See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/5790785

Accuracy of Adolescent Self-report of Height and Weight in Assessing

Overweight Status: A Literature Review

Article  in  Archives of Pediatrics and Adolescent Medicine · January 2008

DOI: 10.1001/archpedi.161.12.1154 · Source: PubMed

CITATIONS READS

326 744

3 authors:

Bettylou Sherry Maria Elena del Socorro Jefferds

Centers for Disease Control and Prevention Centers for Disease Control and Prevention
69 PUBLICATIONS   6,011 CITATIONS    87 PUBLICATIONS   1,877 CITATIONS   

SEE PROFILE SEE PROFILE

Laurence M Grummer-Strawn
World Health Organization WHO
186 PUBLICATIONS   26,389 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Maria Elena del Socorro Jefferds on 04 August 2014.

The user has requested enhancement of the downloaded file.

 REVIEW ARTICLE

Accuracy of Adolescent Self-report of Height

and Weight in Assessing Overweight Status
A Literature Review
Bettylou Sherry, PhD, RD; Maria Elena Jefferds, PhD; Laurence M. Grummer-Strawn, PhD

Objective: To examine the accuracy of self-reported
height and weight data to classify adolescent over-
weight status. Self-reported height and weight are com-
monly used with minimal consideration of accuracy.
Data Sources: Eleven studies (4 nationally represen-
tative, 7 convenience sample or locally based).
Study Selection: Peer-reviewed articles of studies con-
ducted in the United States that compared self-reported
and directly measured height, weight, and/or body mass
index data to classify overweight among adolescents.
Main Exposures: Self-reported and directly measured
height and weight.
Main Outcome Measures: Overweight prevalence;
missing data, bias, and accuracy.
Results: Studies varied in examination of bias. Sensi-
tivity of self-reported data for classification of over-
weight ranged from 55% to 76% (4 of 4 studies). Over-
weight prevalence was −0.4% to −17.7% lower when body
mass index was based on self-reported data vs directly
measured data (5 of 5 studies). Females underestimated
weight more than males (ranges, −4.0 to −1.0 kg vs −2.6
to 1.5 kg, respectively) (9 of 9 studies); overweight in-
dividuals underestimated weight more than nonover-
weight individuals (6 of 6 studies). Missing self-
reported data ranged from 0% to 23% (9 of 9 studies).
There was inadequate information on bias by age and race/
ethnicity.
Conclusions: Self-reported data are valuable if the only
source of data. However, self-reported data underestimate
overweight prevalence and there is bias by sex and weight
status. Lower sensitivities of self-reported data indicate that
one-fourth to one-half of those overweight would be missed.
Other potential biases in self-reported data, such as across
subgroups, need further clarification. The feasibility of col-
lecting directly measured height and weight data on a state/
community level should be explored because directly mea-
sured data are more accurate.
Arch Pediatr Adolesc Med. 2007;161(12):1154-1161

Author Affiliations: Maternal
Child Nutrition Branch,
Division of Nutrition and
Physical Activity, Centers for
Disease Control and Prevention,
Atlanta, Georgia.
O
NE IMPORTANT ASPECT OF
addressing pediatric
overweight as a national
public health priority is
to document and moni-
tor the prevalence.1 Prevalence data de-
scribe the extent of the problem and show
trends over time; they are also useful for
targeting interventions to geographic areas
or subgroups with the highest preva-
lence. Monitoring changes in prevalence
is an important component of evaluating
intervention programs to prevent and treat
overweight; in particular, these data are
useful for understanding and guiding pro-
gram effectiveness, improvement, and
maintenance. The most commonly used
anthropometric screening tool for over-
weight is body mass index (BMI) (calcu-
lated as weight in kilograms divided by
height in meters squared). Thus, height
and weight data are essential for classify-
ing overweight status. How these data are
collected becomes an important issue as
state health departments and researchers
more actively try, with limited resources,
to monitor the impact of programs to re-
duce the obesity epidemic.
Self-reported height and weight are
sometimes used to determine and/or moni-
tor overweight. The self-reported method
is chosen in preference to direct measure-
ment because self-reported data are easier,
cheaper, and quicker to collect. More-
over, direct measurement of height and
weight requires appropriate training and
monitoring of personnel and the use of ac-
curate, well-maintained equipment. Be-
cause of limited resources, self-reported
data may be the only data available. Un-
fortunately, minimal attention is given to
potential inaccuracies or bias introduced
with the self-reported method. More-
over, articles in the literature may give no

