8.3: Errors in anthro­pometry

Errors can occur in nutri­tional anthro­pometry which may affect the precision, accuracy, and validity of the mea­sure­ments, and thus indices and indicators. Three major sources of error are significant: mea­sure­ment errors, alterations in the compo­sition and physical properties of certain tissues, and the use of invalid assumptions in the derivation of body compo­sition from anthro­pometric mea­sure­ments (Heymsfield and Casper, 1987).

Measure­ment errors arise from examiner error resulting from inadequate training, instru­ment error, and difficulties in making the mea­sure­ment (e.g., skinfold thicknesses). The major sources of mea­sure­ment error in anthro­pometry are shown in Boxes 9.2 and 9.3. Both random and sys­tem­atic mea­sure­ment errors may occur which reduce the validity of the index and any indicator constructed from the index; they have been extensively reviewed by Ulijaszek and Kerr (1999).

Box 9.2: Common errors and possible solutions when measuring length, height, and weight.

All mea­sure­ments

• Inadequate instru­ment: Select method appro­priate to resources.
• Restless child: Postpone mea­sure­ment or involve parent in procedure or use culturally appro­priate procedures.
• Errors in reading equipment: Training and refresher exercises, stressing accuracy, with intermittent oversight by supervisor.
• Errors in recording results: Record results immediately after mea­sure­ment and have record checked by a second person.

Length

• Incorrect method for age: Use length only when child is < 2y.
• Footwear/headwear not removed: Remove as local culture permits (or make allowances).
• Head not in correct Frankfurt plane: Correct position of child before measuring.
• Child not straight along board and/or feet not parallel with movable board: Have assistant and child's parent present; don't take the mea­sure­ment while the child is struggling; settle child.
• Board not firmly against heels: Correct pressure should be practiced.

Height

• Incorrect method for age: Use only when child is ≥ 2y.
• Footwear/headware not removed; Remove as local culture permits (or make allowances).
• Head not in correct Frankfurt plane, subject not straight, knees bent, or feet not flat on floor: Correct technique with practice and retraining; provide adequate assistance; calm noncooperative children.
• Board not firmly against head: Lower head board to compress hair.

Weight

• Room cold, no privacy: Use appro­priate clinic facilities.
• Scale not calibrated to zero: Re-calibrate after every subject.
• Subject wearing heavy clothing: Remove or make allowances for clothing.
• Subject moving or anxious as a result of prior incident: Wait until subject is calm or remove the cause of anxiety.

9.3.1 Random mea­sure­ment errors and precision

Random mea­sure­ment errors limit precision or the extent to which repeated mea­sure­ments of the same variable give the same value. Random mea­sure­ment errors can be minimized by training personnel to use standardized techniques and precise, correctly calibrated instru­ments (Lohman et al., 1988) . Further­more, the precision (and accuracy) of each mea­sure­ment technique should be firmly established prior to use. To improve precision, two or three mea­sure­ments on each individual should be conducted.

A description of the mea­sure­ment techniques used in the WHO Multi­center Growth Reference Study (MGRS) are available in de Onis et al. (2004), as well as in an anthro­pometric training video from WHO. In the WHO MGRS the equipment was calibrated regularly, using standard weights over the full weight range for the portable electronic weighing scales, metal rods of known length for both the infantometer and stadiometer, and calibration blocks of varying widths for the skinfold calipers.

Poor precision often reflects within-examiner error, but between-examiner error may also be significant in surveys with multiple examiners. The precision of a mea­sure­ment technique can be assessed by calculating:

• Technical error of the mea­sure­ment (TEM)
• per­cent­age technical error (%TEM)
• Coefficient of reliability (R)

These parameters can be calculated for each anthro­pometric mea­sure­ment technique from repeated mea­sure­ments on each subject made within a few minutes to avoid physiological fluctuations. A minimum of 10 subjects is recom­mended.

Box 9.3: Common errors and possile solutions when measuring mid-upper-arm circumference, head circumference, and triceps skinfold.

It is partic­ularly important with these mea­sure­ments to use the correct techniques. This requires training, supervision, and regular refresher courses. Always take into account any cultural problems, such as the wearing of arm bands etc.

