International Journal of Obesity (2000) 24, 652±657

ß 2000 Macmillan Publisher Ltd All rights reserved 0307±0565/00 $15.00

www.nature.com/ijo

Reliability of anthropometric measurements in

overweight and lean subjects: consequences for
correlations between anthropometric and other
variables

K Nordhamn1, E SoÈdergren1, E Olsson1, B KarlstroÈm1, B Vessby1 and L Berglund1*

1
Department of Public Health and Caring Sciences, Section for Geriatrics=Clinical Nutrition Research, Uppsala University, Sweden

OBJECTIVE: To estimate the reliability of anthropometric measurements in overweight and lean subjects, and to
examine the in¯uence of this reliability on correlations to other variables, since low reliability leads to under-
estimation of correlations.
DESIGN: Replicate measurements by two observers in 26 overweight and 25 lean subjects measured at two
occasions.
MEASUREMENTS: Sagittal abdominal diameter (SAD), waist circumference (waist), waist-to-hip ratio (W=H) and
skinfold measurements.
RESULTS: Intra-class correlation coef®cients (ICCs) for SAD and waist were higher than for W=H (0.98 vs 0.90,
P < 0.001, and 0.97 vs 0.90, P ˆ 0.001, respectively). For waist, the ICC was lower for overweight than for lean subjects
(0.85 vs 0.95, P ˆ 0.030), but the ICC values were comparable for SAD and W=H (0.92 vs 0.95 and 0.78 vs 0.83,
respectively). Intra-observer variations (IOV) for SAD and waist were lower than for W=H (coef®cients of variation;
1.6%, 1.4% and 2.3%, respectively), as were intra-subject variations (ISV) (2.7%, 3.0% and 3.4%, respectively). ICC
values ranged from 0.84 to 0.93 and were lower for overweight than for lean subjects for biceps, subscapular and
umbilical skinfolds (P ˆ 0.031, P < 0.001 and P ˆ 0.048, respectively). Coef®cients of variations for skinfold measure-
ments ranged between 7.3% and 16.0% for IOV and between 14.9% and 20.8% for ISV.
CONCLUSIONS: The low ICC values imply that correlations can be underestimated in overweight groups. We propose
that, because of their higher reliability, SAD and waist have a higher predictive capacity for cardiovascular risk than
W=H. SAD is the only measurement with high reliability in both weight groups and its use is recommended.
International Journal of Obesity (2000) 24, 652±657

Keywords: reliability; anthropometry; abdominal fat distribution; correlations; lean vs overweight