(REPRINTED) ARCH PEDIATR ADOLESC MED/ VOL 161 (NO. 12), DEC 2007 WWW.ARCHPEDIATRICS.COM
1154
Downloaded from www.archpediatrics.com at CDC-Information Center, on December 4, 2007
©2007 American Medical Association. All rights reserved.
or limited mention of the quality of self-reported data.
For example, Valenti et al2 reported overweight preva-
lence of Australian children with no mention of the qual-
ity of their parental-reported data; Sen3 examined the
frequency of family dinner and weight status based on
self-reported data from the National Longitudinal Sur-
vey of Youth. Sen cites the Strauss4 National Health and
Nutrition Examination Survey III analysis, which re-
ported that 94% of adolescents had their weight status
correctly classified, as justification for the accuracy of self-
reported data. As will be seen in this review, a single sum-
mary statistic can mask bias.
The goals of this review were to examine and summa-
rize the accuracy of using self-reported data in contrast to
directly measured data for identifying and monitoring over-
weight among US adolescents. We examined consistency
across studies in the magnitude of bias by age, sex, race/
ethnicity, and prevalence of overweight and the extent of
prevalence bias by sensitivity and specificity.
METHODS
LITERATURE SEARCH
We used a comprehensive literature search of the electronic da-
tabases PubMed and MEDLINE, including “in-process
MEDLINE.” The search terms included body weight, body
height, and body mass index; adolescent, youth, or teenage; self-
report, self-disclosure, and self-assessment; reproducibility of
results, sensitivity and specificity, analysis of variance, and mul-
tivariate analysis; valid, reliable, and accurate; and English lan-
guage. We did not limit the years of our search. We identified
60 articles, but only 12 studies met our inclusion criteria of peer-
reviewed articles conducted in the United States that com-
pared self-reported and directly measured height and weight
data or BMIs calculated from these different measures for ado-
lescents. We excluded 1 additional study5 that did not include
correlation coefficients between self-reported and directly
measured data or sensitivity and specificity analyses but only
presented self-reported and directly measured mean differ-
ences. Two studies4,6 used National Health and Nutrition Ex-
amination Survey III data for their analyses but examined dif-
ferent aspects of bias, so we included both.
CODING AND ANALYSIS OF STUDIES
For each study, we identified the geographic location; study
design, including sampling protocol, setting, sequence of mea-
surements, measurement protocols, definitions of at risk for over-
weight and/or overweight status, including reference sources
and exclusion of biologically implausible values; and popula-
tion characteristics, including age, race/ethnicity, sample size,
and directly measured prevalence of overweight. For each study
we extracted percentage of missing data; differences between
the self-reported and directly measured height, weight, and BMI;
stratification of these differences by age, sex, race/ethnicity, and
overweight status, when available; regression analysis to iden-
tify predictors of self-reported bias; differences between self-
reported and directly measured prevalence for both at risk for
overweight and overweight; correlation coefficients between self-
reported and directly measured height and weight and calcu-
lated BMI; and sensitivity, specificity, and positive predictive
value data for prevalence. When data were not available, we
contacted authors in an attempt to obtain complete informa-
tion. For our review, 2 of us (B.S. and M.E.J.) independently
(REPRINTED) ARCH PEDIATR ADOLESC MED/ VOL 161 (NO. 12), DEC 2007
1155
Downloaded from www.archpediatrics.com at CDC-Information Center, on December 4, 2007
©2007 American Medical Association. All rights reserved.
summarized each study and resolved differences by discus-
sion. We analyzed the consistency of patterns across the 11 stud-
ies for the previously listed parameters.
Meta-analysis or a moderator analysis was not appropriate
to define an effect size for this review because there were only
11 studies and they varied substantially in sample size and com-
position, study design, and definition of overweight. Several
studies included small convenience samples or were locally
based, making their findings nongeneralizable to the overall US
population of adolescents.
RESULTS
An overview of the 11 studies (4 based on nationally rep-
resentative samples and 7 convenience sample or lo-
cally based studies) is summarized in Table 1. We strati-
fied our summary of studies into these 2 categories to
differentiate the generalizability of their results. All stud-
ies included children 11 years or older. Human sub-
jects’ approval was obtained for all of the 11 studies.
DIFFERENCES IN STUDY METHODS
Data for the 4 nationally representative studies were col-
lected during the early 1980s through the mid 1990s; all
used standardized protocols for measurements. Inter-
vals between collection of self-reported and directly mea-
sured data ranged from the same day8 to a few weeks.4,6,7
Only Himes and Faricy6 described exclusion of data with
biologically implausible values: they excluded subjects
with differences between self-reported and directly mea-
sured height or weight values more than 5 SDs of the over-
all mean differences.
Among the 7 convenience sample or locally based stud-
ies, data were collected over a wider period (1970s through
2000)9,10,14,15; 3 studies did not report year(s) of data col-
lection.11-13 Various measurement protocols were used: 29,12
used referenced protocols, 214,15 used standardized but un-
referenced protocols, 111 described some procedures, and
210,13 did not describe. The interval between the collection
of self-reported and directly measured data ranged from the
same day to a month or so9-11,13-15; one did not report the
interval.12 Two studies described exclusions of biologi-
cally implausible values: Shannon et al,11 “zero” values and
height values greater than 400 in, and Stewart,9 height and
weight values 4 or more SDs from mean difference of self-
reported and directly measured data.
Moreover, these 11 studies used a variety of refer-
ences and/or cutoffs to define overweight status. These
variations made comparisons among studies difficult.
MISSING DATA
Among the national surveys, missing data (unknown/
unusable) for self-reported values ranged from 14% to
37%.4,6-8 Himes and Faricy6 showed an effect by age: 41%
of 12-year-olds and 25% of 13-year-olds did not report
their weight, in contrast to only 2% of the 14- to 16-year-
olds. Their logistic regression model showed that younger
age, being male, shorter directly measured height, and
lighter directly measured weight were all independently
and significantly associated with missing data.6
WWW.ARCHPEDIATRICS.COM
 Table 1. Characteristics of Studies in Reviewed Literature