Arm circum­ference

• Subject not standing in correct position: Position subject correctly.
• Tape too thick, stretched, or creased: Use correct instru­ment.
• Wrong arm: Use left arm.
• Mid-arm point incorrectly marked: measure and remark midpoint carefully.
• Arm not hanging loosely by side during mea­sure­ment: ask subject to allow arm to hang loosely.
• Examiner not comfortable or level with subject: position subject correctly relative to examiner.
• Tape around arm not at midpoint: reposition tape.
• Tape too tight (causing skin contour indentation): loosen tape.
• Tape too loose: carefully tighten tape.

Head circum­ference

• Occipital protuberance / supraorbital landmarks poorly defined: position tape correctly.
• Hair crushed inadequately: carefully tighten tape.
• Ears under tape, or tension position poorly maintained at time of reading: repeat after positioning tape correctly.
• Headwear not removed: remove as local culture permits.

Triceps fatfold

• Wrong arm: use left arm.
• Mid-arm point or posterior plane incorrectly measured or marked: measure and remark midpoint carefully.
• Arm not loose by side during mea­sure­ment: ask subject to allow arm to hang loosely.
• Finger-thumb pinch or caliper placement too deep (muscle) or too superficial (skin): correct technique with training and supervision.
• Caliper jaws not at marked site; reading done too early, pinch not maintained, caliper handle not released: correct technique with training and supervision.
• Examiner not comfortable or level with subject: ensure examiner is correctly positioned.

From: (Zerfas AJ, 1979)

The technical error of the mea­sure­ment (TEM) is the square root of the mea­sure­ment error variance. TEM is expressed in the same units as that of the anthro­pometric mea­sure­ment under study and is often age dependent. The value is also related to the anthro­pometric characteristics of the study group. The calculation varies according to the number of replicate mea­sure­ments made. For one examiner making two mea­sure­ments, TEM =

\[\sqrt{\left(\Sigma \mathrm{D}^2\right) / 2 \mathrm{~N}}\nonumber\]

where D = the difference between two mea­sure­ments and N = number of subjects. For more than two mea­sure­ments, the equation is more complex, and TEM =

\[\sqrt{\sum_1^N\left[\left(\sum_1^K M^2\right)-\left(\left(\sum_1^K M\right)^2 / K\right)\right] /[N(K-1)]}\nonumber\]

where N = number of subjects, K is the number of deter­minations of the variable taken on each subject, and M_n is the n^th replicate of the mea­sure­ment, where n varies from 1 to K.

Table 9.2: shows the calculation of TEM from mea­sure­ments of stature performed four times on 10 subjects by a single anthro­pometrist.

Table 9.2 Sample calculation of technical error of mea­sure­ment (TEM) from repeat mea­sure­ments of stature (m) carried out by one anthro­pometrist on 10 subjects. M= mea­sure­ment, K= number of replicates, N= number of subjects.

Subject	Stature (m) as deter­mined on repeat				(1)	(2)	Diff.
no.	1	2	3	4	ΣM^2	(ΣM)^2/K	(1) − (2)
1	0.865	0.863	0.863	0.864	2.984259	2.984256	0.000003
2	1.023	1.023	1.027	1.025	4.198412	4.198401	0.000011
3	0.982	0.980	0.989	0.985	3.873070	3.873024	0.000046
4	0.817	0.816	0.812	0.817	2.660178	2.660161	0.000017
5	0.901	0.894	0.900	0.903	3.236446	3.236401	0.000045
6	0.880	0.876	0.881	0.881	3.094098	3.094081	0.000017
7	0.948	0.947	0.947	0.946	3.587238	3.587236	0.000002
8	0.906	0.905	0.907	0.908	3.286974	3.286969	0.000005
9	0.924	0.924	0.926	0.924	3.418804	3.418801	0.000003
10	0.969	0.987	1.002	0.993	3.942343	3.942210	0.000133
							Σ = 0.000282
TEM = = = 0.00307

Note that the size of the mea­sure­ment also influences the size of the associated TEM, so that comparisons of precision of different anthro­pometric mea­sure­ments using TEM cannot be made easily. This is high­lighted in Table 9.3 in which the TEM for five anthro­pometric mea­sure­ments taken during the initial standardization session conducted at the Brazilian site of the WHO Multi­centre Growth Reference Study (MGRS) are presented (de Onis et al., 2004). Table 9.3 also depicts the maximum allowable differences between the mea­sure­ments of two observers that were used in the WHO MGRS, and set based on TEMs achieved during the standardization session.