Introduction are simple, inexpensive and commonly used methods

for indirect assessment of the body fat distribution.4,5
W=H has been the most widely used method for
Epidemiological studies have shown that excess body routine measurements of abdominal fat; today, how-
fat, particularly abdominal fat, is related to the devel- ever, there seems to be a shift towards the use of
opment of metabolic complications such as athero- SAD4 or waist circumference alone.6 Recent studies
sclerosis and type 2 diabetes.1 ± 3 Reliable methods for have shown that SAD and waist circumference appear
measurement of body fat and fat distribution are to be more strongly related to metabolic variables than
therefore of importance. Methods for direct assess- W=H.4,7 ± 11 Skinfold thickness was originally used
ment of abdominal fat include ultrasound, dual energy with prediction equations presented by Durnin and
X-ray absorptiometry, magnetic resonance imaging Womersley12 to assess the percentage of body fat, but
and computerised tomography.4,5 However, these they are also used to estimate the subcutaneous fat
methods are expensive and, in the case of computer- distribution. Although the reliability of anthropo-
ized tomography, the subjects are exposed to ionizing metric measurements has been extensively studied,
radiation. Measurements of anthropometric variables information on different sources of variation and on
such as sagittal abdominal diameter (SAD), waist the way in which variations are related to body mass
circumference, hip circumference, waist-to-hip ratio index (BMI) is not always available.
(W=H) and different measures of skinfold thickness Poor precision in measurement of an anthropo-
metric variable will lead to underestimation of corre-
lations to other variables.13 This effect can be
*Correspondence: L Berglund, Department of Public Health and
Caring Sciences, Section for Geriatrics=Clinical Nutrition measured with the intra-class correlation coef®cient
Research, Uppsala University, PO Box 609, SE-751 25 Uppsala, (ICC). The level of ICC might be different for lean
Sweden. and overweight subjects, which could result in differ-
E-mail: [email protected]
Received 15 March 1999; revised 1 December 1999; accepted ential underestimations. The ICC depends on the pure
5 January 2000 measurement error, the biological variation over time
 Reliability of anthropometric measurements
K Nordhamn et al
653
within subjects and the between-subject variation. height (in meters) squared. Height was rounded to
Thus, it is of importance to estimate these compo- 0.5 cm. Body weight was measured on a digital scale
nents. To our knowledge such estimates have not been with an accuracy of 0.1 kg.
presented separately for lean and overweight subjects. Waist and hip circumferences, like SAD, were
The aim of this study was to assess intra-observer measured with the subject in a supine position.
variations (IOV), intra-subject variations (ISV) and Waist circumference was measured midway between
ICC for a selected number of anthropometric mea- the lower rib and the iliac crest. Hip circumference
surements (SAD, waist and hip circumferences, W=H, was measured at the widest part of the hip. The waist
and skinfold measurements Ð biceps, triceps, sub- and hip measurements were read to the nearest 5 mm.
scapular, suprailiac, umbilical, anterior and posterior W=H was calculated as waist divided by the hip
thigh) in groups of lean and overweight subjects. circumference. SAD was measured both with the
Also, the effect of ICC on estimates of correlation legs stretched and bent. The perpendicular distance
coef®cients was to be evaluated. SAD was measured between the plane of support and the highest point of
both with legs stretched and bent to see whether this the abdomen was measured and read to the nearest
could in¯uence the result. Hypotheses to be tested 1 mm.
were: whether ICC values were different for over- Skinfolds were measured with two calibrated Har-
weight and lean subjects for all measurements; and if penden skinfold callipers. The skinfolds were mea-
ICC values for SAD, waist and W=H were different in sured at seven different locations and were read to the
a combined group of overweight and lean subjects. nearest 0.1 mm. All measurements were made on the
non-dominant side. The calliper was positioned, and
was read after approximately 3 s. No marks were done
at the actual skinfold site before measurement. The
Subjects and methods skinfolds used were:
 Biceps Ð midway from the armpit to the elbow on
Subjects the front of the arm;
Fifty-one subjects were divided into two groups; lean  Triceps Ð midway from the joint of the shoulder
(n ˆ 25, BMI < 26) and overweight (n ˆ 26, to the elbow on the back of the arm;
BMI  26). The lean and the overweight groups con-  Subscapular Ð 2 cm below the shoulder-blade and
sisted of 14 women and 11 men, and 13 women and 2 cm from the spine;
13 men, respectively. The overweight subjects were  Suprailiac Ð on the mid-axillary line just above
mainly recruited from the metabolic unit at Uppsala the iliac spine;
University Hospital. The lean subjects were either  Umbilical Ð 2 cm to the left of the umbilicus;
staff or students at Uppsala University or recruited  Anterior thigh Ð 10 cm above the top of the
from companies located nearby. Participation in the patella, front aspect;
study was voluntary.  Posterior thigh Ð 10 cm above the top of the
patella, back aspect.

Design
The measurement of weight, height and central obe-
Each patient was examined on two occasions, 1 ± 3 sity were performed by ®ve trained observers from the
weeks apart. The medical history, including informa- metabolic unit, two observers for each subject accord-
tion about large weight changes during the previous ing to the measurement scheme. The measurements of
month, was obtained with a questionnaire on the ®rst skinfold thickness were performed by two of the
occasion. The subjects were asked to continue their authors (EO and KN), who were trained by an
usual daily life during the study. The measurements experienced observer.
were performed as follows:

