Body mass index (BMI) is the most commonly used method to estimate overweight and obesity in children and adolescents, using standardized classification criteria based on the child’s height, weight, sex, and age [13]. BMI is often a critical variable included in worldwide surveillance systems and interventions to document outcomes of a program or policy, to describe epidemiology (i.e., person, place, and time) of childhood obesity, and/or to quantify the magnitude of obesity status within and across populations. In surveillance systems and interventions that include large and/or population-based sample sizes, adolescents’ height and weight are often obtained via self-report due to its low cost, ease of data collection, and the ability to efficiently collect data from a large number of individuals [46].

Some surveillance systems and other population-based studies of children and adolescents, including the National Longitudinal Study of Adolescent Health (U.S.), have incorporated ancillary studies where either all, or a subset of, participants’ heights and weights were directly measured and compared with self-reported estimates to examine validity. These comparison studies have been done in the U.S. [57], Wales [8], Portugal [9], Germany [10, 11], and Australia [12]. In general, results of these studies have shown that, while adolescent-reported estimates of height and weight are correlated with objective measurements, they typically generate a lower estimate of overweight and obesity prevalence [6, 7, 1316]

Some differences in the validity of self-reported height and weight data, by age or other socio-demographic factors, are well established. For example, studies have generally shown limited accuracy of self-reported height and weight among children aged younger than fourteen years [4, 6, 17, 18]. Further, self-reported height and weight collected from girls tends to result in greater BMI underestimation than self-reported height and weight from boys [5, 7, 10, 14, 16, 1922]. The relatively few studies that have investigated differences by race/ethnicity have not yielded consistent results [7, 13, 14, 16, 23]. A few studies targeting specific ethnic subpopulations have also been conducted, including studies of Mexican Americans [24] and American Indians [15, 25]. Despite numerous studies assessing validity of child-reported height and weight, gaps in understanding remain, particularly in regard to differences across subpopulations. A 2007 review of studies assessing the accuracy of self-reported height and weight in adolescents identified the lack of understanding about subpopulation differences as the primary gap in the literature on this subject [26]. To date, this gap has not been fully addressed.

The primary goal of this study was to examine, by subpopulations, the precision and accuracy of self-reported height and weight compared to objective measures of height and weight, in addition to the diagnostic validity of weight status (e.g., assumed objective measures as gold standard for estimating overweight and obesity), among a large, diverse and representative population of 8th- and 11th-grade adolescents in Texas, USA. Additionally, since population-based or intervention research often necessitates the collection of self-reported data, a secondary objective was to develop correction equations to estimate height, weight and BMI from self-reported height and weight data. These estimates could be used in lieu of objective measurement and improve the usefulness of self-reported measures in obesity prevention and intervention studies.


The School Physical Activity and Nutrition (SPAN) project was designed to establish a surveillance system to monitor the prevalence of overweight and obesity among Texas school children in grades 4, 8 and 11. The description and design of SPAN has been previously reported [2729]. Briefly, the first statewide SPAN survey was conducted over two academic years: 2000–2001 and 2001–2002, while the second statewide survey was administered in 2004–2005. SPAN utilized a sampling strategy involving nine Texas Health Service Region (HSR) levels, three types of communities (urban center, other urban/suburban and rural) and three grade levels, to yield representative data at the Texas state, Texas Health Service Region (HSR) levels, and for three major racial/ethnic groups in Texas: African-American, Hispanic and white/other. Unfortunately, other races/ethnicities were not considered as a subpopulation due to the low prevalence in Texas and large sample size needed to make a representative sample of other race/ethnicities. The sampling frame was created based on school and school district-level data made available from the Texas Education Agency (TEA) from the academic year preceding each respective SPAN survey. Sampling weights and post-stratification adjustments accounted for the complex design, differential representation, the use of stratification and sampling clusters, as well as updates in the sampling frame for each survey administration [2729]. The SPAN survey included items to assess (1) demographic characteristics (e.g., sex, grade, and race/ethnicity); (2) dietary intake, including meal patterns and nutrition knowledge; (3) physical activity; and (4) reported height and weight. The SPAN survey instruments for grades 8 and 11 are identical and have been previously shown to be valid and reliable [30, 31].

The first administration of the statewide SPAN survey included a sample of 5,362 and 3,576 8th- and 11th-grade children, respectively. This sample was representative of a population of 288,584 and 249,363 8th- and 11th-grade children, respectively. The second statewide SPAN survey included a sample of 8,827 and 6,456 8th- and 11th-grade children, respectively, representing their respective grade populations of 291,672 and 233,753 students.

Human subjects and consent procedures

Approval for this study was obtained from (1) the Committee for the Protection of Human Subjects at The University of Texas Health Science Center at Houston (HSC-SPH-00-056), (2) the institutional review board of the Texas Department of State Health Services (04-062) and (3) participating school districts. Depending on the school or school district, parental consent was obtained via either active or passive methods, and study participants (i.e., children) provided assent prior to data collection.


