ORIGINAL ARTICLE

Estimating body fat in NCAA Division I female athletes:a five-compartment model validation of laboratory methods

Jordan R. Moon Æ Joan M. Eckerson Æ Sarah E. Tobkin Æ Abbie E. Smith ÆChristopher M. Lockwood Æ Ashley A. Walter Æ Joel T. Cramer ÆTravis W. Beck Æ Jeffrey R. Stout

Accepted: 22 September 2008 / Published online: 21 October 2008

� Springer-Verlag 2008

Abstract The purpose of the present study was to

determine the validity of various laboratory methods for

estimating percent body fat (%fat) in NCAA Division I

college female athletes (n = 29; 20 ± 1 year). Body

composition was assessed via hydrostatic weighing (HW),

air displacement plethysmography (ADP), and dual-energy

X-ray absorptiometry (DXA), and estimates of %fat

derived using 4-compartment (C), 3C, and 2C models were

compared to a criterion 5C model that included bone

mineral content, body volume (BV), total body water, and

soft tissue mineral. The Wang-4C and the Siri-3C models

produced nearly identical values compared to the 5C model

(r [ 0.99, total error (TE) \ 0.40%fat). For the remaining

laboratory methods, constant error values (CE) ranged

from -0.04%fat (HW-Siri) to -3.71%fat (DXA); r values

ranged from 0.89 (ADP-Siri, ADP-Brozek) to 0.93 (DXA);

standard error of estimate values ranged from 1.78%fat

(DXA) to 2.19%fat (ADP-Siri, ADP-Brozek); and TE

values ranged from 2.22%fat (HW-Brozek) to 4.90%fat

(DXA). The limits of agreement for DXA (-10.10 to

2.68%fat) were the largest with a significant trend of -0.43

(P \ 0.05). With the exception of DXA, all of the equa-

tions resulted in acceptable TE values (\3.08%fat).

However, the results for individual estimates of %fat using

the Brozek equation indicated that the 2C models that

derived BV from ADP and HW overestimated (5.38,

3.65%) and underestimated (5.19, 4.88%) %fat, respec-

tively. The acceptable TE values for both HW and ADP

suggest that these methods are valid for estimating %fat in

college female athletes; however, the Wang-4C and Siri-3C

models should be used to identify individual estimates of

%fat in this population.

Keywords Air displacement plethysmography �Underwater weighing � Multi-compartment �Dual-energy X-ray absorptiometry

Introduction

Estimates of body composition characteristics are com-

monly used in both athletic and non-athletic populations to

identify general health status. For example, in athletes,

body fat has been shown to be highly related to the physical

demands of their respective sports (Fahey et al. 1975). In

addition, reference body composition values established for

professional athletes may be used by collegiate athletes and

their coaches to design nutrition and training programs in

an attempt to optimize performance.

Although laboratory methods such as hydrostatic

weighing (HW), air displacement plethysmography (ADP)

via the BOD POD�, and dual-energy X-ray absorptiometry

(DXA) are considered to provide the most accurate esti-

mates of body composition and are, oftentimes, accessible

to athletes, field techniques, such as skinfold measure-

ments, are more commonly used due to their low cost,

J. R. Moon � S. E. Tobkin � A. E. Smith �C. M. Lockwood � J. R. Stout (&)

Metabolic and Body Composition Laboratories,

Department of Health and Exercise Science,

University of Oklahoma, 115 Huston Huffman Center,

1401 Asp Ave, Norman, OK 73019, USA

e-mail: jrstout@ou.edu

A. A. Walter � J. T. Cramer � T. W. Beck

Biophysics Laboratory, Department of Health and Exercise

Science, University of Oklahoma, Norman, OK, USA

J. M. Eckerson

Department of Exercise Science and Athletic Training,

Creighton University, Omaha, NE, USA

123

Eur J Appl Physiol (2009) 105:119–130

DOI 10.1007/s00421-008-0881-9

convenience, and ease of use. Conflicting findings regard-

ing the validity of the equations associated with the

laboratory methods mentioned above (Arngrimsson et al.

2000; Ballard et al. 2004; Clasey et al. 1999; Economos

et al. 1997; Ellis 2000; Genton et al. 2002; Oldroyd et al.

2003; Silva et al. 2006; Tylavsky et al. 2003; Van Der

Ploeg et al. 2003; Vescovi et al. 2002; Wang et al. 1998;

Williams et al. 2006; Withers et al. 1998) may also dis-

courage coaches and trainers from using these techniques

to assess the body composition of their athletes.

In the past, two compartment (2C) models, such as HW,

served as the criterion method to validate various field and

laboratory techniques (Vescovi et al. 2002). More recently,

however, multi-compartment models have become more

widely accepted as the preferred criterion method, and

equations have been developed using as many as six

compartments of body composition. In addition, Wang

et al. (1998) have reported that, when compared to a six-

compartment (6C) model, the most accurate equations for

estimating body composition are four-compartment models

(4C) and three-compartment (3C) models that include an

estimate of total body water (TBW). Recently, Wang et al.

(2002) developed a five-compartment model (5C) and a 4C

model that include TBW and also account for both bone

mineral and soft tissue mineral components, which may

provide even more accurate estimates of body composition

compared to earlier 4C models that use older estimations of

soft tissue mineral (Baumgartner et al. 1991; Friedl et al.