Aware of DM/SR
Sample Size Data Before
Source Sample Source (Males/Females) Age Range, y a Race/Ethnicity DM b
Nationally Representative Studies
Davis and Gergen7 HHANES 829 (392/437) 12-19 Mexican American Yes/Yes
Strauss4 NHANES III 1657 (767/890) 12-16 White/Hispanic, Yes/Yes
black, other
Himes and Faricy6 NHANES III 1635 (759/876) 12-16 Non-Hispanic white, Yes/Yes
non-Hispanic
black, Mexican
American
Goodman et al8 Add Health 11 495 (5748/5747) Grades 7-12 Non-Hispanic white, NR/No
non-Hispanic
black, Hispanic,
Asian, other
Convenience Sample or Locally Based Surveys
Stewart9 6 Sites, Ohio, 428 (NA) 14-17 85% White for ages NR/Yes
Washington, 14-61 y
Massachusetts,
South Carolina c
Brooks-Gunn et al10 4 Private day 151 (0/151) 11-13 White Yes/Yes
schools, 1 city
Shannon et al11 School districts in 726 (337/389) 11-13 White NR/Yes
Pennsylvania
Himes and Story12 Indian reservation 69 (41/28) 12-19 Native American No/Yes
school in
Minnesota
Hauck et al13 3 Indian Health 536 (NA/NA) 12-19 Native American NR/Yes
Service areas
Brener et al14 Subsample, 20 2032 (957/1075) Grades 9-12 White, black, Yes/Yes
states and Hispanic, other
Washington, DC
Himes et al15 Comprehensive for 3797 (1936/1861) 12-18 White, black, NR/Yes
31 schools in Hispanic, Asian,
Minnesota d other

Abbreviations: Add Health, National Longitudinal Study of Adolescent Health; DM, directly measured; HNANES, Hispanic Health and Nutrition Examination
Survey; NA, not available; NHANES, National Health and Nutrition Examination Survey; NR, not reported; SR, self-reported.
a As reported in published articles. Includes highest year noted (eg, 12-16 includes up to 16.99 years).
b Participants aware that they would/might be measured.
c Rand Health Insurance Study.
d Project EAT (Eating Among Teens).

Only 5 convenience sample or locally based studies re-
ported missing data, which ranged from 0% to 31%.10,11,13-15
Shannon et al11 found no differences in missing self-reported
data by height, weight, or sex, but they did not examine the
effect by age in their study of 11- to 13-year-olds.
DIFFERENCES IN MEANS
Table 2 shows the differences in mean (self-reported−di-
rectly measured) height, weight, and BMI by sex for the 9
studieswiththesedataavailable.Forheight,weight,andBMI,
the magnitude of mean bias by sex was greater for partici-
pants in the convenience sample or locally based studies than
in the nationally representative studies.
Height
Height was both overestimated and underestimated. For
the 7 studies for both males and females, the range of mean
height differences was relatively small at −1.1 cm to 2.4
cm, with the exception of Brener et al14 who found dif-
ferences of 6.6 cm for males and 6.9 cm for females. Shan-
non et al11 described tall males and females underesti-
mating their weight more than their shorter counterparts
but did not control for age.
Weight
Differences in mean self-reported and directly mea-
sured weights for the 9 studies that stratified by sex
showed that females generally underestimated their weight
more than males (Table 2). Differences in means varied
widely: females ranged from −4.0 to −1.0 kg and males,
from −2.6 to 1.5 kg.
Body Mass Index
Two nationally representative studies6,8 and 3 conve-
nience sample or locally based studies12,14,15 reported mean
differences by sex for BMI (Table 2). Both males and fe-
males showed minimal mean differences in BMI (0 to 0.2
for males; −0.1 to −0.3 for females) in the nationally rep-

(REPRINTED) ARCH PEDIATR ADOLESC MED/ VOL 161 (NO. 12), DEC 2007 WWW.ARCHPEDIATRICS.COM
1156
Downloaded from www.archpediatrics.com at CDC-Information Center, on December 4, 2007
©2007 American Medical Association. All rights reserved.
 Table 2. Sex-Specific Differences Between SR and DM Means for Height, Weight, and BMI

Differences in Means (SR − DM)

Height, cm Weight, kg BMI

Source M F M F M F
Nationally Representative Surveys
Strauss4 −0.4 −0.5 0.1 −1.0 NR NR
Himes and Faricy6 −1.0 −1.1 −0.5 −1.2 0.2 −0.1
Goodman et al8 NR NR −0.2 −1.0 0 −0.3
Convenience Sample or Locally Based Surveys a
Brooks-Gunn et al10 NA 1.4 NA −1.2 NA NR
Shannon et al11 0.5 −0.5 −2.1 −4.0 NR NR
Himes and Story12 0.9 0.0 −2.6 −2.8 −1.2 −1.0
Hauck et al13 1.5 0.5 1.5 −2.1 NR NR
Brener et al14 6.6 6.9 −1.1 −2.0 −2.3 −3.0
Himes et al15 b 1.2 2.4 −1.6 −3.8 −2.2 −2.5
All studies, range −1.0 to 6.6 −1.1 to 6.9 −2.6 to 1.5 −4.0 to −1.0 −2.3 to 0.2 −3.0 to −0.1

Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); DM, directly measured; F, female; M, male;
NA, not available, study included only females; NR, not reported; SR, self-reported.
a Stewart9 did not report if participants were aware of DM data or stratify by sex (height = 0.9 cm, weight = −1.1 kg).
b Means adjusted from mixed linear models by sex and included school, race, age, socioeconomic status, and categories of measured dimensions that were
specific to each of the variables (eg, schools = random, stature = 6 categories).

