMI ≥95th percentile according to the 2000 CDC sex-specific BMI-for-age growth charts.

^b n = 21, 61, and 123 for grades 1, 3, and 5, respectively. After removing unreasonable values, children in all three grade levels significantly underreported their height and weight; the report only discussed effect on prevalence of overweight and obesity (collectively) among students in grade 5 providing reasonable height and weight values.

^c n = 426.

^d n = 3,797.

^e n = 11,495; data from the nationally representative National Longitudinal Study of Adolescent to Adult Health (Add Health). Investigators noted that obesity status was correctly classified for 96 percent of assessed adolescents.

^f n = 24,221 students in grades 8 and 11.

^g n = 2,032 from a convenience sample of students in grades 9-12.

^h n = 7,160; study sample included students in grades 9-12.

2006; Sherry et al., 2007). Underestimation of weight and overestimation of height by self-report have also been observed among adults, leading to an underestimation of obesity prevalence (Gillum and Sempos, 2005; Kuczmarski et al., 2001). In some cases, correction equations have been developed in order to improve estimates of obesity prevalence from self-reported data (Mozumdar and Liguori, 2016; Pérez et al., 2015).

Evidence on changes in reporting bias in children is limited. One longitudinal study concluded that females tended to underreport weight in adolescence, a trend that increased into early adulthood, while no change was observed in males in adolescence or early adulthood (Clarke et al., 2014).

Although adolescent self-reported height and weight data generally do not lead to obesity estimates comparable to those generated from directly measured data, analyses of national YRBS data suggest adolescent self-reported data may provide insight into the directionality of the overall trend. The national YRBS trend analysis suggests a significant linear trend in obesity prevalence among high school students between 1999 through 2013, but no statistically significant change between the two most recent cycles of data collection (2011, 2013) (CDC, 2014c; YRBS, 2014). These findings are similar in directionality to what has been reported among adolescents with directly measured heights and weights in NHANES (Ogden et al., 2014; Skinner and Skelton, 2014).

Collecting Data Related to Demographic Factors

Demographic characteristics are used to describe individuals included in the study’s population, to determine whether the study population is representative of the target population of interest, and to divide the study population into subgroups for comparisons. Consistent with its task, the committee evaluated the methodological approaches that have been used in recent reports to characterize diverse U.S. populations, particularly those that are socially disadvantaged (see Box 3-5).

Sex

For children, adolescents, and young adults, sex is a required demographic factor for classifying obesity status. Although most reports evaluate obesity prevalence and trends across both sexes, some reports evaluate just one.

Age

Like sex, age is required for classifying obesity status among children, adolescents, and young adults. Most studies and surveillance systems deter-

mine exact age based on recorded date of birth and date of measurement. Reports using de-identified or publicly available data may have access only to an age in years rather than year and months. Skinner and Skelton (2014), for example, explained that a cycle of NHANES data provided age in years only for those ages 2 to 19 years and could not be used in the same way

as the available data for previous cycles.² To address this limitation, the investigators chose to use the midpoint of the provided age in year (“e.g., an 11-year-old child would be considered 11.5 years of age”; Skinner and Skelton, 2014).

Race and Ethnicity

Race and ethnicity are frequently collected as descriptors of a population, but the way these data are collected varies. In some studies and assessments, race is a selection criterion (e.g., study of American Indian children). In broader and more general assessments, race and ethnicity are generally presented as a list of categories from which the participant chooses. One standard for race and ethnicity classification, developed by the Office of Management and Budget (OMB), offers five race categories (American Indian or Alaska Native, Asian, Black or African American, Native Hawaiian or Other Pacific Islander, White) and two ethnicity categories (Hispanic or Latino, Not Hispanic or Latino) (OMB, 1994). The standard high school YRBS questionnaire, for example, uses this classification.

Some studies go beyond the OMB approach to deconstruct the diversity classification of one or multiple race or ethnicity categories. This refinement often reflects the broader diversity of the target population as well as efforts to identify population groups at highest risk. Stingone et al. (2011), for example, further differentiated the participants identifying as Latino into Puerto Rican, Dominican, Mexican, or other Latino groups. Similarly, the 2013-2014 NHANES demographic questionnaire provided participants identifying as of Hispanic, Latino, or Spanish origin 29 ancestry groups from which to select, those identifying as of Asian origin 35 ancestry groups from which to select, and those identifying as Native Hawaiian or Pacific Islander origin 4 ancestry groups from which to select (NHANES, 2013). During data collection, participants are often offered more race and ethnicity categories to select from than appear in the results of reports. This is typically an analytic decision made due to sample size (for further explanation, see Chapter 5).