Table 9.3. Maximum allowable differences between the mea­sure­ments of two observers.
TEM: Technical error of mea­sure­ment
From: de Onis et al., 2004.

mea­sure­ment	Brazil TEM from pilot study	Maximum allowable difference
Weight	Not available	100g
Length	2.5mm	7.0mm
Head circum­ference	1.4mm	5.0mm
Arm circum­ference	1.8mm	5.0mm
Triceps skinfold	0.44mm	2.0mm
Subscapular skinfold	0.43mm	2.0mm

Per­cent­age TEM has been recom­mended to overcome the difficulty of the TEM being dependent on the size of the original mea­sure­ment (Norton and Olds, 1996). The per­cent­age technical error of the mea­sure­ment is analogous to the coefficient of variation and is calculated as:

\[\% \mathrm{TEM}=(\mathrm{TEM} / \text { mean }) \times 100 \%\nonumber\]

Note that %TEM has no units and can be used to make direct comparisons of all types of anthro­pometric mea­sure­ments. It cannot be used, however, when more than one examiner is involved, as then both within- and between-examiner errors are involved. Ulijaszek and Kerr (1999) describe ways to deal with this more complex case.

The coefficient of reliability (R) is an alternative approach that is widely used for comparing mea­sure­ment errors among anthro­pometric mea­sure­ments. It ranges from 0 to 1 and can be calculated using the following equation:

\[\mathrm{R}=1-\left((\mathrm{TEM})^2 / \mathrm{s}^2\right)\nonumber\]

where s² is the between-subject variance. The coefficient indicates the proportion of between-subject variance in a measured population which is free from mea­sure­ment error. Hence, a mea­sure­ment with R = 0.95 indicates that 95% of the variance is due to factors other than mea­sure­ment error.

Whenever possible, a coefficient of reliability > 0.95 should be sought. Coefficients of reliability can be used to compare the relative reliability of different anthro­pometric mea­sure­ments, and the same mea­sure­ments in different age groups, as well as for calculating sample sizes in anthro­pometric surveys.

More details of standardization procedures and calculation of precision using TEM, per­cent­age TEM, and coefficient of reliability are given in Lohman et al. (1988). In general, the precision of weight and height mea­sure­ments are high. However, for waist and hip circum­ferences, between-examiner error tends to be large and it is preferable for only one examiner to take these mea­sure­ments. Because skinfolds are notoriously imprecise, both within- and between-examiner errors can be large. Therefore, rigorous training using standardized techniques and calibrated equipment are critical when skinfold mea­sure­ments are taken.

9.3.2. Sys­tem­atic mea­sure­ment errors and accuracy

Sys­tem­atic mea­sure­ment errors affect the accuracy of anthro­pometric mea­sure­ments or how close the mea­sure­ments are to the true value. The most common form of sys­tem­atic error in anthro­pometry results from equipment bias. For example, apparent discrepancies in skinfold mea­sure­ments performed on the same person but with different calipers may be due to compression differences arising from variations in spring pressure and surface area of the calipers (Schmidt & Carter, 1990); Harpenden and Holtain skinfold calipers consistently yield smaller values than Lange calipers(Gruber et al., 1990). Errors arising from bias reduce the accuracy of the mea­sure­ment by altering the mean or median value, but have no effect on the variance. Hence, such errors do not alter the precision of the mea­sure­ment.

The timing of some anthro­pometric mea­sure­ments of body size and compo­sition is also known to be critical, partic­ularly for short-term growth studies: progressive decreases in the height of an individual during the day as a consequence of compression of the spinal column, for example, may seriously compromise the accuracy of height velocity mea­sure­ments.

The deter­mination of accuracy in anthro­pometry is difficult because the correct value of any anthro­pometric mea­sure­ment is never known with absolute certainty. In the absence of absolute reference standards, the accuracy of anthro­pometric mea­sure­ments is estimated by comparing them with those made by a criterion anthro­pometrist (Ulijaszek & Kerr, 1999), a person who has been highly trained in the standardized mea­sure­ment techniques and whose mea­sure­ments compare well to those from another criterion anthro­pometrist.

In preparation for the compilation of the new WHO Child Growth Standard, four anthro­pometrists were trained and standardized against a criterion anthro­pometrist, designated as the “lead” anthro­pometrist; see de Onis et al. (2004) for more details. All anthro­pometric mea­sure­ments were taken and recorded inde­pen­dently by two designated anthro­pometrists, and their mea­sure­ment values compared for maximum allowable differences (Table 9.3). Targets for sports anthro­pometrists are also available (Gore et al., 1996).