resentative studies, whereas there were greater underes-
timations (−1.2 to −2.3 for males; −1.0 to −3.0 for fe-
males) in the convenience sample or locally based studies.
Age or School Grade
Bias by age or school grade was reported by 4 studies. For
the 3 studies that statistically examined bias by increasing
age, 2 found increasing differences between self-reported
and directly measured height,14,15 2 found decreasing dif-
ferences between self-reported and directly measured
weight,12,15 and 2 found decreasing differences between self-
reported and directly measured BMI12,14; for each of these
indexes, 1 of these 3 studies found no difference. Using a
mixed-effects general linear model, Himes and Story12 re-
ported that age was an independent predictor of reporting
error for weight and BMI, but not for height, with the de-
gree of underestimation of both self-reported weight and
calculated BMI based on self-reported data decreasing with
increasing age (errors less negative for each additional year
in age by 0.78 kg for weight and 0.35 for BMI, respec-
tively). Brener et al14 found that, after controlling for sex
and race/ethnicity in linear regression analyses, each year’s
increase in school grade was positively associated with an
overestimate of self-reported height (␤=0.22; P⬍.001) and
inversely associated with an underestimate of BMI based
on self-reported values (␤=−0.19; P=.002); they did not
report any difference in weight bias by school grade.
ling for sex and school grade in their regression analy-
sis, Brener et al14 found that white individuals overesti-
mated their height more than black or Hispanic individuals
or those of other racial or ethnic groups (␤ = 0.09;
P⬍.001). Himes et al15 found significant effects for males
only, with Asian individuals more accurately self-
reporting height and underestimating weight more than
white, African American, or Hispanic individuals or those
of other racial or ethnic groups.
Weight Status
Six of the 11 studies examined reporting bias by weight
status and found that overweight youth underestimated
their weight4,7,11-13,15 and, if available, their resulting BMI,7,12
more than normal-weight youth. For example, Strauss4
found that mean differences between self-reported and
directly measured weight were significantly greater among
overweight (BMI⬎95th percentile) adolescents vs those
not overweight (−4.6 kg vs 0.2 kg, respectively; P⬍.001).
An additional study9 found underreporting of weight in-
creased as directly measured weight increased for their
entire sample (aged 14-61 years). Studies varied in these
comparisons, making it impossible to examine bias by
percentage of underestimation of directly measured weight
by directly measured weight status.
DIFFERENCES IN PREVALENCE

Race/Ethnicity We examined both absolute error (absolute difference be-

Bias by self-reported race/ethnicity was examined statis-
tically in 3 studies. Findings were disparate. Strauss4 re-
ported no statistically significant differences in report-
ing of height and weight by race/ethnicity (white/
Hispanic individuals, black individuals, and individuals
of other race/ethnicity) stratified by sex. After control-
tween self-reported and directly measured prevalence)
and relative error (absolute error divided by directly mea-
sured prevalence) in the 5 studies that had data avail-
able (Table 3). These studies showed that adolescent
self-reported height and weight data underestimated
prevalence of overweight, risk for overweight, or both,
with wide variability in both absolute and relative er-

(REPRINTED) ARCH PEDIATR ADOLESC MED/ VOL 161 (NO. 12), DEC 2007 WWW.ARCHPEDIATRICS.COM
1157
Downloaded from www.archpediatrics.com at CDC-Information Center, on December 4, 2007
©2007 American Medical Association. All rights reserved.
 Table 3. SR and DM Prevalence of Overweight, Absolute Error, and Relative Error

Bias in Prevalence Overweight, %

Reference Used Overweight SR Prevalence DM Prevalence Absolute Error a Relative Error b
to Classify Classification
Source Weight Status by BMI M F All M F All M F All M F All
Nationally Representative Surveys
Davis and Gergen7 Internal ⱖ 85th Percentile 16.5 8.9 NR 16.9 13.4 NR − 0.4 − 4.5 NR − 2.4 − 33.5 NR
Goodman et al8 Must et al16 and ⱖ95th Percentile NR NR 8.1 NR NR 9.7 NR NR − 1.6 NR NR −16.5
Rosner et al17
Convenience Sample or Locally Based Surveys
Hauck et al13 Gelbach18 ⬎85th Percentile NR NR 28.4 NR NR 34.0 − 6.7 − 4.3 − 5.6 NR NR −16.5
Brener et al14 Kuczmarski et al19 ⱖ85th Percentile c NR NR 29.7 NR NR 47.4 NR NR − 17.7 NR NR −37.3
ⱖ85th to ⬍ 95th NR NR 14.8 NR NR 21.4 NR NR − 6.6 NR NR −30.8
Percentile d
ⱖ95th Percentile NR NR 14.9 NR NR 26.0 NR NR − 11.1 NR NR −42.7
Himes et al15 Kuczmarski et al19 ⱖ85th Percentile c 26.6 24.3 NR 32.3 33.0 NR − 5.7 − 8.7 NR − 20.9 − 26.4 NR
ⱖ95th Percentile 11.0 8.9 NR 14.8 11.8 NR − 3.8 − 2.9 NR − 25.7 − 24.6 NR

Abbreviations: BMI, body mass index; DM, directly measured; F, female; M, male; NR, not reported; SR, self-reported.
a Absolute error = SR−DM.
b Relative error = absolute error divided by DM.
c Overweight and risk of overweight.
d Risk of overweight.