Socioeconomic Status

Measures of socioeconomic status (SES) vary across published reports of obesity prevalence and trends. Different measures exist at the individual-level and the population-level (see Appendix D, Table D-5).

___________________

² The datasets have since been updated and include age in months for the variable of age for participants ages 0 to 19 years (CDC, 2015b). Date of birth, however, is not publicly released information.

At the individual-level, household income can be used to produce different variables of SES. These include the income-to-poverty ratio; the child or family’s eligibility for and/or participation in assistance programs (e.g., WIC); and the child’s individual eligibility for free or reduced-price school meals. In many cases, these measures are preferable to assessing income directly because they offer more information regarding the family’s income status in a broader community context, and also generally have a higher response rate in questionnaires (Potter et al., 2005; Shavers, 2007).

Education also is used as a proxy of SES at the individual level, but the measures are not clearly defined. For example, parental/caregiver education has been reported as the highest level of education attained by either parent, both parents, or just the mother (Delva et al., 2007; Halloran et al., 2012; Huh et al., 2012; Sherwood et al., 2009). Education is more stable and often easier information to access than income, but the correlation with income is not direct, and same investment in education will likely yield a different income and SES in individuals of a different sex and race or ethnicity. Parent education, as well as parent employment status, age, and family structure, also have been used in reports as demographic variables on their own, and not as a proxy for SES (Fakhouri et al., 2013). Parental employment status is used in some cases as a variable to assess SES but has been regarded as a weaker assessment than income or education (Nuru-Jeter et al., 2010). Some studies also have used measures of geography to assess SES, based on trends that show disparities between urban and rural populations, as well as differences based on geographic location within the United States.

SES also has been examined at the community level. In school-based assessments, the percentage of students receiving free or reduced-price school meals and the racial or ethnic majority population have been used (Sanchez-Vaznaugh et al., 2015). Estimates of neighborhood income level and neighborhood education level also are used to gauge neighborhood SES level. Community SES measures offer additional information about an individual’s environment, and some studies have suggested that they may be used in some cases to substitute where individual data are not available, but more research is needed (Krieger et al., 2003). In studies examining the impact of SES on obesity, the investigators have even developed unique measures of community SES level to fit their needs and facilitate answering the investigators’ specific research question(s). For example, Sekhobo et al. (2014) divided New York City boroughs into low-risk and high-risk areas depending on their proximity or inaccessibility, respectively, to a District Public Health Office providing services for maternal and child health, and compared obesity rates between children in those communities.

Geography

The geographic locale of the population evaluated is an inherent characteristic of a data source and is directly linked to the representativeness of the sample. Some data collection efforts are purposely sampled in a way to represent the nation, a specific state, or defined community. Other data sources, for example WIC administrative data or school-based BMI assessments, are defined because their collection at the state or local level is required by law. Some data sources, however, are not representative of a geographic location but rather a physical location. Obesity prevalence and trends analyses of EHRs from a single medical practice would be one such example. The evaluated population may include all patients seen for a well-child appointment in a given year, for example, but would not be representative of all children in the city or town in which the medical practice is located. Beyond a single geographic location, some data sources are designed to capture multiple states or localities, which can be used for comparative assessments. Specific examples are described in Chapter 4.

Rurality

The rurality of a defined geographic region has been classified and used in different ways in reports on obesity prevalence and trends. The level of rurality (or urbanicity) is typically defined by the total population or population density. For example, an evaluation of obesity among white and American Indian school children in South Dakota dichotomized the samples as residing in urban (two cities with a population >50,000) or rural (rest of state) locations (Hearst et al., 2013). In contrast, an evaluation in Pennsylvania had four categories: urban (≥1,000 population per square mile), suburban (999 to 300 population per square mile), rural (299 to 100 population per square mile), and ultrarural (<100 population per square mile) (Bailey-Davis et al., 2012). In some reports, the level of rurality or urbanicity defines the entire sampled and target population. Data from the Bogalusa Heart Study, for example, are described by investigators as being from a “semirural” population in Louisiana (Broyles et al., 2010).