1992; Heymsfield et al. 1996). In an earlier study, Wang

et al. (1998) also reported that the Lohman 3C model

(Lohman 1986), Siri 2C (Siri 1961), Brozek 2C (Brozek

et al. 1963), and DXA produced similar estimates of body

composition because each of these models assume a con-

stant fat-free mass hydration (0.73%). In agreement with

Wang et al. (1998), several other studies have shown that,

when compared to multi-compartment models, DXA and

2C models, such as HW and ADP, provide valid group

estimates of percent body fat (%fat); however, the error

associated with the individual results may be too large to be

of practical value (Clasey et al. 1999; Fields et al. 2001;

Wang et al. 1998; Withers et al. 1998). Furthermore, Prior

et al. (2001) reported that athletes may have ‘‘systematic

deviations’’ in their density of fat free mass (FFM)

(=1.1 g/cm3) which, in turn, resulted in large group mean

errors (2–5%fat) when they compared a 2C model to a 4C

model. Moreover, Prior et al. (2001) suggests that 2C

models may not be accurate for estimating %fat in athletic

populations.

Although multi-compartment models are widely con-

sidered to be more accurate than 2C models for estimating

body composition, questions remain regarding whether 5C

and 6C models add any meaningful accuracy to 3C and 4C

models, and whether potential deviations in the density of

the FFM in athletes preclude the use of 2C models for

assessing body composition in this population. Moreover,

there is a lack of research in college age female athletes

and laboratory methods compared to a multiple-compart-

ment model, and, to the best of our knowledge, there have

been no validation studies comparing recently updated

DXA software, the Lunar Prodigy Advance DXA model, or

ADP in this population to a 4C or 5C model. Therefore, the

purpose of the current investigation was to compare %fat

estimates from the Wang-5C (Wang et al. 2002) model to

estimates from the Wang-4C (Wang et al. 2002), Siri-3C

(Siri 1961), Lohman-3C (Lohman 1986), Siri 2C (Siri

1961), Brozek 2C (Brozek et al. 1963), and DXA in

female, NCAA Division I, Caucasian athletes, and to

compare the density of FFM, and water, mineral, and

protein content in this population to the density of FFM

derived from a ‘‘reference body’’(Brozek et al. 1963). In

addition, this investigation sought to compare the Siri 2C

(Siri 1961) and Brozek 2C (Brozek et al. 1963) models as

calculated using both ADP and HW.

Methods

Study participants

Twenty-nine female, NCAA Division I, Caucasian athletes

�x� SD ¼ 20� 1 yearð Þ volunteered to participate in the

study. Descriptive characteristics of the subjects can be

found in Table 1. All 29 subjects were currently partici-

pating in one of three collegiate level sports: volleyball

(n = 7), softball (n = 16), or track and field (n = 6). Next to

professional athletes, NCAA Division I collegiate athletes

are of the highest caliber. Moreover, all athletes in the

current investigation were varsity level and currently

enrolled in an offseason strength and conditioning program

and had been training and participating in said sport for at

least one year prior to study participation. The purpose of

the study and a description of the testing protocol were

explained to each subject. Additionally, the study was

approved by the Institutional Review Board for human

subjects, and written informed consent was obtained from

all subjects prior to testing.

Table 1 Descriptive characteristics of subjects (n = 29)

Variable Mean SD Range

Age (years) 20 1 18–23

Height (cm) 172.0 6.0 159.0–182.0

Body mass (kg) 73.63 10.66 56.50–104.55

FFM (kg) 54.93 5.76 44.42–70.49

FM (kg) 18.71 5.86 8.99–34.06

120 Eur J Appl Physiol (2009) 105:119–130

123

Measurements

TBW was assessed using bioimpedance spectroscopy (BIS)

and incorporated into the Siri-3C (Siri 1961), and the 4C

and the 5C equations of Wang et al. (2002). Both ADP and

DXA procedures were performed prior to HW in no par-

ticular order. All body composition assessments were

performed on the same day following a 12-h fast (ad libi-

tum water intake was allowed), and the subjects were also

instructed to avoid exercise for at least 12 h prior to testing.

Height (HT) was measured to the nearest 0.5 cm using a

calibrated stadiometer, and body mass (BM) was deter-

mined using a calibrated clinical scale to nearest 0.01 kg

with subjects wearing a tight-fitting bathing suit or com-

pression shorts and sports bra.

Dual-energy X-ray absorptiometry

Dual-energy X-ray absorptiometry (enCORETM

2006,

v.10.50.086, Lunar Prodigy Advance, Madison, WI) was

used to estimate total body bone mineral content (BMC)

and %fat. BMC was then converted to total body bone

mineral (Mo) (Wang-4 and 5C) and total body mineral (M)

(Lohman-3C) using the following equations (Heymsfield

et al. 1989; Lohman 1986; Wang et al. 1998):

Mo ¼ total body BMC kgð Þ � 1:0436

M ¼ Mo � 1:235

.

Each day prior to testing, a quality assurance phantom

was performed to ensure calibration. Before each test, the

subjects’ HT, BM, sex, and race were entered into the

computer program. The subjects were positioned supine on

the DXA table with hands pronated and flat on the table.

Total body mode was selected for each scan, and scanning

thickness was determined by the DXA software. All DXA

scans were performed by a certified enCORETM

software

operator. Previous test-retest scans of 11 men and women

measured 24–48 h apart for Mo and %fat produced a

standard error of measurement (SEM) of 0.05 kg and

0.75%fat with ICC’s greater than 0.99, respectively.