Table 4. Sex-Specific Correlation Coefficients Between Self-report and Direct Measurement of Height, Weight, and BMI

Correlation Coefficient

Height Weight BMI

Source M F M F M F
Nationally Representative Surveys a
Davis and Gergen7 b 0.86 0.86 0.95 0.93 0.87 0.85
Strauss4 c 0.91 0.82 0.94 0.90 0.89 0.84
Himes and Faricy6 d 0.89 0.79 0.97 0.93 0.93 0.87
Convenience Sample or Locally Based Surveys e
Brooks-Gunn et al10 b NA 0.75 NA 0.98 NA NA
Shannon et al11 b 0.74 0.62 0.90 0.84 NA NA
Himes and Story12 c 0.91 0.71 0.96 0.91 0.90 0.80
Hauck et al13 c 0.83 0.62 0.95 0.90 0.88 0.79
Brener et al14 c 0.87 0.82 0.92 0.94 0.89 0.89
Himes et al15 c 0.90 0.80 0.96 0.94 0.89 0.85
All studies, range 0.74-0.91 0.62-0.86 0.90-0.97 0.84-0.98 0.87-0.93 0.79-0.89

Abbreviations: BMI, body mass index; F, female; M, male; NA, not available.
a Goodman et al8 not stratified by sex. Pearson coefficients for height (0.94), weight (0.95), and BMI (0.92) were all statistically significant (P ⬍ .001).
b Type of correlation coefficient not specified.
c Pearson correlation coefficients; Himes and Faricy6 refer to “r”; thus, may be Pearson coefficients.
d Intraclass correlation coefficients.
e Stewart9 not stratified by sex. Pearson coefficients for height= 0.95 and weight = 0.97. Significance was not reported.

rors. The wide variety of references and definitions of over-
weight used limited comparisons. For example, for the
prevalence of overweight based on the Centers for Dis-
ease Control and Prevention 2000 growth charts,19 for
the BMI cutoff of the 95th percentile or higher, absolute
errors ranged from −1.6% to −11.1% and relative errors,
from −16.5% to −42.7% (Table 3).
and BMI for 9 studies. The correlation coefficients for
height, weight, and BMI were lower for females than males,
with 3 exceptions. Including both males and females, cor-
relation coefficients for height ranged from 0.62 to 0.91;
for weight, 0.84 to 0.98; and for BMI, 0.79 to 0.93.
Age

CORRELATION COEFFICIENTS Four studies examined self-reported and directly mea-

sured correlation coefficients by age6,7,10,14 and in gen-
Table 4 shows sex-specific correlation coefficients be- eral found higher coefficients for the older adolescents,
tween self-reported and directly measured height, weight, yet none reported statistical differences. For example,

(REPRINTED) ARCH PEDIATR ADOLESC MED/ VOL 161 (NO. 12), DEC 2007 WWW.ARCHPEDIATRICS.COM
1158
Downloaded from www.archpediatrics.com at CDC-Information Center, on December 4, 2007
©2007 American Medical Association. All rights reserved.
 Table 5. Classification of Overweight by Self-reported vs Directly Measured Data: Sensitivity, Specificity, and Positive Predictive Value

%
Overweight Classification Sensitivity Specificity Positive Predictive Value
by BMI (Anthropometric
Source References) M F All M F All M F All
Nationally Representative Surveys
Davis and Gergen7 ⱖ 85th Percentile (internal) 76.0 59.0 a,b NR 96.0 99.0 NR NR NR NR
Goodman et al8 ⱖ 95th Percentile16,17 74.0 70.2 a,c NR 98.9 99.3 NR 88.4 88.0 NR
Convenience Sample or Locally Based Surveys
Hauck et al13 ⬎ 85th Percentile18 76.2 70.4 d 73.6 NR NR NR 93.9 81.4 88.2
Brener et al14 ⱖ 85th Percentile19 NR NR 60.5 NR NR 98.0 NR NR 96.5
ⱖ 95th Percentile19 NR NR 54.9 NR NR 99.2 NR NR 96.0

Abbreviations: BMI, body mass index; F, female; M, male; NR, not reported.
a Not statistically different from males.
b Reported no differences by age.
c Reported no differences by race/ethnicity.
d Did not conduct statistical testing of differences between males and females.

Himes and Faricy6 reported variability by age stratified
by sex and found lower values for 12-year-old boys and
girls for height (0.57 vs 0.62, respectively) and BMI (0.70
vs 0.76, respectively) compared with 13- to 16-year-old
boys and girls (both sexes, ⬎ 0.75 for height and ⬎ 0.83
for BMI), whereas Brener et al14 found high correlation
coefficients (range, 0.86-0.96) for height, weight, and BMI
for ninth- through 12th-grade adolescents.
Race/Ethnicity
tile or higher, Davis and Gergen7 reported no significant
difference in sensitivity among Mexican American indi-
viduals by sex or by age (12-14 years, 15-17 years, or 18-19
years) for males and females (values not reported). Using
a similar BMI cutoff for overweight, Hauck et al13 found
that sensitivity for Native American individuals was lower
among younger (12-14 years) than among older (15-19
years) males, yet higher among younger vs older females.
Specificity was greater than 95% for the 4 studies, regard-
less of the cutoff or stratification (Table 5).