SUMMARY

Data on height and weight have been collected for research studies using cross-sectional, repeated cross-sectional, and, to a lesser extent, longitudinal designs. These data also have been collected for other purposes including school-based screenings, surveillance, and as part of routine health care.

The ability to generalize estimates beyond the study population is contingent, in part, on how the individuals were selected for inclusion.

Sampling procedures have been used to not only arrive at a representative sample, but also to generate more precise estimates for relatively small subgroups of interest.

Approaches used to capture height and weight data direct measurement include self-report and proxy-report. Estimates of obesity prevalence calculated from these data collection approaches are generally not interchangeable. Obesity trends based on self-reported heights and weights from nationally representative samples of high school students suggest such data may provide insight into the general directionality of obesity trends over time, similar to those calculated from directly measured data.

Collecting and using data on key demographic characteristics varies across data sources and reports. Obesity status classification for children, adolescents, and young adults depends on both sex and age of the participant, and these two variables appear in most reports. The number of race and ethnicity categories offered to participants varies across data sources, with some studies capturing specific origin and ancestry groups. Similarly, measures of SES used across reports differ and may not be directly comparable. The geographic boundaries of estimates are reliant on the representativeness of the sampled population; not all data sources represent a broader geography. Finally, approaches to classifying levels of rurality and the number of categories included vary across reports.

REFERENCES

Adler, N. E., and J. M. Ostrove. 1999. Socioeconomic status and health: What we know and what we don’t. Annals of the New York Academy of Sciences 896:3-15.

Akinbami, L. J., and C. L. Ogden. 2009. Childhood overweight prevalence in the United States: The impact of parent-reported height and weight. Obesity 17(8):1574-1580.

Al Kazzi, E. S., B. Lau, T. Li, E. B. Schneider, M. A. Makary, and S. Hutfless. 2015. Differences in the prevalence of obesity, smoking and alcohol in the United States Nationwide Inpatient Sample and the Behavioral Risk Factor Surveillance System. PLoS ONE 10(11):e0140165.

The ARIC (Atherosclerosis Risk in Communities) Investigators. 1989. The Atherosclerosis Risk in Communities (ARIC) Study: Design and objectives. American Journal of Epidemiology 129(4):687-702.

Arterburn, D. E., G. L. Alexander, J. Calvi, L. A. Coleman, M. W. Gillman, R. Novotny, V. P. Quinn, M. Rukstalis, V. J. Stevens, E. M. Taveras, and N. E. Sherwood. 2010. Body mass index measurement and obesity prevalence in ten U.S. health plans. Clinical Medicine & Research 8(3-4):126-130.

Bailey-Davis, L., M. Horst, M. M. Hillemeier, and A. Lauter. 2012. Obesity disparities among elementary-aged children: Data from school-based BMI surveillance. Pediatrics 130(6):1102-1109.

Beck, J., C. A. Schaefer, H. Nace, A. D. Steffen, C. Nigg, L. Brink, J. O. Hill, and R. C. Browning. 2012. Accuracy of self-reported height and weight in children aged 6 to 11 years. Preventing Chronic Disease 9:E119.

Bethell, C., D. Read, E. Goodman, J. Johnson, J. Besl, J. Cooper, and L. A. Simpson. 2009. Consistently inconsistent: A snapshot of across- and within-state disparities in the prevalence of childhood overweight and obesity. Pediatrics 123(Suppl 5):S277-S286.

Brener, N. D., T. McManus, D. A. Galuska, R. Lowry, and H. Wechsler. 2003. Reliability and validity of self-reported height and weight among high school students. Journal of Adolescent Health 32(4):281-287.

Broyles, S., P. T. Katzmarzyk, S. R. Srinivasan, W. Chen, C. Bouchard, D. S. Freedman, and G. S. Berenson. 2010. The pediatric obesity epidemic continues unabated in Bogalusa, Louisiana. Pediatrics 125(5):900-905.

CDC (Centers for Disease Control and Prevention). 2013a. Methodology of the Youth Risk Behavior Surveillance System—2013. Morbidity and Mortality Weekly Report 62(1).

CDC. 2013b. National Health and Nutrition Examination Survey (NHANES) anthropometry procedures manual. http://www.cdc.gov/nchs/data/nhanes/nhanes_13_14/2013_Anthropometry.pdf (accessed June 7, 2016).