Occasion Observer
Statistical methods
The difference between the two measurements made
1 A by observer A at occasion 1 was calculated (do) as
1 B
1 A well as the difference between the measurements
2 B made by observer B at occasions 1 and 2 (ds). There-
after, measures of IOV and ISV (sum of biological
Thus, the set of measurements was performed four variation and IOV) were calculated as standard devia-
times in each subject, three times on the ®rst occasion tions of do and ds , respectively, and as coef®cients of
and once on the second. In this way each observer variation (CV), ie standard deviations in percent of the
made two sets of measurements in each subject. total mean.14 These calculations were performed for
the whole study group and for the overweight and lean
Methods groups separately.
BMI was calculated as body weight (in kilograms) Low reliability in measurement of an anthro-
without shoes and with light clothing, divided by pometric variable will lead to underestimation of
International Journal of Obesity
 Reliability of anthropometric measurements
K Nordhamn et al
654
correlations to other variables. ICC and the closely Table 2 Intra-subject variations (ISV) of weight, height and BMI
related attenuation factor, which is a measure of the Overweight (n ˆ 24) Lean (n ˆ 25) All (n ˆ 49)
actual relative bias in correlations, are measures of
reliability. For example, an attenuation factor of 0.9 s.d. CV% s.d. CV% s.d. CV%
means that the expected underestimation of a correla- Weight (kg)
tion will be 10%. These measures were calculated from ISV 1.3 1.4 0.5 0.7 1.0 1.2
the ratio of the between-subject variation to ISV.13 Height (cm)
ICCs and attenuation factors were estimated for all ISV 0.4 0.2 0.2 0.1 0.3 0.2
anthropometric variables for the whole study group and BMI (kg=m2)
ISV 0.4a 1.3a 0.1 0.6 0.3b 1.2b
for the lean and overweight groups separately.
Statistical signi®cance tests of the hypotheses were a
n ˆ 23. bn ˆ 48.
made for comparisons of ICC values between over-
weight and lean groups13 and for comparisons of ICC Table 3 Intra-subject variations (ISV) and intra-observer
variations (IOV) of central obesity measurements
values between waist, SAD with bent legs and W=H
measurements within the whole study group.13 All Overweight (n ˆ 26a) Lean (n ˆ 25) All (n ˆ 51)a
tests were two-tailed and P-values < 0.05 were con- s.d. CV% s.d. CV% s.d. CV%
sidered statistically signi®cant. All other comparisons
were descriptive. The calculations were made with the Waist (cm)
IOV 1.1 1.1 1.3 1.7 1.2 1.4
statistical program package SAS version 6.12 (SAS ISV 3.6 3.4 1.5 2.0 2.8 3.0
Institute Inc., Cary, NC, USA). Hip (cm)
IOV 1.7 1.5 2.1 2.2 1.9 1.8
ISV 2.2 2.0 2.3 2.4 2.2 2.1
Results W=H
IOV 0.01 1.6 0.02 3.0 0.02 2.3
ISV 0.04 3.8 0.02 2.8 0.03 3.4
The characteristics of the investigated groups are SAD bent (cm)
presented in Table 1. Mean BMI values differed IOV 0.3 1.2 0.4 2.1 0.4 1.6
ISV 0.8 2.8 0.4 2.3 0.6 2.7
between the overweight men and women (32.7 vs
SAD stretched (cm)
35.5) with P-value 0.018. IOV 0.5 1.8 0.3 1.3 0.4 1.7
For the measurements of weight, height and BMI, ISV 0.9 3.2 0.4 2.1 0.7 2.9
only ISV was calculated. As seen in Table 2, the a
For ISV n ˆ 25 for overweight group and n ˆ 50 for all subjects.
variation in height measurements was very small. In
this sample, the variation in BMI was mostly due to
variation in weight, even though BMI is de®ned with The variation in measurements of skinfolds, shown in
height squared. The contribution from height CV is Table 4, were larger than those in the other anthro-
17% of the BMI CV. pometric measurements (Table 2 and 3). The triceps
The variations in measurements of central obesity and umbilical skinfolds showed the lowest IOV of the
are compared between the subgroups in Table 3. IOV skinfolds measurements (8%). ISV was larger for the
was similar between the overweight and lean groups. triceps and biceps than for the other skinfolds. The
The coef®cient of variation for W=H was larger than
those for waist and hip circumferences. This observa- Table 4 Intra-subject variations (ISV) and intra-observer
tion was the same in both groups and for both IOV variations (IOV) of skinfold measurements
and ISV. For W=H, the ISV and IOV were higher than Overweight Lean All
the variations for the SAD measurements. For SAD
with stretched and bent legs the variations were s.d. CV% n s.d. CV% n s.d. CV% n
similar. Biceps (mm)
IOV 2.2 14.4 26 0.8 10.2 25 1.7 14.3 51
ISV 3.4 21.6 26 0.9 11.1 25 2.5 20.8 51
Table 1 Characteristics of the groups Triceps (mm)
IOV 1.4 7.5 26 1.1 8.7 25 1.2 8.0 51
Overweight (n ˆ 26) Lean (n ˆ 25) All (n ˆ 51)
ISV 3.3 18.1 26 2.9 23.8 25 3.1 20.4 51
Age (y) Subscapular (mm)
Mean (s.d.) 47.7 (12.6) 39.8 (12.8) 43.8 (13.2) IOV 4.9 16.3 22 1.4 9.9 25 3.5 15.8 47
Range 20.8 ± 67.6 23.4 ± 63.3 20.8 ± 67.6 ISV 5.0 16.4 24 1.0 6.9 25 3.5 15.8 49
Weight (kg) Suprailiac (mm)
Mean (s.d) 99.3 (11.4) 65.8 (10.0) 82.8 (20.0) IOV 3.5 11.7 25 1.1 8.0 25 2.6 11.9 50
Range 81 ± 132 50 ± 87 50 ± 132 ISV 4.3 14.2 26 1.5 11.1 23 3.3 14.9 49
Height (cm) Umbilical (mm)
Mean (s.d) 172 (8) 171 (8) 171 (8) IOV 2.7 7.5 13 1.5 7.8 23 2.0 7.3 36
Range 154 ± 186 157 ± 191 154 ± 191 ISV 5.4 14.8 22 2.8 14.5 21 4.3 15.5 43
BMI (kg=m2) Anterior thigh (mm)
Mean (s.d.) 33.7 (4.2) 22.5 (2.2) 28.2 (6.6) IOV 4.1 16.6 19 2.0 14.3 23 3.2 16.0 42
Range 26 ± 45 18 ± 26 18 ± 45 ISV 3.6 14.7 24 1.8 12.9 16 3.0 15.5 40