Total body water

Bioimpedance spectroscopy was used to estimate TBW

following the procedures recommended by the manufac-

turer (ImpTM SFB7, ImpediMed Limited, Queensland,

Australia). This technique, explained elsewhere (Matthie

et al. 1998), uses a range of frequencies, encompassing

both low and high ranges that allow electrical current to

pass around and through each cell, and has produced valid

estimates of TBW when compared to a criterion method,

such as deuterium oxide (Matthie et al. 1998; Moon et al.

2008; Van Loan et al. 1993). Results from our laboratory

have also shown that the BIS device used in the current

investigation is a valid measurement tool for determining

TBW in healthy females when compared to deuterium

oxide (Moon et al. 2008). Furthermore, BIS has been used

to assess TBW for multi-compartment equations in previ-

ous validation studies (Minderico et al. 2007; Moon et al.

2007; Sardinha et al. 2003). Measurement of TBW was

taken while the subject was lying in a supine position on a

table with arms C308 away from the torso and legs sepa-

rated. Electrodes were placed at the distal ends of each

subject’s right hand and foot following the manufacturer

guidelines. Prior to electrode placement, excess body hair

was removed, and the skin was cleaned with alcohol at

each site. The average of two trials within ±0.05 L was

used as the representative TBW. Prior to analysis, each

subject’s HT, BM, age, and sex were entered into the BIS

device. The BIS utilized 256 frequencies internal to the

device to estimate TBW. The TBW estimate was then used

to estimate total body soft tissue mineral (Ms) using the

equation from Wang et al. (2002):

Ms ¼ 0:882� 12:9 TBWð Þð Þ þ 37:9½ �=1;000

Previous test-retest measurements of 11 men and

women measured 24–48 h apart for TBW using the

ImpTM SFB7 BIS produced a SEM of 0.48 L and an ICC

greater than 0.99.

Hydrostatic weighing

Body volume (BV) was assessed from HW as previously

described by our laboratory and others (Clasey et al. 1999;

Moon et al. 2007; Prior et al. 2001; Wang et al. 1998).

Residual volume was determined with the subject in a

seated position using the oxygen dilution method of Wil-

more et al. (1980) via a metabolic cart with residual

volume software (True One 2400�, Parvo-Medics, Inc.

Sandy, UT.). Subjects completed a minimum of two trials

and the average of the closest two trials within 5% was

used to represent residual volume.

Underwater weight was measured to the nearest

0.025 kg in a submersion tank in which a PVC seat was

suspended from a calibrated Chatillon� 15-kg scale (Model

# 1315DD-H, Largo, FL.), and the average of the three

highest values (six to ten trials) was used as the represen-

tative underwater weight. Percent body fat was calculated

using the 2C equations of Siri (Siri 1961) and Brozek et al.

(1963). Previous test-retest reliability data for HW from

our laboratory indicated that, for 11 young adults

(24 ± 2.4 years) measured on separate days, the ICC was

0.99 with a SEM of 0.8%fat and 0.34 L for BV with an ICC

greater than 0.99.

Eur J Appl Physiol (2009) 105:119–130 121

123

Air-displacement plethysmography

Body volume determined from ADP was assessed using the

BOD POD�, which was calibrated before each test using

the manufacturer’s instructions with the chamber empty

and using a cylinder of known volume (49.558 L). The

subject, wearing a swimming cap and tight-fitting bathing

suit or compression shorts and sports bra, entered and sat in

the fiberglass chamber. All ADP measurements were per-

formed by a BOD POD� certified investigator. Percent

body fat was calculated from ADP using the 2C equations

of Siri (1961) and Brozek et al. (1963). Previous test-retest

reliability data for ADP from our laboratory indicated that,

for 14 young adults (24 ± 3 years) measured on separate

days, the ICC was 0.99 with a SEM of 0.47%fat, which is

similar to the ICC reported by Fields et al. (2001)

(ICC = 0.98).

Five-compartment model calculations

The criterion estimate of %fat was calculated using the

5C model described by Wang et al. (2002), which

includes measurements of BV, TBW, Mo, Ms, and BM.

Fat mass (FM) and %fat were derived using the following

equations:

FM kgð Þ ¼ 2:748 BVð Þ � 0:715 TBWð Þ þ 1:129 Moð Þþ 1:222 Msð Þ � 2:051 BMð Þ

%fat ¼ FM=BMð Þ � 100

Fat-free mass calculations

Fat-free mass (FFM) was calculated by subtracting FM

from BM, while density of FFM was calculated using the

following equation (Wang et al. 2002):

FFM density ¼ 1= TBW=0:9937ð Þ þ Mo=2:982ð Þ½þ Ms=3:317ð Þ þ Protein=1:34ð Þ�

Protein ¼ BM� BF�Mo �Ms � TBW

Propagation of error

Although multi-component models are recommended over

2C models for assessing and tracking changes in body

composition, the potential propagation of errors due to the

inherent measurement error of each device used to assess

each variable may offset the improved accuracy of 5C

model estimates of body composition (Wang et al. 2005).

Wang et al. (2005) suggested calculating the propagated

error, sometimes referred to as the total error of measure-

ment (TEM) (Moon et al. 2007) to account for the accuracy

of the 4C equation. In the current study, the SEM from the

reliability data for the measurement of BV, TBW, and Mo

were used to calculate propagated errors for %fat using the

following equation:

5C TEM ¼ TBW SEM2þHW BV SEM2þMo SEM2� �1=2

5C TEM ¼ 0:482 þ 0:342 þ 0:052� �1=2

The results showed that the TEM was 0.59%fat, which

is similar (\1%fat) to values reported for 4C and 5C

models in other laboratories (0.70–0.89%fat) (Silva et al.