CDC. 2014a. Behavioral Risk Factor Surveillance System questionnaire. http://www.cdc.gov/brfss/questionnaires/pdf-ques/2014_brfss.pdf (accessed March 14, 2016).

CDC. 2014b. A guide to conducting your own Youth Risk Behavior Survey. http://www.cdc.gov/healthyyouth/data/yrbs/pdf/yrbs_conducting_your_own.pdf (accessed March 28, 2016).

CDC. 2014c. Youth Risk Behavior Surveillance—United States, 2013. Morbidity and Mortality Weekly Report 63(4).

CDC. 2015a. BMI percentile calculator for child and teen. http://nccd.cdc.gov/dnpabmi/Calculator.aspx (accessed March 10, 2016).

CDC. 2015b. Demographic variables and sample weights (DEMO_G). https://wwwn.cdc.gov/Nchs/Nhanes/2011-2012/DEMO_G.htm (accessed April 25, 2016).

CDC. 2015c. Results from the School Health Policies and Practices Study 2014. http://www.cdc.gov/healthyyouth/data/shpps/pdf/shpps-508-final_101315.pdf (accessed March 28, 2016).

Chartier, M., A. V. Stoep, E. McCauley, J. R. Herting, M. Tracy, and J. Lymp. 2008. Passive versus active parental consent: Implications for the ability of school-based depression screening to reach youth at risk. Journal of School Health 78(3):157-186.

Cheng, T. L., and E. Goodman. 2015. Race, ethnicity, and socioeconomic status in research on child health. Pediatrics 135(1):e225-e237.

CHIS (California Health Interview Survey). 2015. CHIS 2013-2014 adolescent questionnaire. http://healthpolicy.ucla.edu/chis/design/Documents/chis2013teenquestionnaire.pdf (accessed March 24, 2016).

Clarke, P., N. Sastry, D. Duffy, and J. Ailshire. 2014. Accuracy of self-reported versus measured weight over adolescence and young adulthood: Findings from the National Longitudinal Study of Adolescent Health, 1996-2008. American Journal of Epidemiology 180(2):153-159.

Dalenius, K., E. Borland, B. Smith, B. Polhamus, and L. Grummer-Strawn. 2012. Pediatric Nutrition Surveillance 2010 report. Atlanta: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention.

Delva, J., L. D. Johnston, and P. M. O’Malley. 2007. The epidemiology of overweight and related lifestyle behaviors: Racial/ethnic and socioeconomic status differences among American youth. American Journal of Preventive Medicine 33(Supplement 4):S178-S186.

Economos, C. D., R. R. Hyatt, J. P. Goldberg, A. Must, E. N. Naumova, J. J. Collins, and M. E. Nelson. 2007. A community intervention reduces BMI z-score in children: Shape Up Somerville first year results. Obesity 15(5):1325-1336.

Ennis, S. R., M. Ríos-Vargas, and N. G. Albert. 2011. The Hispanic population: 2010. Washington, DC: U.S. Government Printing Office.

Fakhouri, T. H., J. P. Hughes, D. J. Brody, B. K. Kit, and C. L. Ogden. 2013. Physical activity and screen-time viewing among elementary school-aged children in the United States from 2009 to 2010. JAMA Pediatrics 167(3):223-229.

George, M. G., X. Tong, E. V. Kuklina, and D. R. Labarthe. 2011. Trends in stroke hospitalizations and associated risk factors among children and young adults, 1995-2008. Annals of Neurology 70(5):713-721.

Gillum, R. F., and C. T. Sempos. 2005. Ethnic variation in validity of classification of overweight and obesity using self-reported weight and height in American women and men: The third National Health and Nutrition Examination Survey. Nutrition Journal 4:27.

Goodman, E. 1999. The role of socioeconomic status gradients in explaining differences in US adolescents’ health. American Journal of Public Health 89(10):1522-1528.

Goodman, E., B. R. Hinden, and S. Khandelwal. 2000. Accuracy of teen and parental reports of obesity and body mass index. Pediatrics 106(1):52-58.

Halloran, D. R., N. E. Marshall, R. M. Kunovich, and A. B. Caughey. 2012. Obesity trends and perinatal outcomes in black and white teenagers. American Journal of Obstetrics and Gynecology 207(6):492.e1-492.e7.