International Journal of Obesity

 Reliability of anthropometric measurements
K Nordhamn et al
655
variations were generally larger in the overweight group Table 6 Intra-class correlation coef®cients (ICC) and attenuation
factors for skinfold measurements
than in the lean group. This was most pronounced for the
biceps and subscapular measurements. Due to technical Overweight Lean All
problems in measuring the posterior thigh this variable subjects subjects subjects Pa
was excluded from all results. Biceps (n) 26 25 51
The ICC values and attenuation factors, shown in Attenuation factor 0.90 0.97 0.94
Table 5 for waist and hip circumferences and SAD ICC 0.81 0.94 0.89 0.031
were high (ICC: 0.96 ± 0.98; attenuation factors: Triceps (n) 26 25 51
Attenuation factor 0.91 0.86 0.92
0.98 ± 0.99) and lower for W=H (0.90 and 0.95, ICC 0.82 0.74 0.84 0.446
respectively). P-values for comparisons of SAD with Subscapular (n) 24 25 49
bent legs and waist vs W=H were < 0.001 and 0.001, Attenuation factor 0.82 0.98 0.93
respectively, and the P-value for comparison of SAD ICC 0.67 0.96 0.87 0.000
with bent legs versus waist was 0.472. The ICC values Suprailiac (n) 26 23 49
Attenuation factor 0.87 0.96 0.95
and the attenuation factors were higher for the lean ICC 0.76 0.91 0.90 0.065
group than for the overweight group (except for hip Umbilical (n) 22 21 43
measurements). These differences in ICC reached Attenuation factor 0.80 0.94 0.93
statistical signi®cance for waist and weight. The ICC 0.64 0.88 0.86 0.048
smallest difference in ICC between lean and over- Anterior thigh (n) 24 16 40
Attenuation factor 0.96 0.88 0.97
weight subjects was observed for SAD with bent legs. ICC 0.92 0.77 0.93 0.089
The ICC values and the attenuation factors for a
P-value for comparison of ICC values between overweight and
skinfold measurements ranged from 0.84 to 0.93 and lean groups.
0.92 to 0.97, respectively, with the lowest value for
triceps measurements (Table 6). For biceps, subscap-
ular and umbilical skinfolds lean subjects had larger between overweight and lean groups regarding
ICC and attenuation factors than overweight subjects. reliabilities have been made using signi®cance tests.
One important observation in this study was that
several anthropometric measurements had lower relia-
Discussion bility in overweight subjects than in lean subjects.
This was the case for the trunkal measurements Ð
waist, SAD with bent and stretched legs, subscapular,
This is the ®rst study to formally show that SAD and suprailiac and umbilical skinfolds Ð although the dif-
waist measurements have higher reliabilities than ference did not reach statistical signi®cance for the
W=H. It is also the ®rst time that comparisons SAD measurements and suprailiac skinfolds. It is
therefore recommended that replicate measurements
Table 5 Intra-class correlation coef®cients (ICC) and attenu-
ation factors for central obesity measurements of anthropometric variables be made in studies of
overweight subjects. As the IOV component of the
Overweight Lean All total measurement error is in general large (Tables 3
subjects subjects subjects Pa
and 4), it will be suf®cient to make replicate measure-
Waist (n) 25 25 50 ments on one occasion. Replicate measurements
Attenuation factor 0.92 0.98 0.99
ICC 0.85 0.96 0.97 0.030 appear to be especially important when correlation
Hip (n) 25 25 50 coef®cients between anthropometric measurements
Attenuation factor 0.97 0.93 0.98 and metabolic factors are compared between lean
ICC 0.94 0.87 0.96 0.146 and overweight groups. The difference in measure-
W=H (n) 25 25 50 ment precision, as measured by the ICC, can
Attenuation factor 0.89 0.91 0.95
ICC 0.78 0.83 0.90 0.658 otherwise give rise to a greater underestimation of
SAD bent (n) 25 25 50 correlations in overweight groups than in lean groups.