Only Strauss4 and Brener et al14 examined correlation co-

efficients by race/ethnicity, stratified as white, black, and
Hispanic individuals; Brener et al14 also included an “other”
category. Strauss4 reported that BMI correlation coeffi-
cients for black individuals were significantly lower than
those for white or Hispanic individuals (0.79 vs 0.89 and
0.85, respectively; P⬍.05 for black vs white or black vs His-
panic individuals). Brener et al14 did not carry out signifi-
cance testing but reported lower height correlations for His-
panic individuals than for individuals from the 3 other racial
or ethnic groups (0.77 vs ⱖ0.88); BMI correlation coeffi-
cients for the “other” category were lower (0.79 vs ⱖ0.89).
SENSITIVITY, SPECIFICITY, AND POSITIVE
PREDICTIVE VALUE
Two nationally representative studies7,8 and 2 conve-
nience sample studies13,14 examined the validity of using
self-reported height and weight-calculated BMI to iden-
tify overweight (Table 5). The values for sensitivity
among these 4 studies suggest that only between 55% and
75% of those adolescents who were truly overweight
would be classified as overweight if their self-reported
weight and height were used to determine BMI.
Three studies reported sensitivity by sex7,8,13 (Table 5),
one reported by race/ethnicity,8 and another reported by
age13; no consistent patterns emerged for these sub-
groups. Using a BMI cutoff of the 95th percentile or higher,
Goodman et al8 reported no difference in sensitivity by sex
or race/ethnicity. Using a BMI cutoff of the 85th percen-
COMMENT
Our review identified 4 consistent patterns in bias across
the studies when self-reported data were used to classify
overweight among adolescents: (1) the prevalence of over-
weight was consistently underestimated by self-reported
data, although the relative error ranged widely, from −2.4%
to −42.7%; (2) sensitivity data showed that 25% to 45% of
those overweight would be missed if self-reported data were
used; (3) females underestimated their weight and BMI more
than males; and (4) overweight youth underestimated their
weight and BMI more than normal-weight youth.
We identified 3 additional issues, relevant to surveil-
lance, targeting, program evaluation, and research, in need
of further examination: (1) lack of or inconsistent infor-
mation on other subgroup differences; (2) lack of infor-
mation on whether biases are consistent over time; and
(3) lack of consensus regarding an acceptable level of dif-
ference between self-reported and directly measured
height and weight data and how this might vary depend-
ing on purpose of use.
The sensitivity data show that self-reported height and
weight are relatively weak estimates of directly mea-
sured values for categorizing overweight status, which
is similar or slightly worse than findings for adults. In
comparison with adolescent data, the sensitivity data for
adult US men and women aged 20 to 59 years are gen-
erally at the upper end of the range or higher than that
for adolescents (Bowlin et al,20 77% and 72%, respec-