2006; Withers et al. 1998).

Equations used for validation

The following equations were used to calculate FM and

were converted to %fat (Wang et al. 1998):

Wang-4C (Wang et al. 2002):

FM (kg) ¼ 2.748(BV)� 0.699(TBW) + 1.129ðMoÞ� 2.051(BM)

Siri-3C (Siri 1961):

FM (kg) = 2.118(BV)� 0.78(TBW)� 1.351(BM)

Lohman-3C (Lohman 1986, 1998):

FM (kg)¼ 6.386(BV) + 3.916[total body mineral (M)]

� 6.09(BM)

Siri 2C (HW-Siri, ADP-Siri) (Siri 1961):

FM (kg) = 4.95(BV)� 4.50(BM)

Brozek 2C (HW-Brozek, ADP-Brozek) (Brozek et al. 1963):

FM (kg) = 4.570(BV)� 4.142(BM)

%fat ¼ FM=BMð Þ � 100

Statistical methods

The validity of the %fat estimates (Wang-4C, Siri-3C,

Lohman-3C, HW-Brozek, ADP-Brozek, HW-Siri, ADP-

Siri, and DXA) was based upon the evaluation of predicted

values versus the criterion or actual value from the 5C

model by calculating the constant error (CE = actual

(5C) - predicted %fat), r value, standard error of esti-

mate SEE ¼ SDffiffiffiffiffiffiffiffiffiffiffiffiffi1� r2p� �

; and total error TE ¼ðffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiR½predicted� actual�

p 2=nÞ (Heyward and Wagner 2004).

The mean difference (CE) between the predicted and actual

(5C) %fat values was analyzed using dependent t-tests with

the Bonferroni alpha adjustment (P B 0.00625) (Keppel

and Wickens 2004). One-sample t-tests were performed to

122 Eur J Appl Physiol (2009) 105:119–130

123

determine if there was a difference between FFM density

and water, mineral, and protein content relative to the

values derived from the ‘‘reference body’’ of Brozek et al.

(1963). Differences between all variables regarding sport

were analyzed using a one-way ANOVA with a Tukey post-

hoc analysis. In addition, the method of Bland and Altman

was used to identify the 95% limits of agreement between

the criterion and predicted %fat values (Bland and Altman

1986).

Results

With the exception of height between softball and volleyball

athletes (P \ 0.05), all dependent variables were not sig-

nificantly different between athlete and sport. Therefore,

athletes from all sports were combined to include a single

sample of 29 Division I Caucasian female athletes. Com-

pared to the ‘‘reference body’’ cadavers of Brozek et al.

(1963), the Caucasian female athletes used in the current

study demonstrated a similar (non-significant, P [ 0.05)

FFM density, water, mineral, and protein content (Table 2).

Table 3 presents the results of the validation analyses for

each equation. Compared to the 5C model, the Wang-4C,

and Siri-3C produced the most accurate estimates of %fat

(r [ 0.99, SEE \ 0.39%fat, TE \ 0.40%fat and an agree-

ment of less than -0.08 ± 0.78%fat, CE/Bias ± 1.96 SD).

Of the remaining methods, DXA produced the largest CE

(-3.71%fat), which was significantly different than the 5C

(P \ 0.00625). All methods, with the exception of DXA,

produced CE values \0.82%fat, which are similar

(\1.0%fat) or less than the TEM reported by this laboratory

(0.59%fat) and others (0.70–0.89%fat) (Silva et al. 2006;

Withers et al. 1998). Of the multiple-compartment models,

the Lohman-3C equation produced the highest SEE

(1.91%fat), TE (2.54%fat), and CE (-0.72%fat) values,

with the lowest r value (0.91) compared to the 5C model. Of

the 2C models, the Brozek et al. (1963) equation produced

lower TE values (HW-Brozek = 2.22%fat, ADP-Brozek =

2.65%fat) compared to the Siri (1961) equation (HW-Siri =

2.33%fat, ADP-Siri = 3.07%fat) for both HW and ADP.

Although multi-compartment models, particularly those

that include a measure of TBW (Wang et al. 1998), are

preferred over 2C models for estimating %fat, many uni-

versity laboratories are limited to the use of 2C models and,

perhaps, DXA to assess the body composition of their

athletes. Both CE values for ADP-Brozek (-0.09%fat) and

HW-Brozek (0.62%fat) were less than 1.0%fat. The lowest

validity coefficient (r) was 0.89 (ADP-Brozek), and the

highest was 0.93 (DXA), while the SEE values ranged from

1.78%fat (DXA) to 2.19%fat (ADP-Brozek). The Brozek

et al. (1963) equation using BV estimates from either HW

or ADP resulted in TE values that were B2.65%fat, while

Table 2 Fat-free mass characteristics of subjects (n = 29)

Variable Mean SD

5C Model FFM Density (g/cm3) 1.100 0.008

Water/FFM (%) 73.69 2.18

Mineral/FFM (%) 6.88 0.43

Bone 5.97 0.43

Soft Tissue 0.91 0.02

Protein/FFM (%) 19.41 2.13

Reference body FFM Density (g/cm3) 1.100

Water/FFM (%) 73.80

Mineral/FFM (%) 6.80

Protein/FFM (%) 19.40

Table 3 Validation of all methods for predicting % body fat compared to the Wang 5C

