Correction of the self-reported BMI in a teenage population

International Journal of Obesity (1998) 22, 673±677
ß 1998 Stockton Press All rights reserved 0307±0565/98 $12.00
http://www.stockton-press.co.uk/ijo
Correction of the self-reported BMI in a teenage
population
M Giacchi1, R Mattei2 and S Rossi1
1
Dpt. Scienze Medico-Legali e Socio-Sanitarie and 2Cattedra di Fisiologia della Nutrizione, Ist. Scienze Chirurgiche,
UniversitaÁ di Siena, Italy
OBJECTIVE: To investigate the issue of systematic bias in self-reported weight and height, and produce a simple
procedure which can be used to correct reporting bias.
DESIGN: Cross-sectional, with self-reported questionnaires.
SUBJECTS: A sub-sample (n 143) of secondary school students in Siena, Italy, taken from the Food Behaviour
Survey (sample size, n 779).
RESULTS: In the teenage sub-sample, both males and females under-reported their weight and over-reported their
height, such that underestimation of the overweight prevalence was in the order of about 8% for both genders. For
both weight and height, the correlations between self-reported and measured values were over 0.90. Conversion
factors were derived to correct the reported body mass index (BMI) distribution by adjusting the percentages of
erroneously classi®ed subjects in the four BMI categories.
CONCLUSION: High correlation coef®cients (r 0.75), showing a systematic tendency for erroneous self-reporting of
a `slim-body shape', justify the use of conversion factors (measured/self-reported) to correct BMI distributions
calculated from self-reported values.
Keywords: self-reported; BMI; teenagers; conversion factors
Introduction
Methods
The accuracy of self-reported data in describing the
overweight prevalence is very important. There is a
general feeling that teenage populations under-report
their weight and over-report their height; in particular,
females are more likely to under-report their weight,
compared to males, and males are more likely to
overestimate their height.1±3
Recent studies show that self-reported height and
weight data from teenage populations, should be used
with caution and suggest that further investigations
should be conducted on the correction of reporting
bias.1,4 Due to important implications of using selfreported data on the determination of body mass index
(BMI), we performed the present study.
The present study suggests a simple and inexpensive procedure for correcting reporting bias. This
method can be especially useful to those groups
which operate in the health promotion ®eld, whose
objectives include community preventive interventions on speci®c population groups, rather than objectives involving data which can be generalized, such as
those regarding prevalence.
The data used in this study were taken from the Food
Behaviour Survey (University of Siena, 1995 ±1996),
which collected information on diet and lifestyle from
a sample of teenage students (age: 15±17 y) in Siena.
The original study began with a co-operation
between the University of Siena and Sienese secondary schools to add food education to existing didactic
programmes. The aim of the original study was
twofold: cognitive and educational. In the ®rst
phase, data relating to students' food habits were
collected. The results were subsequently shown and
discussed with the students during some food education lessons, carried out by GSAA group experts (a
dietitian, a psychologist and an epidemiologist).
The study sample consisted of 779 secondary
school students (325 female and 454 male). The
students were asked to complete a questionnaire,
which collected data on a wide number of variables:
age, parents' occupational role and educational history, height, weight, dieting habits, food and drink
preferences, perception of self body image, alcohol
consumption and sport activities, and a weekly dietary
diary.
The general sample was obtained as follows:
Correspondence: Dott. Mariano Giacchi, Dpt. Scienze MedicoLegali e Socio-Sanitarie, Policlinico `Le Scotte', 53100 Siena,
Italy.
Received 11 September 1997; revised 26 February 1998;
accepted 3 March 1998
1) Secondary schools in Siena were divided into two
main groups according to different studies (High
School and Technical School); both groups were
Correction of self-reported BMI
M Giacchi et al
674
homogeneous in terms of the students' township
and their family's cultural background: most of the
high school students were from Siena and their
families had a mid±high level of education, whilst
the majority of those who attended technical
schools were commuters belonging to families
with a mid±low level of education. One school
was randomly selected from each group.
2) At each school, only a few classes were involved;
as such, it was impossible to proceed with random
sampling. For didactic continuity purposes, we
only included the food education programme for
those classes (779 students) in which a Science
course was scheduled for the current semester.
In the questionnaire, the students were asked to
indicate their height in cm and their weight in kg to
one decimal point.
Considering the aims of the study, the teenagers
were advised to answer the questions as accurately as
possible. It was predictable, at least as far as weight
and height were concerned, that the answers would
not be totally accurate, since they were in¯uenced by
different variables, ®rstly by the type of instrument
used.
Most of the values re¯ected in the questionnaires
were expressed in whole numbers (that is, 50 kg,
71 kg, 187 cm; otherwise 50.5 kg, 71.5 kg, 187.5 cm,
etc.). The reported measures were not rounded during
data processing.