(REPRINTED) ARCH PEDIATR ADOLESC MED/ VOL 161 (NO. 12), DEC 2007 WWW.ARCHPEDIATRICS.COM
1159
Downloaded from www.archpediatrics.com at CDC-Information Center, on December 4, 2007
©2007 American Medical Association. All rights reserved.
tively; Kuczmarski et al,21 86%-95%; and Nieto-Garcı́a
et al,22 74%). Furthermore, the adult data show evi-
dence of lower values among certain subgroups. For ex-
ample, Bowlin et al20 found that for 20- to 29-year-old
women, sensitivity was 60%, whereas Nieto-Garcı́a et al22
found that for 60- to 69-year-olds, sensitivity was 64%.
In contrast to sensitivity analysis, we found that the cor-
relation coefficients between self-reported and directly mea-
sured height, weight, and BMI were relatively high. How-
ever, correlation coefficients are strongly affected by extreme
values, and they do not examine the accuracy of the clas-
sification of overweight status, our goal.
Additional research needs are to assess (1) the accu-
racy of self-reported data by order of self-reported and di-
rectly measured data collection; (2) whether advance knowl-
edge of directly measured data influences self-reported data;
(3) whether other potentially important factors not con-
sidered in this review, such as mental health status, socio-
economic status, or physical activity status, may also affect
bias of self-reported data; and (4) comparisons of data based
on standard accepted definitions of overweight status and
anthropometric reference. One example of an additional
important factor could be depression. Overweight people
are more likely to be depressed,23 and we found greater un-
derestimates of weight among overweight adolescents. If
depression is independently associated with underreport-
ing of overweight status, then self-reported bias could also
vary by level of depression.
Our findings raise questions about the practical im-
plications for the use of self-reported data. The preva-
lence of overweight is assessed in populations for a va-
riety of purposes, including identifying the problem,
monitoring trends over time, program targeting, pro-
gram evaluation, research, and advocating for funds for
interventions. The utility of self-reported height and
weight data depends on the purpose for which they are
used. Self-reported data are almost always easier to col-
lect and provide documentation of the problem of over-
weight even if the data are underestimates.
Prevalence based on self-reported data can be useful
for monitoring overweight over time if the magnitude of
biases remains constant over time. For short-term follow-
up, the assumption of constant bias is probably reason-
able, but no studies examined whether there have been
changes in reporting bias over long periods. The fact that
the degree of underreporting of prevalence varies widely
among subgroups, from being almost undetectable in
some studies (eg, males in Davis and Gergen7) to miss-
ing almost half of the problem in others,14 certainly raises
the concern that reporting biases among subgroups are
not constant across populations and settings.
The utility of self-reported data for targeting pro-
grams toward those at greatest risk depends on whether
different population subgroups underreport BMI simi-
larly. Females tend to underestimate their weight more
than males, potentially leading to a false conclusion that
males should be targeted more directly for overweight
prevention programs. Heavier youth underestimate weight
more than normal-weight youth, leading to an underes-
timate of the size of the target group. Other subgroup dif-
ferences (in particular age, race/ethnicity, and missing
data) have not been adequately studied, so it is impos-
(REPRINTED) ARCH PEDIATR ADOLESC MED/ VOL 161 (NO. 12), DEC 2007
1160
Downloaded from www.archpediatrics.com at CDC-Information Center, on December 4, 2007
©2007 American Medical Association. All rights reserved.
sible to know whether targeting based on self-reported
data are appropriate.
Program evaluation of the effectiveness of obesity pre-
vention and treatment interventions typically occurs over
a relatively short time, such that it may be reasonable to
expect that the degree of bias associated with self-
reported data will not change between baseline and fol-
low-up assessments. However, this expectation prob-
ably depends on the nature of the intervention, and one
needs to think about the potential for bias based on the
content of the intervention. For example, an interven-
tion focused on BMI awareness or weekly monitoring of
a person’s own weight would probably change that per-
son’s knowledge of his or her directly measured weight
and his or her ability to accurately report it. In such a
case, evaluation might show increases in overweight be-
cause of the improved reporting even if the prevalence
of overweight went down. Alternatively, there might be
other factors associated with underreporting, such as the
social desirability to be thin or poor body image, so that
some participants would still underreport weight de-
spite knowing their directly measured weight. Because
the reasons for underreporting are multiple and com-
plex, it is difficult to anticipate how an intervention might
affect self-reported height and weight data.
Research often attempts to identify factors associated
with overweight and nonoverweight as a dichotomous
variable. To the extent that self-reported data misclas-
sify adolescent overweight status and that this misclas-
sification is unrelated to the factors of interest, research
findings will be biased toward the null hypothesis. This
bias implies that important factors associated with over-
weight will be overlooked. Where significant associa-
tions are found, however, one can expect that a real as-
sociation does exist.
Where the estimated prevalence of overweight is high,
the existence of self-reported height and weight data can
be useful in advocating for programs and policies that
address overweight and may be a prerequisite for secur-
ing funds to address the problem. However, policy mak-
ers and funders may be unimpressed with understated
prevalence numbers and give inadequate attention to the
problem, particularly in the long-term as the newness of
the obesity epidemic passes.
One example of frequently used self-reported height
and weight data is the Center for Disease Control and
Prevention Youth Risk Behavior Surveillance System.24
This surveillance system regularly collects self-reported
height and weight data and is often the only source of
information on the prevalence of overweight among youth
attending high school in a state. Without feasible alter-
native mechanisms to collect directly measured data, these
self-reported data fill an important gap and are neces-
sary for documenting the problem, advocating for funds
for interventions, and monitoring trends over time.
Finally, this review has both strengths and limita-
tions. To the best of our knowledge, this is the first lit-
erature review of this topic for adolescents. We identi-
fied 4 consistent biases and described the implications
of these findings for using self-reported data. As for weak-
nesses, the lack of comparability among studies for cut-
offs and references used to define overweight status, mea-
WWW.ARCHPEDIATRICS.COM
 surement protocols, intervals between self-reported and
directly measured data, and data stratification and sta-
tistical analyses made it impossible to produce overall
summary results. Second, the lack of adequate data in sev-
eral studies prevented us from examining additional
needed subgroup analyses and other potential impor-