Demographic characteristics

Demographic variables collected include sex, age, grade, and race/ethnicity. Categories of response for self-reported race/ethnicity were: Black or African-American; Mexican-American, Latino or Hispanic; White, non-Hispanic, non-Latino; American Indian or Alaska Native; Asian; Native Hawaiian or Other Pacific Islander; White, non-Hispanic, non-Latino; and Other. These were collapsed into three main race/ethnicities: African-American, Hispanic, or White/other. For international comparison purposes, in the U.S., children begin their first year of formal education (kindergarten) at age 5. Eighth grade is the ninth year of formal education, also known as the third year of middle school, or lower secondary education (level 2) as classified by the United Nations Educational, Scientific and Cultural Organization’s International Standard Classification of Education (ISCED) [32]. Similarly, in the U.S., 11th grade is the twelfth year of formal education, also known as the third year of U.S. high school, or upper secondary education (ISCED level 3). As in many countries worldwide [33], U.S. students typically begin 8th-grade at age 13 years and 11th-grade at age 16 years.

Self-reported measures of height and weight

Self-reported height, recorded in feet and inches, was converted to centimeters and self-reported weight, recorded in pounds, was converted to kilograms to standardize units of expression for comparison with the objective measures of height and weight. Self-reported height (without shoes) and weight (without heavy clothes and shoes) data were collected from students in 8th- and 11th-grade. These grades were chosen in line with recommendations to not collect these measures from 4th-grade children (aged approximately 9-10 years) due to their general inability to give accurate or reasonable values for height or weight [4, 6, 17].

Objective measures of height and weight

Students’ heights and weights were measured using standardized procedures. Children removed any heavy clothes and shoes before having their height and weight measured. Height was measured to the nearest 0.1 centimeter with a portable stadiometer (Perspective Enterprises Portable Adult Measuring Unit PE-AIM-101) and weight was measured to the nearest 0.1 kg with a portable digital scale with remote display (Tanita Professional Digital Scales with Remote Display, BWB-800S) calibrated to 113 kg (i.e. 250 pounds) before each series of measurements. Study staff recorded both measures on the student questionnaires.

Using both the self-reported and objective measures, BMI was computed as weight (kilograms) divided by height (meters) squared. Then, both BMI estimates (self-reported and objective measures) were collapsed to categories reflecting weight status (i.e., underweight/normal (<85th percentile), overweight (≥85th percentile to <95th percentile) and obese (≥95th percentile)) using the Centers for Disease Control and Prevention growth charts [34, 35].

Statistical analyses

All statistical analysis for estimation takes the form of weighted statistics, using the sampling weights from each statewide survey. This provides the opportunity to look at differences in reported height and weight by (1) sex, (2) race/ethnicity, and (3) grade level. SPAN did not sample students by age, which precludes the ability to present and examine the parameters of interest by age. Differences between the self-reported and objectively-measured height and weight were explored between the two administrations of the SPAN survey (i.e., 2000–2001 versus 2004–2005 academic years) using weighted regression analysis. There were no statistically significantly differences between the first and second SPAN surveys in the self-reported and objectively-measured data adjusting by sex, grade, and race/ethnicity. Therefore, data were pooled to enhance statistical power.

Second, descriptive statistics including (1) age (mean ± standard error), (2) self-reported and objectively-measured BMI (including height and weight estimates; median ± standard error), and (3) proportion within sex, race/ethnicity, and weight status categories (% ± standard error) were calculated for each statewide survey separately and then pooled across both survey administrations. Median values with 95% confidence intervals (CI) of height, weight, and BMI were reported due to lack of normality. Third, weight status prevalence estimates, using self-report and objectively-measured estimates were reported by (1) grade and (2) sex. Then, the absolute error (bias), calculated as the difference between the self-reported prevalence and the objectively-measured prevalence, and relative error, calculated as the absolute error divided by the objectively-measured estimate, were reported. The prevalence of weight status is presented for normal, overweight and obese status, as well as for combined overweight and obese status.

Fourth, agreement between self-reported and objectively-measured height, weight, and BMI was assessed using Lin’s concordance correlation coefficient (rho_c) by (1) sex and (2) race/ethnicity for 8th- and 11th-grade students, separately. Because Lin’s concordance correlation coefficient combines measures of precision (Pearson correlation) and accuracy (bias correction factor=C-bias), the overall correlation as well as estimates reflecting accuracy and precision are reported.