Method �X � SDð Þ Slope Intercept r SEE TE Agreement

CE/Bias ± (1.96 9 SD) Upper Limits Lower Limits Trend

Wang-4C 24.98 ± 4.63 1.000 -0.06 [0.99 0.01 0.05 -0.05 ± 0.02 -0.03 -0.07 -0.00

Siri-3C 25.00 ± 4.50 1.027 -0.74 [0.99 0.38 0.39 -0.07 ± 0.77 0.70 -0.84 0.03

Lohman-3C 25.65 ± 5.85 0.724 6.35 0.91 1.91 2.54 -0.72 ± 4.86 4.14 -5.58 -0.24**

HW-Brozek 24.31 ± 5.03 0.832 4.71 0.90 2.04 2.22 0.62 ± 4.26 4.88 -3.65 -0.09

HW-Siri 24.97 ± 5.44 0.767 5.76 0.90 2.04 2.33 -0.04 ± 4.64 4.60 -4.68 -0.17

ADP-Brozek 25.02 ± 5.73 0.716 7.01 0.89 2.19 2.65 -0.09 ± 5.29 5.19 -5.38 -0.22**

ADP-Siri 25.74 ± 6.21 0.661 7.91 0.89 2.19 3.07 -0.81 ± 5.90 5.09 -6.71 -0.31**

DXA 28.63 ± 7.05 0.609 7.49 0.93 1.78 4.90 -3.71* ± 6.39 2.68 -10.10 -0.43**

5C 24.93 ± 4.63

*Represents significance at (P B 0.00625), **represents significance at (P B 0.05), HW hydrostatic weighing, ADP air displacement plethys-

mography via the BOD POD�, DXA dual-energy X-ray absorptiometry, CE/Bias constant (mean) error, TE total error, SEE standard error of

estimate, r = Pearson product-moment correlation coefficient, Limits 95% limits of agreement [CE ± 1.96 SD of residual scores (predicted-

actual)], Trend relationship between the difference of the criterion and laboratory method and the mean of both methods

Eur J Appl Physiol (2009) 105:119–130 123

123

DXA produced an unacceptable TE value of 4.90%fat. The

results for the regression analysis between the 5C model

and laboratory methods are depicted in Fig. 1.

The individual estimates for %fat derived from the

equations presented in Table 3 were analyzed using Bland

and Altman plots, and the results are shown in Figs. 2, 3, 4.

The 95% limits of agreement (CE ± 1.96 SD of residual

scores (predicted–actual)) were large for DXA (Fig. 4)

(-3.71 ± 6.39%fat, -10.10 to 2.68%fat), while ADP-

Brozek and HW-Brozek produced smaller limits of agree-

ment (B0.62 ± 5.29%fat, less than or equal to -5.38 to

5.19%fat) (Fig. 3). Additionally, DXA and ADP-Brozek

produced a significant trend, indicating an overestimation

of %fat as total body fat increased (Figs. 3, 4). Of the

multiple-compartment models, the individual estimations

of %fat for the Lohman-3C model were similar to both the

ADP- and HW-Brozek 2C models (Fig. 3), while the

Siri-3C and Wang-4C models produced individual esti-

mations of %fat with greater accuracy (Fig. 2).

Discussion

The primary findings of the current study indicated that the

4C and 5C equations of Wang et al. (2002) resulted in

nearly identical results (Table 3, Fig. 2), and suggest that

the 5C model provides minimal added accuracy compared

to the most recent 4C model of Wang et al. (2002). The

Siri-3C (Siri 1961) model was the next most accurate

equation with an error of less than or equal to

-0.07 ± 0.77%fat (CE ± 1.96 SD) (Fig. 2); therefore, the

Wang-4C and Siri-3C models are recommended for esti-

mating %fat in this population. Specifically, of the 29

subjects, no subject produced different whole number

values (\0.1%fat) when comparing the Wang-5C to the

Wang-4C model, indicating that these models produce

virtually identical values. Nine of the 29 subjects produced

different whole number values when comparing the Wang

5C to the Siri-3C model; however, these differences were

less than 0.85%fat. Therefore, data suggests that, due to the

minimal difference between the Wang-5C compared to the

Wang-4C and Siri-3C models and the near-perfect corre-

lation (r [ 0.99), these three models can be considered

criterion methods for estimating %fat in this population.

Additionally, due to these nominal errors, any one of these

three methods can potentially be used as an accurate rou-

tine follow-up method to track changes in body

composition during training and over time. However, the

Siri-3C model is more practical than the Wang-4C model

because there is no required DXA measurement for BMC.

Therefore, considering relative degrees of cost, radiation

exposure, technical skill, and simplicity, the Siri-3C model

is suggested for routine use in female NCAA Division I,

Caucasian athletes. The Lohman-3C equation produced

similar accuracy to both ADP- and HW-Brozek 2C models

(Fig. 3), which suggests that the inclusion of BMC in a 2C

model (Siri 1961; Brozek et al. 1963) does not improve the

accuracy of %fat estimations in Division I NCAA female

athletes. Although the Siri (1961) and the Brozek et al.