tant sources of bias, such as consistency of bias over time.
CONCLUSIONS
Researchers and public health practitioners should care-
fully consider the caveats described in this article when
interpreting results based on self-reported data. Because
of the limitations of relying on self-reported height and
weight data for surveillance of overweight, public health
practitioners should explore the feasibility of collecting
directly measured heights and weights at the state and
local levels for surveillance purposes. Historically, states
collected directly measured height and weight data for a
variety of purposes, particularly for screening. Arkansas
measures the height and weight of children in schools
for screening and surveillance.25 Missouri collected height
and weight data during physical education classes as part
of the curriculum for surveillance purposes.26 In New
York, education law requires that children have health
appraisals, including measurement of height and weight,
at school entrance, as well as when students are in the
second, fourth, seventh, and 10th grades. The New York
Department of Health and some school districts are in
discussions as to how to systematically collect and re-
port these data to the Department of Health (Barbara A.
Dennison, MD, written communication, April 2006). Ex-
periences with these state programs, as well as other pro-
grams, related to data quality, representativeness, and cost
could help determine the feasibility of developing state
and local systems for collecting directly measured height
and weight data for surveillance purposes.
Self-reported data are valuable if the only source of
data. However, self-reported data underestimate over-
weight prevalence and there is bias by sex and weight sta-
tus. Lower sensitivities of self-reported data indicate that
one-fourth to one-half of those overweight would be
missed. Other potential biases in self-reported data, such
as differences across subgroups and inconsistency over
time, need further clarification. The feasibility of collect-
ing directly measured height and weight data on a state
or community level should be explored because directly
measured data are more accurate.
Accepted for Publication: May 24, 2007.
Correspondence: Bettylou Sherry, PhD, RD, Division of
Nutrition and Physical Activity, Centers for Disease Con-
trol and Prevention, 4770 Buford Hwy, MS K25, At-
lanta, GA 30341 (bsherry@cdc.gov).
Author Contributions: Study concept and design: Sherry,
Jefferds, and Grummer-Strawn. Acquisition of data: Sherry
and Jefferds. Analysis and interpretation of data: Sherry,
Jefferds, and Grummer-Strawn. Drafting of the manu-
script: Sherry and Jefferds. Critical revision of the manu-
script for important intellectual content: Sherry, Jefferds,
and Grummer-Strawn. Statistical analysis: Sherry and
(REPRINTED) ARCH PEDIATR ADOLESC MED/ VOL 161 (NO. 12), DEC 2007
1161
Downloaded from www.archpediatrics.com at CDC-Information Center, on December 4, 2007
©2007 American Medical Association. All rights reserved.
View publication stats
Grummer-Strawn. Administrative, technical, and mate-
rial support: Jefferds. Study supervision: Sherry and
Grummer-Strawn.
Financial Disclosure: None reported.
Additional Contributions: Nancy D. Brener, PhD, Laura
K. Kann, PhD, and Deborah A. Galuska, PhD, reviewed
the manuscript.
REFERENCES
1. Institute of Medicine. Preventing Childhood Obesity: Health in Balance. Wash-
ington, DC: National Academies Press; 2005.
2. Valenti L, Charles J, Britt H. BMI of Australian general practice patients. Aust Fam
Physician. 2006;35(8):570-571.
3. Sen B. Frequency of family dinner and adolescent body weight status: evidence
from the National Longitudinal Survey of Youth, 1997. Obesity (Silver Spring).
2006;14(12):2266-2276.
4. Strauss RS. Comparison of measured and self-reported weight and height in a
cross-sectional sample of young adolescents. Int J Obes Relat Metab Disord.
1999;23(8):904-908.
5. Fortenberry JD. Reliability of adolescents’ reports of height and weight. J Ado-
lesc Health. 1992;13(2):114-117.
6. Himes JH, Faricy A. Validity and reliability of self-reported stature and weight of
US adolescents. Am J Hum Biol. 2001;13(2):255-260.
7. Davis H, Gergen PJ. The weights and heights of Mexican-American adoles-
cents: the accuracy of self-reports. Am J Public Health. 1994;84(3):459-462.
8. Goodman E, Hinden BR, Khandelwal S. Accuracy of teen and parental reports of
obesity and body mass index. Pediatrics. 2000;106(1)(pt 1):52-58.
9. Stewart AL. The reliability and validity of self-reported weight and height. J Chronic
Dis. 1982;35(4):295-309.
10. Brooks-Gunn J, Warren MP, Rosso J, Gargiulo J. Validity of self-report mea-
sures of girls’ pubertal status. Child Dev. 1987;58(3):829-841.
11. Shannon B, Smiciklas-Wright H, Wang M. Inaccuracies in self-reported weights
and heights of a sample of sixth-grade children. J Am Diet Assoc. 1991;91
(6):675-678.
12. Himes JH, Story M. Validity of self-reported weight and stature of American In-
dian youth. J Adolesc Health. 1992;13(2):118-120.
13. Hauck FR, White L, Cao G, Woolf N, Strauss K. Inaccuracy of self-reported weights
and heights among American Indian adolescents. Ann Epidemiol. 1995;5(5):
386-392.
14. Brener ND, McManus T, Galuska D, Clowry R, Wechsler H. Reliability and valid-
ity of self-reported height and weight among high school students. J Adolesc
Health. 2003;32(4):281-287.
15. Himes JH, Hannan P, Wall M, Neumark-Sztainer D. Factors associated with er-
rors in self-reports of stature, weight, and body mass index in Minnesota
adolescents. Ann Epidemiol. 2005;15(4):272-278.
16. Must A, Dallal GE, Dietz WH. Reference data for obesity: 85th and 95th percen-
tiles of body mass index (wt/ht2) and triceps skinfold thickness. Am J Clin Nutr.
1991;53(4):839-846.
17. Rosner B, Prineas R, Loggie J, Daniels SR. Percentiles for body mass index in
US children 5 to 17 years of age. J Pediatr. 1998;132(2):211-222.
18. Gelbach S. Interpreting the Medical Literature: Practical Epidemiology for Clinicians.
New York, NY: MacMillan; 1988:139-160.
19. Kuczmarski RJ, Ogden CL, Grummer-Strawn LM, et al. CDC growth charts: United
States. Adv Data. 2000;(314):1-27.
20. Bowlin SJ, Morrill BD, Nafziger AN, Jenkins PL, Lewis C, Pearson TA. Validity of
cardiovascular disease risk factors assessed by telephone survey: the Behav-
ioral Risk Factor Survey. J Clin Epidemiol. 1993;46(6):561-571.
21. Kuczmarski MF, Kuczmarski R, Najjar M. 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. J Am Diet Assoc. 2001;
101(1):28-34.
22. Nieto-Garcı́a FJ, Bush T, Keyl P. Body mass definitions of obesity: sensitivity and
specificity using self-reported weight and height. Epidemiology. 1990;1(2):
146-152.
23. Onyike CU, Crum RM, Lee HB, Lyketsos CG, Eaton WW. Is obesity associated
with major depression? Results from the Third National Health and Nutrition Ex-
amination Survey. Am J Epidemiol. 2003;158(12):1139-1147.
24. YRBSS: Youth Risk Behavior Surveillance System. CDC Web site. http://www
.cdc.gov/healthyyouth/yrbs/index.htm. Accessed December 20, 2005.
25. Centers for Disease Control and Prevention (CDC). Overweight among students
in grades K-12: Arkansas, 2003-04 and 2004-05 school years. MMWR Morb Mor-
tal Wkly Rep. 2006;55(1):5-8.
26. Missouri Coordinated School Health Coalition. Promoting healthy weight in Mis-
souri’s children: a guide for schools, families and communities. State of Mis-
souri Department of Health and Senior Services Web site. http://www.dhss.mo
.gov/Nutrition_School/Publications.html. Accessed December 20, 2005.
WWW.ARCHPEDIATRICS.COM