Fifth, we developed correction equations to estimate objectively measured height, weight, and BMI (i.e., dependent variables) from self-reported height and/or weight by grade, sex, and race/ethnicity using weighted linear regression. Although, SPAN did not sample by age, age was included as a covariate in the initial correction equations. However, age was not a statistically significant contributor to the equations and was removed from the final correction equations. Because there were statistically significant differences noted by grade (2 levels), sex (2 levels), and race/ethnicity (3 levels), 12 (=2*2*3) linear correction equations for these combinations are reported. The coefficient of determination (R2) reported how much of the variability of the dependent variables was explained by the independent variables for each correction equation.

Sixth, the validity of BMI from self-reported height and weight data to appropriately classify overweight and obesity status was assessed by estimating the (1) sensitivity, (2) specificity and (3) positive predictive value by grade level and by sex. Finally, self-reported measures and correction equations were used to estimate weight status. Sensitivity, specificity and positive predictive values of the correction equations were computed.



As shown in Table 1, the mean age for 8th- and 11th-grade students was 13.7, and 16.7 years, respectively. These mean ages were consistent with the age based on grade level. Around 53% of the sample was male, and the majority (56.6%) were non-Hispanic White. Using the objective measurements, the median and standard error (SE) values for BMI were 21.5 kg/m2 (±0.18SE) and 23.1(±0.16SE) kg/m2 among 8th- and 11th-graders, respectively. Among 8th-grade students, the prevalence of overweight and obesity was 17.9% and 18.1%, respectively. The prevalence of overweight and obesity in 11th-grade students was 16.5% and 15.8%, respectively.

Table 1 Descriptive statistics of SPAN 2000-2002 and 2004-2005 by grade

Differences in medians of height, weight, and BMI

With regards to height, among 8th-grade students, statistically significant differences between self-reported and objective measurements were seen in (1) Hispanic boys, (2) Hispanic girls and (3) African-American girls (Table 2). Across all categories of race/ethnicity, 11th-grade students overestimated their height, ranging from 0.68 to 1.04 cm among girls and from 1.87 to 2.02 cm among boys.

Table 2 Differences between self-reported and objective measures in SPAN 2000-2002 and 2004-2005 pooled data

With regards to weight, among girls there were statistically significant differences between self-reported and objectively-measured estimates across all categories of race/ethnicity in both grades, with the exception of 11th-grade Hispanic girls. Hispanics boys in 8th-grade underreported their weight [median -0.68 kg 95% CI (-1.03; -0.33)]. With regard to BMI, estimates obtained from self-reported measurements were lower than those from objectively-measured data across all grade, sex and race/ethnicity categories, with the exception of 8th-grade African-American boys. Differences ranged from -0.23 to -0.70 kg/m2.

Differences in prevalence of weight status

Overall self-reported height and weight data underestimated the prevalence of overweight and obesity when compared to the objective measures (Table 3). Further, the absolute and relative error estimates varied by grade and sex. The largest underestimation of the prevalence of obesity was shown among 8th-grade girls (absolute and relative error of -4.1% and -0.25, respectively).

Table 3 Prevalence of weight status categories (Standard error of percent), absolute error and relative error

Concordance coefficients

Table 4 shows (1) Lin’s concordance coefficients, (2) precision (Pearson correlation) and (3) accuracy (bias-correction factor), between self-reported and objectively-measured height, weight, and BMI for 8th- and 11th-grade students. Lin’s concordance coefficients for height ranged from 0.60 to 0.90; for weight from 0.82 to 097; and for BMI, from 0.78 to 0.95. The lowest precision was found in height among 8th- and 11th-grade Hispanic girls (0.66) and the greatest precision observed was in weight among 11th-grade African-American boys (0.97).

Table 4 Lin’s concordance coefficients1(95%CI), Pearson correlation coefficient and bias correction factor(BCF) between self-reported and objective measurements

Linear regression equations

The lowest amount of variation in objectively-measured height explained by self-reported height was observed among Hispanics, regardless of their sex (43.1% for 8th-grade girls, 47.4% for 8th-grade boys, 44% for 11th-grade girls and 56.9% for 11th-grade boys) (Table 5). This indicates that there are some additional factors that were not measured in SPAN to predict objectively-measured height. The variation of objective measures explained by self-reported measures was higher for weight than for height for all sex, grade, and across race/ethnicity categories (Table 6). Across sex, grade, and race/ethnicity, the BMI variability explained when using self-reported weight and the square inverse of self-reported height was above 77% with the exception of 8th-grade African-American girls (65.9%) (see Table 7)

Table 5 Correction equations to estimate objectively-measured height (cm) using self-reported data
Table 6 Correction equations to estimate objectively-measured weight (Kg) using self-reported data
Table 7 Correction equations to estimate objectively-measured BMI (kg/m2) using self-reported data

Sensitivity, specificity and positive predictive value

As shown in Table 8, sensitivity, which is the proportion of overweight students correctly classified by self-reported weight status, was 60.5% and 62.0% for 8th- and 11th-grade students, respectively. The sensitivity for obese weight status was 70.8% and 81.9% for 8th- and 11th-grade students, respectively. The sensitivity for combined overweight/obese status was 81.3% and 83.1% for 8th- and 11th-grade students, respectively. Further, specificity, which is the proportion of students who are not overweight/obese who are classified as not overweight/obese by self-reported weight and height was 91.1% and 95.6% for 8th- and 11th-grade students, respectively. The specificity for obese weight status was 97.0% and 98.1% for 8th- and 11th-grade students, respectively.