Attenuation factor 0.96 0.98 0.99 We are aware of the fact that the higher ICC values in
ICC 0.92 0.95 0.98 0.297 the whole study group is an effect of increased ranges to
SAD stretched (n) 25 25 50 some extent. As comparisons of ICCs were not per-
Attenuation factor 0.94 0.97 0.99
ICC 0.89 0.95 0.97 0.170 formedbetween the whole studygroup and the subgroups
Weight (n) 24 25 49 thisissuewasofnomajorconcern.Thecomparisonswere
Attenuation factor 0.99 1.00 1.00 made between the two subgroups and between different
ICC 0.99 1.00 1.00 0.001 measures in the whole study group.
Height (n) 24 25 49 When we compared the different anthropometric
Attenuation factor 1.00 1.00 1.00
ICC 1.00 1.00 1.00 0.239 measurements of central obesity regarding ICC values
BMI (n) 23 25 48 (Table 5), we found that these values were generally
Attenuation factor 0.99 1.00 1.00 lower in the overweight group. However, we observed
ICC 0.99 1.00 1.00 0.065 that the ICC values for SAD measurements with bent
a
P-value for comparison of ICC values between overweight and legs showed ICC values above 0.9 in both groups. In
lean groups. a prospective study on 45 ± 50-y-old male Swedish
International Journal of Obesity
 Reliability of anthropometric measurements
K Nordhamn et al
656
Volvo employees it was shown that SAD was a strong with other reports.19,20 Regarding IOV for W=H, we
indicator of coronary risk.15 The fact that SAD was a obtained similar results to those in two other stu-
strong indicator may be explained by a high reliability dies.18,20 The IOV of the triceps and subscapular
of the SAD measurements, as shown in our data. skinfolds in this study are comparable to those
Since SAD measurements with bent legs showed found by Sullivan et al.21 Ferrario et al 20 noted
somewhat higher ICC values than measurements lower variation than ours for subscapular but compar-
with stretched legs, we would recommend that SAD able variation for triceps skinfolds.
be measured with bent legs (unless a slide-calliper is In conclusion, the ICCs were in general lower in
used) in order to obtain a measure with high reliability overweight subjects, indicating that correlations can
over a wide range of BMI. be underestimated in overweight groups. We propose
Results from a previous study3 indicated that waist that SAD and waist circumference have a higher
circumference is a more predictive measure than W=H predictive capacity for cardiovascular morbidity risk
for development of type 2 diabetes. In a weight loss than W=H, partly for the reason that the former
study16 it was concluded that changes in waist cir- measures have higher reliability. Speci®cally, waist
cumference correlated signi®cantly to changes in circumference is more precise than W=H, since a ratio
cardiovascular risk factors such as total cholesterol, generally has larger measurement errors than its
low density lipoprotein cholesterol and diastolic blood nominator and denominator. In a comparison of all
pressure, whereas the correlations between W=H measurements, SAD with bent legs showed the high-
changes and changes in these variables were less est reliability, which was unaffected by BMI. We
pronounced. Pouliot et al 7 showed that waist circum- therefore recommend the use of SAD with bent legs
ference is a better anthropometric index than W=H for as a cheap and accurate assessment of metabolic and
predicting abdominal visceral adipose tissue and car- cardiovascular risk. It remains to be investigated how
diovascular morbidity risk. They concluded that waist different diseases, such as type 2 diabetes, in¯uence
circumference and W=H provide different informa- the degree of variability in anthropometric measure-
tion: the former determining the extent of abdominal ments as well as the difference between overweight