To improve the accuracy of the BMI, calculated
from self-reported weight and height, we used data
from a simple random sub-sample including 143
students (56 female and 87 male) belonging to the
study sample. Subjects from this sub-sample were
visited the following week and their weight and
height measured. Students were told a few days
before that they would undergo a medical check-up.
Due to missing weight and/or height data on the
questionnaire, 10 subjects (®ve male and ®ve
female) were excluded from the study. Weight and
height measurements were carried out by the dietitian
and assistants. The technique used for the measurements followed ISS criteria.11 Subjects wearing
no clothes or shoes were weighed by means of a
`platform scale' and their height measured with a
`stadiometer' after correct positioning of the body.
During measurement, weight values were rounded to
the nearest 0.1 kg and height values to the nearest cm.
Data for females and males were analysed separately.
We ®rst studied the predictive validity of selfreported BMI. Then, we attempted to correct the
BMI distributions. According to the Recommended
Dietary Allowances (L.A.R.N.-1996)5 indications, the
BMI categories were the following:
Underweight: BMI < 18.5
Normal weight: BMI 18.5±25.0
Over weight: BMI 25.01±29.99
Obesity: BMI 30
The data analysis consisted of cross-tabulations,
paired t-tests and regression analysis, all performed
using BMDP software.6
Results
On average, both males and females under-reported
their weight by about 2.0 kg (ÿ1.8 kg for males and
ÿ1.9 kg for females), whilst they over-reported their
height by an average of 2.3 cm and 0.8 cm, respectively (Table 1).
For males and females, both height and weight
differences between self-reported and measured
values were statistically signi®cant (paired t-test,
P < 0.001).
The differences in self-estimation of body size,
described above, in¯uenced the mean BMI values
calculated from the reported data. The females'
mean BMI was underestimated by 1.3 kg/m2 (or
ÿ5.8%) and the males' BMI by 0.8 kg/m2 (or
ÿ3.5%) (Table 1). If we compare the percentage
distribution of self-reported BMI vs measured BMI,
we can observe that the size of the overweight group
is underestimated by self-reported data. The degree of
the overweight prevalence underestimation accounted
for nearly 8% (11.8% measured vs 4.0% self-reported)
for the females and nearly 7% (17.1% measured vs
9.8% self-reported) for the males (Table 2). With
regard to the males' self-reported distribution, the
overweight and obesity prevalence values in the
Table 1 Measured and self-reported mean values of weightb, heightb, and body mass index (BMI) and mean per cent differences
(D%)a in the sub-sample (n 133)
Females (n 51)
Parameters
Weight (kg)*
Height (cm)*
BMI (kg/m2)
Males (n 82)
Measured
mean s.d.
Self-reported
mean s.d.
D%
Measured
mean s.d.
Self-reported
mean s.d.
D%
60.3 (8.8)
163.8 (5.8)
22.4 (2.9)
58.4 7.9
166.1 6.1
21.1 2.3
ÿ3.1%
1.4%
ÿ5.8%
69.5 12.1
174.8 7.4
22.7 3.8
67.7 12.1
175.6 7.2
21.9 3.5
ÿ2.6%
0.5%
ÿ3.5%
* Statistically signi®cant differences between measured and self-reported mean values for both sex (paired t-test, P < 0.001).
(Self-reported 7 Measured)/Measured 100.
b
For females, the ranges are: 48±89 (weight measured); 45±85 (weight reported); 148±174 (height measured); 150±174 (height
reported). For males, the ranges are: 45±114 (weight measured); 45±115 (weight reported); 158±202 (height measured); 160±201
(weight reported).
a
Correction of self-reported BMI
M Giacchi et al
675
Table 2 Measured and self-reported body mass index (BMI) distributions (percentage values) according to gender in the
sub-sample (n 133)
Females (n 51)
a
BMI categories
Self-reported
Measured
Self-reported
Measured
7.8%
88.2%
4.0%
±
5.8%
82.4%
11.8%
±
13.4%
74.4%
9.8%
2.4%
6.1%
72.0%
17.1%
4.8%
Underweight (BMI < 18.5)
Normal weight (BMI 18.5±25.0)
Overweight (BMI 25.01±29.99)
Obesity (BMI 30.0)
a
Males (n 82)
According to LARN 1996.7
subsample were rather high with respect to those
belonging to the study sample. In fact, by the
random selection, 2/2 obese cases and 8/20 overweight cases were chosen.
Even if the magnitude of the differences between
self-reported and measured data is small, the effect of
these biases in reporting may heavily affect the BMI
distribution.7,8 Calculation of the BMI, results both in
ampli®cation of biases and misclassi®cation of subjects into incorrect BMI categories. We utilized Kuskowska-Wolk's statistical model,7 in order to show
additional empirical evidence for reporting bias in our
population. Of interest, is the utility of KuskowskaWolk's model to underline the effect of reporting bias
in terms of height. Our aim was, in fact, to point out
the importance of not neglecting even small differences between measured and self-reported values.