(1963) 2C models resulted in similar and valid results, the

TE values associated with the Brozek et al. (1963) 2C

model were lower when using either HW or ADP to assess

Fig. 1 Regression analysis of percent body fat (%fat) estimations by

laboratory methods [hydrostatic weighing (HW), air displacement

plethysmography via the BOD POD� (ADP), dual-energy X-ray

absorptiometry (DXA)], and the criterion five-compartment model

(5C). The solid line represents the line of identity with a slope of one

and a y-intercept of zero

Fig. 2 Bland and Altman plots

of percent body fat (%fat)

estimations by laboratory

methods. The solid linesrepresent the upper and lower

limits of agreement

(±1.96 SD). The dotted/dashedline represents the constant error

or mean bias. The dashedregression line represents the

trend between the differences of

methods and the mean of both

methods

124 Eur J Appl Physiol (2009) 105:119–130

123

BV (Table 3); thus, the Brozek et al. (1963) 2C model is

recommended over the Siri 2C model for use in this pop-

ulation. When comparing HW-Brozek to ADP-Brozek,

both produced similar values, but HW-Brozek was more

accurate, and ADP-Brozek produced a significant trend

(Table 3). Therefore, HW-Brozek is recommended over

ADP-Brozek due to the slightly improved accuracy. Due to

these diverse error values and the number of methods and

models used in the current investigation, all models have

been subjectively ranked based on accuracy and are

depicted in Table 4.

In the current study, DXA resulted in %fat values that

were considered ‘‘fair’’ (Heyward and Wagner 2004)

(TE = 4.90%fat) and produced the largest limits of

agreement (Table 3, Fig. 4). Therefore, caution is war-

ranted when utilizing the DXA model used in the current

investigation (enCORETM

2006, v.10.50.086, Lunar Prod-

igy Advance) as a criterion or laboratory method to

estimate %fat in Caucasian female collegiate athletes.

The values for the density of FFM, water/FFM, mineral/

FFM, and protein/FFM derived for the female athletes in the

current investigation were not significantly different than

the ‘‘reference body’’ reported in the cadaver analysis of

Brozek et al. (1963) (Table 2). These findings are in

agreement with a similar study by Prior et al. (2001) who

used estimates of body composition from a 4C model to

determine the density of the FFM in female athletes and

reported that, when all female athletes (swimmers, gym-

nasts, volleyball, basketball, and track and field) were

combined, density of FFM (1.099 ± 0.011 g/cm3) was not

significantly different (P [ 0.05) compared to the ‘‘refer-

ence body’’ of Brozek et al. (1963) (1.100 g/cm3). However,

in contrast to the investigation by Prior et al. (2001), the

current study produced similar, non-significant values for

mineral/FFM and protein/FFM compared to the cadaver

analysis of Brozek et al. (1963). Discrepancies in the find-

ings of the current investigation compared to those of Prior

et al. (2001) may be due to differences in the samples of

female athletes. For example, Prior et al. (2001) found that

swimmers and gymnasts had significantly different FFM

densities than those of the ‘‘reference body’’ of Brozek et al.

(1963) and that swimmers had a significantly lower mineral/

FFM density and gymnasts had a higher protein/FFM den-

sity and a lower water/FFM density compared to the same

‘‘reference body’’. These varying densities could have

contributed to the overall group mean difference for

mineral/FFM and protein/FFM densities compared to the

current investigation that did not include gymnasts or

swimmers. In addition, Prior et al. (2001) used an older 4C

model (Lohman 1986) rather than the most recent 5C model,

which also may have contributed to the difference in results.

Air-displacement plethysmography

Due to its many salient features and improved subject

compliance, ADP measured via the BOD POD� provides

an attractive alternative to HW for estimating BV. Because

it is a relatively new technique, the validity ADP for esti-

mating %fat compared to a multi-compartment model has

not been fully investigated, and the results of a few studies

that have examined its validity in college female athletes

have resulted in conflicting findings (Ballard et al. 2004;

Fig. 3 Bland and Altman plots

of percent body fat (%fat)

estimations by laboratory

methods. The solid linesrepresent the upper and lower

limits of agreement

(±1.96 SD). The dotted/dashedline represents the constant error

or mean bias. The dashedregression line represents the

trend between the differences of

methods and the mean of both

methods

Fig. 4 Bland and Altman plots of percent body fat (%fat) estimations

by laboratory methods. The solid lines represent the upper and lower

limits of agreement (±1.96 SD). The dotted/dashed line represents

the constant error or mean bias. The dashed regression line represents

the trend between the differences of methods and the mean of both

methods

Eur J Appl Physiol (2009) 105:119–130 125

123

Vescovi et al. 2002). For example, Vescovi et al. (2002)

compared estimates of %fat from ADP to HW in female

college athletes and reported that ADP over-predicted %fat

by as much as 8% in subjects who were greater than

18.3%fat and over-predicted by as much as 16% in leaner

subjects. In contrast, the data from the current study

showed that ADP resulted in an overestimation of %fat as

%fat increased, indicated by a significant trend (trend =

-0.22, P \ 0.05; Table 2, Fig. 3) and underestimated %fat

in leaner subjects.