Attempts should always be made to minimize mea­sure­ment errors. In longitudinal studies involving sequential anthro­pometric mea­sure­ments on the same group of individuals (e.g., surveillance), it is preferable, whenever possible, to have one person carrying out the same mea­sure­ments throughout the study to eliminate between-examiner errors. This is partic­ularly critical when increments in growth and body compo­sition are calculated; such increments are generally small and are associated with two error terms, one on each mea­sure­ment occasion. Recom­mendations for the minimum intervals necessary to provide reliable data on growth increments during infancy and early childhood (Guo et al., 1991) and adolescence(WHO, 1995) are available.

In large regional surveillance studies, it is often necessary to use several well-trained anthro­pometrists. In such circumstances, the between-examiner differences among anthro­pometrists must be monitored throughout the study to maintain the quality of the mea­sure­ments and thereby to identify and correct sys­tem­atic errors in the mea­sure­ments. This practice was followed during the WHO MGRS (de Onis et al., 2004).

In studies involving two longitudinal mea­sure­ments, the TEM can be calculated to estimate the proportion of the difference that can be attributed to mea­sure­ment error. For example, with a TEM of 0.3 for a given anthro­pometric mea­sure­ment, the TEM for the difference between two mea­sure­ments is:

\[\sqrt{(0.3)^2+(0.3)^2}=0.42\nonumber\]

because both TEM values contribute to the variance in the difference. Only if the difference exceeds 2 × 0.42 = 0.84 is there a 95% probability that the difference exceeds the mea­sure­ment error alone.

Once assured that such differences are not a function of mea­sure­ment error, then any changes in growth and body compo­sition can be correlated with factors such as age, the onset of disease, response to nutri­tion inter­vention therapy, and so on.

The collection of longitudinal anthro­pometric data is more time consuming, expensive, and laborious than from cross-sectional surveys, and, as a result, the sample size is generally smaller. Hence, the probability of sys­tem­atic sampling bias (Section 1.4.2) is generally greater than in more extensive cross-sectional surveys.

For cross-sectional studies, the examiners should be rotated among the subjects to reduce the effect of mea­sure­ment bias of the individual examiners. Statistical methods exist for removing anthro­pometric mea­sure­ment error from cross-sectional anthro­pometric data; details are given in Ulijaszek and Lourie (1994).

Cross-sectional surveys are useful for comparing population groups, provided that probability sampling techniques have been used to ensure that the samples are represen­tative of the populations from which they are drawn (Chapter 1). Recently WHO has provided countries with tools to develop or strengthen their surveillance systems so they have the capacity to monitor changes in the Global Nutri­tion Targets for 2030. They include the following anthro­pometric indicators: stunting, wasting, low birthweight, and childhood over­weight. For more details see: WHO Nutrition Tracking Tool.

9.3.3 Errors from changes in tissue compo­sition and properties

Variation in the compo­sition and physical properties of certain tissues may occur in both healthy and diseased subjects, resulting in inaccuracies in certain anthro­pometric mea­sure­ments. Even among healthy individuals, body weight may be affected by variations in tissue hydration with the menstrual cycle (Heymsfield and Casper, 1987; Madden and Smith, 2016).

Skinfold thickness mea­sure­ments may be influenced by variations in compressibility and skin thickness with age, gender, and the level of tissue hydration (Martin et al., 1992; Ward and Anderson, 1993). For example, repeated mea­sure­ments of skinfolds, over a short period (i.e., 5min), may actually decrease accuracy of skinfolds because later mea­sure­ments are more compressed due to the expulsion of water from the adipose tissue at the site of the earlier mea­sure­ment (Ulijaszek and Kerr, 1999).

The accuracy of waist circum­ference is affected by both the phase of respiration at the point of mea­sure­ment and by the tension of the abdominal wall. The phase of respiration is important because it deter­mines the extent of fullness of the lungs and the position of the diaphragm at the time of the mea­sure­ment. Increasing the tension of the abdominal wall (by sucking in) is frequently an unconscious reaction which is also important because it reduces the waist mea­sure­ment. To minimize these errors, WHO (2011) recom­mends advising the subject to relax and take a few deep, natural breaths before the actual mea­sure­ment and at the end of normal expiration.