Analysing the BMI data according to the statistical
models (separate regression models for males and
females) proposed by Kuskowska-Wolk7:
Hm2 BMIs aHm2 bHm2 BMIm e
where a and b are regression coef®cients, H2m the
measured height, BMIm the measured BMI and BMIs
the self-reported BMI, we found that the systematic
error can result in an underestimation of the true value
of up to 1 kg/m2 in most overweight females, which
corresponds to about 3 kg for an average female
height (164 cm). The underestimation reaches 2 kg/m2
among the underweight females corresponding to
5.5 kg for an average female height. Moreover,
among the extremely obese males, the systematic
error can correspond to an underestimation of about
2 kg for an average male height (175 cm). These
results can be obtained dividing the above regression
equation by H2m. Subsequently, the linear relation is:
BMIs a bBMIm eHmÿ2
which differs with an ordinary regression model in
that the error term depends on the height.
Pearson's correlation between BMI calculated from
reported and measured weight was 0.96 for males and
0.93 for females (Table 3). For height, the correlation
coef®cients for males and females were 0.93 and 0.97,
respectively.
High correlation coef®cients justify the use of
conversion factors (measured/self-reported) to correct
Table 3 Pearson correlations between self-reported and
measured values
Parameters
Weight (kg)
Height (cm)
Females* (n 51)
Males** (n 82)
0.96
0.93
0.93
0.97
* For n 51, correlations are statistically signi®cant (P < 0.001).
** For n 82, correlations are statistically signi®cant (P < 0.001).
Table 4 Conversion factorsa
Parameters
Weight (kg)
Height (cm)
a
Females (n 51)
Males (n 82)
1.032
0.988
1.026
0.996
Conversion factor Measured/Self-reported.
BMI distributions calculated from self-reported values
(Table 4). They have been calculated as a ratio
between the measured and self-reported mean values
and can be used to correct the reported BMI distribution by adjusting the percentages of erroneously
classi®ed subjects into the four BMI categories. The
number of students who were classi®ed as obese,
overweight, normal weight and underweight by selfreported BMI were compared to the number corrected
by conversion factors (Table 5). Both overweight
males and females were incorrectly classi®ed in
about half of all cases, as those subjects actually
belonged to the normal weight group. After correction, the percentage of overweight females increased
from 1.6% to 3.1% and that of overweight males from
4.4% to 7.9%. Among the obese group, the percentage
increased from 0.3% to 0.6% for females and from
0.5% to 0.9% for males. Again, among females the
number of underweight girls decreased from 91 to 30
(28% vs 9.2%) and, among males, from 68 to 43
subjects (15% vs 9.5%).
Discussion
Our data only partly con®rm the results found by other
studies, which show that females are more likely to
under-report weight compared to males and that males
are more likely to overestimate their height.
Correction of self-reported BMI
M Giacchi et al
676
Table 5 Body mass index (BMI) distributions (not corrected and corrected values) according to gender in the study sample n 779
Females
Males
Not corrected
Corrected
Not corrected
Corrected
BMI categories
Freq.
%
Freq.
%
Freq.
%
Freq.
%
Underweight (BMI < 18.5)
Normal weight (BMI 18.5±25.0)
Overweight (BMI 25.01±29.99)
Obesity (BMI 30.0)
Total
91
228
5
1
325
28.0
70.1
1.6
0.3
100.0
30
283
10
2
325
9.2
87.1
3.1
0.6
100.0
68
364
20
2
454
15.0
80.1
4.4
0.5
100.0
43
371
36
4
454
9.5
81.7
7.9
0.9
100.0
The difference regards female students who, in our
study, under-reported their weight more often than
males, but overestimated their height to a larger
extent than males. On average, however, the magnitude of misreporting was rather small. Under-reporting (3.1%) was found in weight measures among
women, whilst among men the misreporting
accounted for 2.6%. In height measures, positive
discrepancies of 1.4% in the females and 0.5% in
males were found. A more consistent underestimation of the BMI values (ÿ5.8% in females and
ÿ3.5% in males) (Table 1) resulted from the misreporting of both height and weight.
Despite the small differences between self-reported
and measured data, the consequences of an erroneous
BMI distribution can be dangerous. In our study, for
example, comparing the self-reported mean values
with the measured ones we observed that the overweight percentages rose from 4.0% to 11.8% for the
females and from 9.8% to 17.1% for the males (Table
2). This underestimation of the true size of speci®c
groups may affect the possibility of their inclusion in
intervention activities.
Very high correlation coef®cients between reported
and measured data (r > 0.90) underlines a homogeneous tendency in height and weight reporting
amongst teenagers (Table 3).