Table 8 Sensitivity, specificity and positive predictive value of self-reported data and correction equations for weight status

The positive predictive value, which is the proportion of students identified by self-reported weight status as overweight/obese that are truly overweight/obese was 83.4% and 89.9% among 8th- and 11th-grade students, respectively; the positive predictive value for obese as weight status was 83.4% and 89.1% among 8th- and 11th-grade students, respectively.

The sensitivity of estimates obtained using the correction equations was improved over the self-reported measures. However, the positive predictive value was not consistently improved across sex, grades and weight status (Table 8).


This study examined the bias, precision and accuracy of self-reported height, weight, and BMI derived from self-reported measures, across all sex, grade and race/ethnicity using data from a large surveillance system of 8th- and 11th-grade Texas adolescents. Due to the lack of prior studies exploring these properties in adolescents using probability samples, it is difficult to discuss our findings within the context of previous research. Despite this, we identified several findings. Prior studies have recommended against the use of self-report data from children under the age of 14 [4, 6, 17, 18]. This study suggests that a slightly lower age cutoff for using self-reported height and weight may be appropriate, but this is arguable. Although statistically significant differences were noted between self-report and objective measures overall and by subpopulations, median differences between self-report and objective measures were small. Overall, height tended to be over-reported relative to objective measures. Self-reported measurements were taken in feet and inches; children are unlikely to report height with a greater precision than 0.5 inches. The increased bias in height and a decreased bias in weight resulted in a small downward bias in BMI calculated from self-report measures in almost all sex, grade, and the race/ethnicity subpopulations. These findings are generally consistent with previous publications [57, 10, 14, 16, 1922, 36]. The slightly lower relative errors in estimating the prevalence of overweight/obese among youth in Texas could be due to the fact that, since 1998, it is standard practice to have height and weight measured during sports team participation or physical education classes in schools and could potentially create greater awareness of actual height and weight in our population [37]. This could also explain why Lin’s concordance coefficients were similar for height and weight for this Texas sample compared to samples from nationally representative data in the U.S. [6]. These sensitivity results in Texas are similar to results from nationally representative data (range 59% to 76% for overweight/obese and 70.2% to 74% for obese) [26].

To our knowledge, this is the first study to develop correction equations to estimate measured height, weight and BMI from self-reported data by each level of sex (two levels), grade (two levels), and race/ethnicity (three levels) among adolescents. With correction equations, sensitivity to classify overweight/obese and obese status rose to a minimum of 85.9% and 76.3%, respectively, across categories. These equations are particularly useful for public health researchers who are restricted to self-reported measures of height and weight due to budget and/or staff resource constraints. However, the positive predictive value of weight status was not improved by the correction equations. Researchers interested in obesity as an outcome will need to weigh the pros and cons of using correction equations instead of self-reported measures within the context of their own research and study design.

This study has several strengths and limitations. One of the strengths is that Texas can be viewed as a representative state in terms of its increasingly diverse racial/ethnic profile and the steadily increasing Hispanic population [38]. Thus, the findings of this study will be applicable to the projected changes in U.S. racial/ethnic demography [39]. The second strength is the large sample size and availability of a representative sample of middle school and high school students in Texas, both of which greatly contribute to the generalizability of these findings. One limitation is the study’s need to collapse non-Hispanic White, Asian, Pacific Islander, Native American, and “other” within the White/other category. Retaining the original categories will require a large sample size. Asian girls usually have lower BMI than other girls and we were unable to capture this in SPAN. As previously mentioned, grade level, rather than age was used in the sampling strategy. While this limits the interpretation of results to an international audience, grade provides a reasonable proxy for age. Finally, menarche can also change BMI as fat distribution changes, but SPAN did not measure age of menarche to account for this. One final limitation is that data on the school-level percentage of students receiving free and reduced lunch, a marker of socio-economic status, were unavailable for the 2000–2002 academic year and, therefore, not included as a term in the correction equations.


Direct measurements of height and weight are preferable over self-report measures when seeking to estimate prevalence of weight status among students. However, when direct measurement is not practical, self-reported measurements provide a reasonable proxy measure across grade, sex, and racial/ethnic subpopulations of adolescents. Researchers should be cautious of the potential for bias, particularly among girls. This study’s findings suggest that the use of correction equations for reported data is a reasonable alternative when direct measurement of height and weight is not feasible.