Healthy Youth Survey. 2014. Washington State Healthy Youth Survey-form B. http://www.doh.wa.gov/portals/1/Documents/Pubs/160-NonDOH-HYS-Form-B2014.pdf (accessed March 24, 2016).

Hearst, M. O., K. Biskeborn, M. Christensen, and C. Cushing. 2013. Trends of overweight and obesity among white and American Indian school children in South Dakota, 1998-2010. Obesity (Silver Spring) 21(1):E26-E32.

Himes, J. H., P. Hannan, M. Wall, and D. Neumark-Sztainer. 2005. Factors associated with errors in self-reports of stature, weight, and body mass index in Minnesota adolescents. Annals of Epidemiology 15(4):272-278.

Hixson, L., B. B. Hepler, and M. Ouk Kim. 2011. The white population: 2010. Washington, DC: U.S. Government Printing Office.

Hoeffel, E. M., S. Rastogi, M. Ouk Kim, and H. Shahid. 2012. The Asian population: 2010. Washington, DC: U.S. Government Printing Office.

Hruby, A., O. T. Hill, L. Bulathsinhala, C. J. McKinnon, S. J. Montain, A. J. Young, and T. J. Smith. 2015. Trends in overweight and obesity in soldiers entering the US Army, 1989-2012. Obesity (Silver Spring) 23(3):662-670.

Huh, D., E. Stice, H. Shaw, and K. Boutelle. 2012. Female overweight and obesity in adolescence: Developmental trends and ethnic differences in prevalence, incidence, and remission. Journal of Youth and Adolescents 41(1):76-85.

Jayawardene, W., D. Lohrmann, and A. YoussefAgha. 2014. Discrepant body mass index: Behaviors associated with height and weight misreporting among US adolescents from the National Youth Physical Activity and Nutrition Study. Childhood Obesity 10(3):225-233.

Johnson, C., S. Dohrmann, V. Burt, and L. Mohadjer. 2014. National Health and Nutrition Examination Survey: Sample design, 2011-2014. Vital and Health Statistics (2)162.

Joyce, N., G. A. Wellenius, D. D. Dore, J. W. Newburger, and J. P. Zachariah. 2015. Patterns of lipid lowering therapy among children ages 8-20 years. Journal of Pediatrics 167(1):113-119.

Kish, L. 1965. Survey sampling. New York: John Wiley & Sons, Inc.

Koebnick, C., D. Getahun, K. Reynolds, K. J. Coleman, A. H. Porter, J. M. Lawrence, M. Punyanitya, V. P. Quinn, and S. J. Jacobsen. 2009. Trends in nonalcoholic fatty liver disease-related hospitalizations in US children, adolescents, and young adults. Journal of Pediatric Gastroenterology and Nutrition 48(5):597-603.

Krieger, N., J. Chen, P. Waterman, M. Soobader, S. Subramanian, and R. Carson. 2003. Choosing area based socioeconomic measures to monitor social inequalities in low birth weight and childhood lead poisoning: The public health disparities geocoding project (US). Journal of Epidemiology and Community Health 57(3):186-199.

Kuczmarski, M. F., R. J. Kuczmarski, and M. Najjar. 2001. Effects of age on validity of self-reported height, weight, and body mass index: Findings from the third National Health and Nutrition Examination Survey, 1988-1994. Journal of the American Dietetic Association 101(1):28-34.

Lundahl, A., K. M. Kidwell, and T. D. Nelson. 2014. Parental underestimates of child weight: A meta-analysis. Pediatrics 133(3):e689-e703.

Matthews, K. A., and L. C. Gallo. 2011. Psychological perspectives on pathways linking socioeconomic status and physical health. Annual Review of Psychology 62:501-530.

MCHB (Maternal and Child Health Bureau). 2015. Dataspeak: Progress on the redesign of the NSCH and the data resource center. PowerPoint presented at DataSpeak Webinars.

Morrissey, S. L., L. M. Whetstone, D. M. Cummings, and L. J. Owen. 2006. Comparison of self-reported and measured height and weight in eighth-grade students. Journal of School Health 76(10):512-515.

Mozumdar, A., and G. Liguori. 2016. Corrective equations to self-reported height and weight for obesity estimates among US adults: NHANES 1999-2008. Research Quarterly for Exercise and Sport 87(1):47-58.