obesity and the latter the regional distribution of and lean subjects. Also, this question needs to be
adipose tissue. Our ®ndings support another possible addressed in elderly people and in a larger randomized
explanation for this discrepancy in the predictive population sample.
ability of waist circumference and W=H. We have
shown that the W=H ratio carries larger measurement
errors than waist and hip measurements alone. Ratio References
measures will in general have lower reliability than 1 Larsson B, SvaÈrdsudd K, Welin L, Wilhelmsen L, BjoÈrntorp P,
Tibblin G. Abdominal adipose tissue distribution, obesity, and
their nominator and denominator.17 When the true risk of cardiovascular disease and death: 13 year follow up of
values of the nominator and denominator are posi- participants in the study of men born in 1913. Br Med J 1984;
tively correlated (in our study the correlation between 288: 1401 ± 1404.
waist and hip was 0.87) and their measurement errors 2 Lapidus L, Bengtsson C, Larsson B, Pennert K, Rybo E,
are uncorrelated, the reliability of the ratio will be SjoÈstroÈm L. Distribution of adipose tissue and risk of cardio-
vascular disease and death: a 12 year follow up of participants
poorer than the reliability of the components. We thus in the population study of women in Gothenburg, Sweden.
conclude that the discrepancies between the predictive Br Med J 1984; 289: 1257 ± 1261.
abilities of waist circumference and W=H are partly 3 Ohlson LO, Larsson B, SvaÈrdudd K, Welin L, Eriksson H,
due to a lower reliability of the latter measure. In Wilhelmsen L, BjoÈrntorp P, Tibblin G. The in¯uence of body
order to achieve an estimate of the proportion of the fat distribution on the incidence of diabetes mellitus. 13.5
years of follow-up of the participants in the study of men born
different predictive abilities that could be accounted in 1913. Diabetes 1985; 34: 1055 ± 1058.
for by different reliabilities, we applied our estimates 4 van der Kooy K, Seidell JC. Techniques for the measurement
of attenuation factors to correlations from a large of visceral fat: a practical guide. Int J Obes 1993; 17:
cross-sectional study on the relation between anthro- 187 ± 196.
pometric variables and cardiovascular risk factors.11 5 Heyms®eld SB, Tighe A, Wang ZM. Nutritional assessment
by anthropometric and biochemical methods. In: Shils ME,
In that study, waist and W=H were associated with a Olson JA, Shike M (eds). Modern nutrition in health and
cardiovascular risk index with correlation coef®cients disease, 8th edn. Lea & Febiger: Malvern, 1994, pp 812 ± 841.
of 0.53 and 0.45, respectively. The predictive ability is 6 Lean MEJ, Han TS, Seidell JC. Impairment of health and
de®ned as the square of the correlation. If the two quality of life in people with large waist circumference. Lancet
measurements had the same attenuation factors, the 1998; 351: 853 ± 856.
7 Pouliot MC, Despres JP, Lemieux S, Moorjani S, Bouchard C,
difference in predictive ability would be reduced by Tremblay A, Nadeau A, Lupien PJ. Waist circumference and
19.5%. This example further implies the importance abdominal sagittal diameter: best simple anthropometric
of measurement reliability when correlation coef®- indexes of abdominal and visceral adipose tissue accumulation
cients are compared. and related cardiovascular risk in men and women. Am J
We found a much lower variation in SAD than in Cardiol 1994; 73: 460 ± 468.
8 Richelsen B, Pedersen SB. Associations between different
W=H, which is in contrast to a previous observation.18 anthropometric measurements of fatness and metabolic risk
The ICC value was lower for W=H than for waist and parameters in non-overweight, healthy, middle-aged men. Int J
hip measurements separately, a ®nding in accordance Obes 1995; 19: 169 ± 174.