Having veri®ed that the differences between measured and self-reported data were statistically signi®cant, we considered the high correlation coef®cients
between measured and self-reported data (r 0.75)9
as good indicators of a wide tendency to strive for a
`slim body shape'. On the basis of these considerations, we calculated the conversion factors. These
could have derived from the random sub-sample data
as a ratio between the measured and self-reported mean
values, and could be used to correct data from the study
sample when the differences between measured and
self-reported data are statistically signi®cant.
Before correction, the number of students who
would have been classi®ed as obese or overweight,
corresponded to half of the actual number for both
genders, while nearly 70% of subjects from the underweight group would have been incorrectly classi®ed
in the female group and 37% in the male group (Table
5). In the female group, in particular, it is important to
notice the reduction in the number of underweight
cases resulting from the re-classi®cation of a considerable percentage of them to the normal weight class.
In fact, despite the contemporary overweight growth,
the percentage of normal weight students increases
from 70.1% to 87.1% after correction.
It would be interesting to compare corrected BMI
data with measured data from other studies of this age
group in Italy. However, to date, similar Italian
studies have only been carried out on limited numbers
of adolescents. The same growth charts used in Italy,
as reference standards, have been obtained from
studies monitoring height and weight of children and
adolescents (2±18 y) in France, Britain, Switzerland
and the USA. A recent article by Zoppi et al10
presented the growth curve for weight and height in
Italian children and adolescents (2±18 y); although the
authors of this article consider their data to be suf®ciently representative of the Italian population, we
found differences between our data and theirs with
regard to the mean measured height and weight. In
particular, our measured BMI mean values are higher
for all ages (15, 16, 17 y) for both males and females
by nearly one unit.
The proposed procedure is simple and economical.
In particular, it can be widely used because it is quick
and easy to handle; complicated calculations are not
required. Moreover, information on inaccuracy, deriving from a representative sub-sample, makes it possible to avoid substituting self-reporting with more
expensive procedures. Obviously, the correct subsample size depends on the population's variability
and the survey characteristics (objectives, time and
budget). Even if it is dif®cult to de®ne general rules,
about 30 cases represent the minimum size for a
statistical analysis. Nevertheless, many researchers
believe that 100 cases constitute an appropriate
sample size.13
Although this methodology is easy to perform,
speci®c correction factors need to be calculated,
following the above procedure for different populations. In fact, our data and the derived conversion
factors are not necessarily descriptive of any single
racial, social, economic or nutritional group.
Correction of self-reported BMI
M Giacchi et al
Conclusion
The results of this study could be extended to different
and/or wider populations. The reliability of height and
weight reports should be investigated and, during
planning stages, particular attention should be paid
to the sampling for a correct selection of the subsample according to different populations and various
characteristics of the surveys.
Acknowledgement
Research developed with the ®nancial support of the
University of Siena ± MURST ± 60% ± 1996.
References
1 Crawley HF, Portides G. Self-reported versus measured
height, weight and body mass index amongst 16±17 year old
British teenagers. Int J Obes 1995; 19: 579±584.
2 Fortenberry JD. Reliability of adolescents' reports of height
and weight. J Adolesc Health 1992; 13: 114±117.
3 Tienboon P, Wahlquist ML, Rutishauser IHE. Self-reported
weight and height in adolescents and their parents. J Adolesc
Health 1992; 13: 528±532.
4 Roberts RJ. Can self-reported data accurately describe the
prevalence of overweight? Public Health 1995; 109: 275±284.
5 SocietaÁ Italiana di Nutrizione Umana (SINU). Livelli di
assunzione raccomandata di energia e nutrienti per la popolazione italiana (LARN). Rev.96. Zest: Rome, 1996.
6 Dixon WJ (ed). BMDP Statistical Software Manual. University of California Press: Los Angeles, 1992.
7 Kuskowska-Wolk A, Karlsson P, Stolt M, Rossner S. The
predictive validity of body mass index on self-reported weight
and height. Int J Obes 1989; 13: 441±453.
8 Tormo-Diaz MJ, Cirera L, Navarro C. The prevalence of
obesity estimated from the self-reporting of weight and
height in Spain: a systematic error. Med Clin (Barc) 1995;
104: 596±597.
9 Colton T. Statistica in medicina (Ital. Transl. Marubini E). (1st
edn). Piccin: Padova, 1979.
10 Zoppi G, Bressan F, Luciano A. Height and weight reference
charts for children aged 2±18 years from Verona, Italy. Eur J
Clin Nutr 1996; 50: 462±468.
11 Giampaoli S, Romiti A, Menotti A. Methodological notes for
some anthropometric measures of epidemiological interest.
Istisan 88/31: Rome, 1988.
12 Bailey KD. Metodi della ricerca sociale (Ital. Transl. Rossi
M.), Il Mulino: Bologna, 1991.
677

Download Report