Nader, N., V. Singhal, A. Javed, A. Weaver, and S. Kumar. 2014. Temporal trends in the diagnosis and management of childhood obesity/overweight in primary care. Journal of Primary Care Community Health 5(1):44-49.

NCES (National Center for Education Statistics). 2014. Private school enrollment. Institute of Education Sciences. http://nces.ed.gov/programs/coe/indicator_cgc.asp (accessed March 7, 2016).

NCHS (National Center for Health Statistics). 2015. 2014 National Health Interview Survey (NHIS) survey description. ftp://ftp.cdc.gov/pub/Health_Statistics/NCHS/Dataset_Documentation/NHIS/2014/srvydesc.pdf (accessed March 7, 2016).

New York State Center for School Health. 2015. 2015-2016 student weight status category (SWSC) reporting. http://www.schoolhealthservicesny.com/datareporting.cfm?subpage=104 (accessed March 7, 2016).

NHANES (National Health and Nutrition Examination Survey). 2013. Questionnaire: Demographics information. http://www.cdc.gov/nchs/data/nhanes/nhanes_13_14/DMQ_H.pdf (accessed March 24, 2016).

NLS (National Longitudinal Surveys). 2008. National Longitudinal Survey of Youth questionnaire. https://www.nlsinfo.org/sites/nlsinfo.org/files/attachments/121214/ChildSuppandCSAS-2008.html (accessed March 24, 2016).

Norris, T., P. L. Vines, and E. M. Hoeffel. 2012. The American Indian and Alaska Native population: 2010. Washington, DC: U.S. Government Printing Office.

NRC (National Research Council). 1995. Measuring poverty: A new approach. Washington, DC: National Academy Press.

NSCH (National Survey of Children’s Health). 2012. 2011 NSCH questionnaire. Quarter 1 2012. http://www.cdc.gov/nchs/data/slaits/2011nschquestionnaire.pdf (accessed March 14, 2016).

Nuru-Jeter, A. M., K. Sarsour, D. P. Jutte, and W. T. Boyce. 2010. Socioeconomic predictors of health and development in middle childhood: Variations by socioeconomic status measure and race. Issues in Comprehensive Pediatric Nursing 33(2):59-81.

O’Connor, D. P., and J. J. Gugenheim. 2011. Comparison of measured and parents’ reported height and weight in children and adolescents. Obesity (Silver Spring) 19(5):1040-1046.

Ogden, C. L. 2015. Obesity in the US: Beyond the data. PowerPoint presentation at Evaluating Approaches to Assessing Prevalence and Trends in Obesity: Data Gathering Public Workshop, Washington, DC.

Ogden, C. L., M. D. Carroll, B. K. Kit, and K. M. Flegal. 2014. Prevalence of childhood and adult obesity in the United States, 2011-2012. Journal of the American Medical Association 311(8):806-814.

OMB (Office of Management and Budget). 1994. Standards for the classification of federal data on race and ethnicity. https://www.whitehouse.gov/omb/fedreg_notice_15 (accessed April 15, 2016).

Pérez, A., K. P. Gabriel, E. K. Nehme, D. J. Mandell, and D. M. Hoelscher. 2015. Measuring the bias, precision, accuracy, and validity of self-reported height and weight in assessing overweight and obesity status among adolescents using a surveillance system. International Journal of Behavioral Nutrition and Physical Activity 12(Supplement 1):S2.

Pokorny, S. B., L. A. Jason, M. E. Schoeny, S. M. Townsend, and C. J. Curie. 2001. Do participation rates change when active consent procedures replace passive consent. Evaluation Review 25(5):567-580.

Potter, B. K., K. N. Speechley, I. A. Gutmanis, M. K. Campbell, J. J. Koval, and D. Manuel. 2005. A comparison of measures of socioeconomic status for adolescents in a Canadian national health survey. Chronic Disease in Canada 26(2-3):80-89.

Project EAT (Eating Among Teens). 2010. Project EAT-II survey for high school students. http://docs.sph.umn.edu/epich/eat/EAT-II_HighSchoolSurvey.pdf (accessed March 24, 2016).

Rastogi, S., T. D. Johnson, E. M. Hoeffel, and M. P. J. Drewery. 2011. The black population: 2010. Washington, DC: U.S. Government Printing Office.

Ruggieri, D. G., and S. B. Bass. 2015. A comprehensive review of school-based body mass index screening programs and their implications for school health: Do the controversies accurately reflect the research? Journal of School Health 85(1):61-72.