International Journal of Obesity

 Reliability of anthropometric measurements
K Nordhamn et al
657
9 Han TS, van Leer EM, Seidell JC, Lean MEJ. Waist circum- 16 Han TS, Richmond P, Avenell A, Lean MEJ. Waist
ference action levels in the identi®cation of cardiovascular risk circumference reduction and cardiovascular bene®ts during
factors: prevalence study in a random sample Br Med J 1995; weight loss in women. Int J Obes 1997; 21: 127 ± 134.
311: 1401 ± 1405. 17 Allison DB, Paultre F, Goran MI, Poehlman ET, Heyms®eld
10 Ashwell M, Cole TJ, Dixon AK. Ratio of waist circumference SB. Statistical considerations regarding the use of ratios to
to height is strong predictor of intra-abdominal fat. Br Med J adjust data. Int J Obes 1995; 19: 644 ± 652.
1996; 313: 559 ± 560 (letter). 18 Hojby-Rasmussen M, Andersen T, Breum L, Hilsted J,
11 OÈ hrvall M, Berglund L, Vessby B. Sagittal abdominal dia- Gotzsche PC. Observer variation in measurements of waist-
meter compared with other anthropometric measurements in hip ratio and the abdominal sagittal diameter. Int J Obes 1993;
relation to cardiovascular risk. Int J Obes 2000; 24: 497 ± 501. 17: 323 ± 327.
12 Durnin JVGA, Womersley J. Body fat assessed from total 19 Sonnenschein EG, Kim MY, Pasternack BS, Toniolo
body density and its estimation from skinfold thickness: PG. Sources of variability in waist and hip measurements
measurements on 481 men and women aged from 16 to in middle-aged women. Am J Epidemiol 1993; 138: 301 ±
72 years. Br J Nutr 1974; 32: 77 ± 97. 309.
13 Rosner B, Willett WC. Interval estimates for correlation 20 Ferrario M, Carpenter MA, Chambless LE. Reliability of
coef®cients corrected for within-person variation: implications body fat distribution measurements. The ARIC study baseline
for study design and hypothesis testing. Am J Epidemiol 1988; cohort results. Atherosclerosis risk in communities study. Int J
127: 377 ± 386. Obes 1995; 19: 449 ± 457.
14 Bland M. Clinical measurement. In: An introduction to medical 21 Sullivan DH, Patch Ga, Baden AL, Lipschitz DA. An
statistics. Oxford University Press: Oxford, 1987, pp 276 ± 278. approach to assessing the reliability of anthropometrics in
15 Kumlin L, Dimberg L, MaÊrin P. Ratio of abdominal sagittal elderly patients. J Am Geriatr Soc 1989; 37: 607 ± 613.
diameter to height is strong indicator of coronary risk. Br Med
J 1997; 314: 830 (letter).

International Journal of Obesity