In addition, during aging, demineral­ization of the bone and changes in body water may result in a decrease in the density of the fat-free mass (Visser et al., 1994; JafariNasabian et al., 2017), which are not always taken into account when calculating total body fat and hence fat-free mass from skinfolds via body density (see Chapter 11 for more details).

9.3.4 Invalid models and errors in body compo­sition

Invalid assumptions may lead to erroneous estimates of body compo­sition when these are derived from anthro­pometric mea­sure­ments, especially in obese or elderly patients and those with protein-energy mal­nu­trition or certain disease states. For instance, use of skinfold thickness mea­sure­ments to estimate total body fat assumes that (a) the thickness of the sub­cu­taneous adipose tissue reflects a constant proportion of the total body fat and (b) the sites selected represent the average thickness of the sub­cu­taneous adipose tissue. In fact, the relationship between sub­cu­taneous and internal fat is nonlinear and varies with body weight, age, and disease state. Very lean subjects have a smaller proportion of body fat deposited sub­cu­taneously than do obese subjects, and in malnourished persons there is probably a shift of fat storage from sub­cu­taneous to deep visceral sites. Variations in the distri­bution of sub­cu­taneous fat also occurs with age, sex, and ethnicity or race (Wagner and Heyward, 2000; He et al., 2002).

Estimates of mid-upper-arm muscle area are used as an index of total body muscle and the fat-free mass (Chapter 11), regardless of age and health status of the subjects. Such estimates are made, despite the known changes in the relationship between arm muscle and fat-free mass with age and certain disease states (Heymsfield and McManus, 1985), and the questionable accuracy of the algorithms used (Martine et al., 1997). Moreover, even the corrected algorithms developed for adults overestimate arm muscle area in obese persons when compared with the deter­mination by computerized tomography (Forbes et al., 1998).

Increasingly, body compo­sition is assessed by laboratory methods; these are described in Chapter 14. Even laboratory methods are based on certain assumptions that have been challenged in recent years. For example, until recently, densitometry, frequently using underwater weighing, has been the gold standard reference method for the deter­mination of the per­cent­age of body fat. The assumptions used in densitometry are that the densities of the fat mass and fat-free mass are constant at 0.90 and 1.10kg/L, respectively (Chapter 11). Several researchers have questioned the validity of using a constant density of the fat-free mass for groups who vary in age, gender, levels of body fatness, and race or ethnicity (Visser et al., 1997). During aging, the density of the fat-free mass may decrease due to demineral­ization of the bone and changes in body water, as noted above (Visser et al., 1994; JafariNasabian et al., 2017), which are not always taken into when calculating total body fat from skinfolds via body density, leading to a1%–2% overestimate of the body fat content in such subjects (Deurenberg et al., 1989); see Chapter 11 for more details.

In contrast, persons of African descent have a larger fat-free mass because they have a greater bone mineral density and body protein content compared to Caucasians (Wagner and Heyward, 2000). Such differences lead to an underestimate of body fat, when generalized equations developed for Caucasians are used.

Per­cent­age of body fat can also be deter­mined using an isotope dilution technique and dual‑energy X‑ray absorptiometry (DXA) (Chapter 14). Both of these methods assume a constant hydration of the fat-free mass (i.e., 73.2% water content), despite knowledge that it varies with age (Wang et al., 1999), obesity, and pregnancy (Hopkinson et al., 1997), and throughout the course of a clinical condition (e.g., inflammation) (Müller et al., 2016). When the actual hydration of fat-free mass is higher than the assumed value, then the per­cent­age of body fat is underestimated by isotope dilution techniques (Chapter 14) (Deurenberg-Yap et al., 2001). For example, even when pregnancy-specific values for hydration have been applied to account for the increased accretion of water that occurs during pregnancy, individual estimates of fat mass using isotope dilution differed by > 3kg from values based on the four-compartment model (Hopkinson et al., 1997). In contrast, hydration effects on estimates of fat by DXA are not significant (Pietrobelli et al., 1998).

Fortunately, the advent of multicomponent models (i.e., the 4‑compartment-model) with minimal assumptions for assessing body compo­sition circumvent the use of older methods, which use assumptions that are not always valid for certain ethnic groups or the elderly (Müller et al., 2016). Nevertheless, the use of multicomponent models is expensive, requiring more time and facilities.