Sanchez-Vaznaugh, E. V., B. N. Sanchez, P. B. Crawford, and S. Egerter. 2015. Association between competitive food and beverage policies in elementary schools and childhood overweight/obesity trends: Differences by neighborhood socioeconomic resources. JAMA Pediatrics 169(5):e150781.

Sekhobo, J. P., L. S. Edmunds, D. K. Reynolds, K. Dalenius, and A. Sharma. 2010. Trends in prevalence of obesity and overweight among children enrolled in the New York State WIC program, 2002-2007. Public Health Reports 125(2):218-224.

Sekhobo, J. P., L. S. Edmunds, K. Dalenius, J. Jernigan, C. F. Davis, M. Giddings, C. Lesesne, and L. Kettel Khan. 2014. Neighborhood disparities in prevalence of childhood obesity among low-income children before and after implementation of New York City child care regulations. Preventing Chronic Disease 11:E181.

Shavers, V. L. 2007. Measurement of socioeconomic status in health disparities research. Journal of the National Medical Association 99(9):1013-1023.

Sherry, B., M. Jefferds, and L. M. Grummer-Strawn. 2007. Accuracy of adolescent self-report of height and weight in assessing overweight status: A literature review. Archives of Pediatrics and Adolescent Medicine 161(12):1154-1161.

Sherwood, N. E., Wall, M., Neumark-Sztainer, D., and M. Story. 2009. Effect of socioeconomic status on weight change patterns in adolescents. Preventing Chronic Disease 6(1):A19.

Short, K. 2012. The research supplemental poverty measure: 2011. US Census Bureau. Current Population Reports (P60-244).

Skinner, A. C., and J. A. Skelton. 2014. Prevalence and trends in obesity and severe obesity among children in the United States, 1999-2012. JAMA Pediatrics 168(6):561-566.

Stingone, J. A., O. F. Ramirez, K. Svensson, and L. Claudio. 2011. Prevalence, demographics, and health outcomes of comorbid asthma and overweight in urban children. Journal of Asthma 48(9):876-885.

Thomson, M. D., and L. Hoffman-Goetz. 2009. Defining and measuring acculturation: A systematic review of public health studies with Hispanic populations in the United States. Social Science and Medicine 69(7):983-991.

UCLA (University of California, Los Angeles) Center for Health Policy Research. 2016. What’s new in the 2013-2014 California Health Interview Survey http://healthpolicy.ucla.edu/chis/design/Documents/Whats-New-CHIS-2013-2014.pdf (accessed March 14, 2016).

U.S. Census Bureau. 2011. Overview of race and Hispanic origin: 2010. Washington, DC: U.S. Government Printing Office. http://www.census.gov/prod/cen2010/briefs/c2010br-02.pdf (accessed March 31, 2016).

Walsh, C. O., C. E. Milliren, H. A. Feldman, and E. M. Taveras. 2013. Sensitivity and specificity of obesity diagnosis in pediatric ambulatory care in the United States. Clinical Pediatrics (Phila) 52(9):829-835.

Weden, M. M., P. B. Brownell, M. S. Rendall, C. Lau, M. Fernandes, and Z. Nazarov. 2013. Parent-reported height and weight as sources of bias in survey estimates of childhood obesity. American Journal of Epidemiology 178(3):461-473.

Wolfenden, L., K. Kypri, M. Freund, and R. Hodder. 2009. Obtaining active parental consent for school-based research: A guide for researchers. Australian and New Zealand Journal of Public Health 33(3):270-275.

YRBS (Youth Risk Behavior Survey). 2014. Trends in the prevalence of obesity, dietary behaviors, and weight control practices: National YRBS: 1991-2013. http://www.cdc.gov/healthyyouth/data/yrbs/pdf/trends/us_obesity_trend_yrbs.pdf (accessed March 24, 2016).

YRBS. 2015. 2015 state and local Youth Risk Behavior Survey. ftp://ftp.cdc.gov/pub/data/yrbs/2015/2015_hs_questionnaire.pdf (accessed March 24, 2016).

Zipf, G., M. Chiappa, and K. Porter, et al. 2013. National Health and Nutrition Examination Survey: Plan and operations, 1999-2010. Vital and Health Statistics 1(56).

This page intentionally left blank.