Similar to the current investigation, Silva et al. (2006)

used the 5C model of Wang et al. (2002) to validate the use

of ADP using the Siri (1961) equation in 32 Portuguese

high school-aged female athletes (mean age ± SD =

15.1 ± 0.3 years) and reported SEE and TE values of 1.55

and 2.23%fat, respectively, which is in agreement with the

findings of the current study using the Brozek et al. (1963)

equation (SEE = 2.19%; TE = 2.65%). In addition, Silva

et al. (2006) reported a high r value (0.98) and a non-

significant and low CE/Bias (1.7%fat), which is comparable

to the results of the current study (r = 0.89; non-significant

CE/Bias = -0.09%fat). Although the limits of agreement

in the current study were larger (5.19 to -5.38%fat) (Fig. 3)

compared to the values reported by Silva et al. (2006) (4.6

to -1.3%fat), these findings suggest that ADP is a valid

procedure for estimating %fat in female athletes ranging

from high school- to college-aged when compared to the 5C

model. The TE and SEE values for ADP in the present study

are also similar to those reported by Fields et al. (2001)

(TE = 2.3%fat, SEE = 2.68%fat), who compared esti-

mates of %fat using ADP to a 4C (Baumgartner et al. 1991)

model in older (mean age = 32.8 years), non-athletic adult

females and concluded that ADP is a valid method for

estimating %fat in a diverse female population. However,

because the wide limits of agreement in the current inves-

tigation indicate that ADP may overestimate %fat by as

much as 5.38%fat and underestimate by as much as

5.19%fat, ADP should be used with caution when esti-

mating %fat in individuals but may be a valid device for use

in small groups or teams when a mean value is desired.

Hydrostatic weighing

To the best of our knowledge, this is the first study to

validate %fat estimates from HW in college female

athletes compared to the 5C model, and previous inves-

tigations that have compared HW to 4C models have

resulted in conflicting findings (Arngrimsson et al. 2000;

Prior et al. 2001; Withers et al. 1998). Withers et al.

(1998) compared %fat estimates from HW to the 4C

model of Heymsfield et al. (1996) in female college-age

middle-distance runners and triathletes and determined

that HW underestimated %fat (CE = 2.8%fat), which is

in contrast to the current findings using a 5C model

(CE = 0.62%fat). Discrepancies in these results could be

attributed to the small number of subjects used in the

study by Withers et al. (1998) (n = 12), the use of a

different 4C model, and the finding that their subjects

were leaner (16.4 ± 2.4%fat) than those used in the

current study (24.93 ± 4.63%fat). However, Arngrimsson

et al. (2000) compared an older 4C model (Lohman 1986)

to HW in middle and long distance runners (n = 10) and

reported a CE value (0.10%fat) that was comparable to

the current findings (CE = 0.62%fat). The subjects in the

investigation by Arngrimsson et al. (2000) were leaner

(21.1 ± 4.0%fat) but more similar to the subjects in the

current investigation than those reported by Withers et al.

(1998), which may explain, in part, the comparable

findings with Arngrimsson et al. (2000). In the study

previously mentioned by Prior et al. (2001) that used 44

college female athletes, HW was also found to be a valid

laboratory technique for estimating %fat, and the reported

CE value (-0.20%fat) was similar to the current inves-

tigation. The results of several studies suggest that HW is

a valid technique for determining group estimates of %fat

in college female athletes, but, due to the wide limits of

agreement reported in this study (-3.65 to 4.88%fat), as

well as others (Arngrimsson et al. 2000; Prior et al. 2001),

caution should be used when utilizing HW alone as a

laboratory method to provide individual estimates of

%fat.

Table 4 Subjective rating for

estimating percent fat in NCAA

Division I female athletes

All methods are based on

comparison to the Wang-5C

model. TE total error, LOAlimits of agreement

Rating requirements Laboratory method Rating

(1 = Most accurate)

TE \ 1.0%fat, LOA \ 1.0%fat, no significant trend Wang-4C 1

Siri-3C 1

TE \ 2.5%fat, LOA \ 4.75%fat, no significant trend HW-Brozek 2

HW-Siri 2

TE \ 3.5%fat, LOA \ 6.0%fat, significant trend Lohman-3C 3

ADP-Brozek 3

ADP-Siri 3

TE [ 3.5%fat, LOA [ 6.0%fat, significant trend DXA Not recommended

126 Eur J Appl Physiol (2009) 105:119–130

123

Dual-energy X-ray absorptiometry

In the current study, DXA resulted in a TE value

(4.90%fat) that was too large to be of practical value. This

is in agreement with the results of the study previously

mentioned by Silva et al. (2006), who also reported a large

TE value of 4.66%fat when comparing DXA (QDR-1500

v.5.67, Hologic pencil beam mode) to the 5C model in

adolescent female athletes. In addition, Silva et al. (2006)

found a significant CE (-3.7%fat, P \ 0.05), wide limits

of agreement (-9.4%fat to 2.0%fat), and a negative trend

(-0.11), which is comparable to the findings of the current

investigation (CE = -3.71%fat, P \ 0.00625, limits =

-10.10 to 2.68%fat, trend = -0.43). Based upon the trend

data, the results from both studies indicate an overestima-

tion of %fat as %fat increases; however, the large CE (CE

greater than -3.70%fat) suggests a systematic bias to

overestimate %fat by more than 3.71%fat, regardless of

overall fatness. However, the SEE value for DXA

(1.78%fat) was lower than those for both 2C models

([2.03%fat), and the r value (0.93) was higher than both

2C models (\0.91), suggesting that the error in DXA may

only be due to this systematic bias (CE). More importantly,

the TE value produced by DXA represents errors from both

the SEE and CE values; thus, reducing the CE values

would ultimately improve the TE value.

As a means to correct this issue, Silva et al. (2006)

suggested developing calibration models for DXA group

mean %fat estimations for each population. This could

reduce the systematic bias (CE), allowing for DXA to

predict more accurately compared to multiple-compart-

ment models. Specifically, the acceptable r and SEE values

indicate that DXA could be a valid method for estimating

%fat in this population if the systematic bias (CE) is

reduced. Therefore, based on the current results, a correc-

tion factor for DXA could be utilized to improve DXA

estimated %fat values. Specifically, based on the current

results (depicted in Fig. 4), the following equation may

reduce the error in %fat estimation in female NCAA

Division I, Caucasian athletes:

%fat ¼ DXA derived %fatð Þ � 3:71

Nonetheless, this correction factor may only be

applicable in athletes with greater than 20%fat as

measured by DXA. Conversely, this correction factor may

not reduce the significant trend for DXA to overestimate

%fat as %fat increases. Therefore, the following regression

equation was developed to reduce both the CE and the

significant trend:

%fat ¼ 0:609 DXA derived %fatð Þ þ 7:49

This new %fat equation for DXA may be used in a wide

range of female athletes that including all level of body

fatness. Thus, when utilizing the current DXA model and

software for estimating %fat in this population, the above

regression equation is suggested.

However, other studies using athletic populations

reported that DXA underestimated %fat by 1.3 (Withers

et al. 1998) and 4.0%fat (Arngrimsson et al. 2000) com-

pared to 4C models; however, the athletes used in these

studies were leaner (16.4%fat Withers et al. 1998 and

21.1%fat Arngrimsson et al. 2000) than the participants in

the current investigation (24.93%fat). Estimates of %fat

from DXA in non-athletic females compared to multi-

compartment models have also been shown to be inferior to

values estimated from other laboratory techniques, such as

HW (Clasey et al. 1999). Furthermore, a number of studies

(Clasey et al. 1999; Silva et al. 2006; Van Der Ploeg et al.

2003; Wang et al. 1998; Williams et al. 2006; Withers et al.

1998) using multi-compartment models as criterion mea-

sures support the current findings that DXA may not be a

valid procedure when estimating %fat in college female

athletes.

Past literature has identified several problems with the

use of DXA for body composition assessment that could

explain to the poor validity found in the current investi-

gation (Economos et al. 1997; Genton et al. 2002; Oldroyd

et al. 2003; Tylavsky et al. 2003). One source of error

could be attributed to the software package used to analyze

the DXA scan and the DXA scanner itself. Although new

software and scanners are continually being introduced to

improve body composition measurements, the updated

software and scanners do not always result in more accu-

rate measurements. For example, studies have shown that

different DXA machines and software packages result in

dissonant body composition values (Genton et al. 2002;

Tylavsky et al. 2003). Therefore, the software used in the

current investigation (enCORETM

2006, v. 10.50.086) and

the fan-beam model (Lunar Prodigy Advance) DXA may

have contributed to the lack of agreement between DXA

and the 5C model. Specifically, disparities in potential error

exist among different DXA machines and models

(Economos et al. 1997; Genton et al. 2002; Oldroyd et al.

2003; Tylavsky et al. 2003), making comparisons between

investigations difficult and suggesting the need for stan-

dardized cross-calibration procedures (Economos et al.

1997; Oldroyd et al. 2003).

Furthermore, another large source of error could be the

inability of DXA to measure soft tissue overlying the bone.

DXA models assume that the soft tissue adjacent to the

bone has the same tissue composition as the soft tissue over

the bone. In muscular athletes, this assumption may

increase the error associated with all soft tissue measure-

ments due to larger variations of soft tissue over bone

associated with various distributions of muscle and fat.

This hypothesis could explain why the error in female

Eur J Appl Physiol (2009) 105:119–130 127

123

college athletes was larger than in high school-age athletes

as described by Silva et al. (2006). Thus, college athletes

are more physically developed and may have more muscle

mass associated with the soft tissue measurement, which

may explain the systematic overestimation in %fat as total

tissue (BM) increased. Specifically, based on the current

findings, it is hypothesized that the more total mass that is

estimated over bone, the larger the potential errors are in

estimating that tissue, regardless of the ratio of FFM and

fat next to the bone.

In addition, DXA is described as a 3C model because it

estimates bone, fat, and other soft tissue, yet, the final

analysis of fat is based on a 2C model of fat and all other

tissue (Ellis 2000). This 2C model has similar errors to HW

and ADP (Fig. 3), due to assumptions of FFM hydration

and TBW. However, Lohman et al. (2000) states that the

hydration of FFM is not a major factor in the agreement

between DXA and multi-compartment models and techni-

cal errors are introduced when the actual BM of the subject

does not match the sum of the tissue mass measured by the

DXA.

Based upon the unacceptably high TE values associated

with DXA compared to the 5C model, and a non-significant

FFM hydration status (73.69% compared to 73.8%), we

analyzed the tissue mass values and actual BM values to

determine if the variations in these values were the source

of error. Compared to actual weight, DXA produced a non-

significant CE (CE = -0.03 kg, P [ 0.05), indicating that

this was not the source of error, and, therefore, the inability

of DXA to estimate %fat is most likely due to several

factors rather than one specific element. Based upon these

results, the DXA model and software used in the current

investigation should be used with caution when estimating

%fat in college female athletes, even when utilizing the

above regression equation, due to the myriad concerns with

the DXA method and lack of research supporting the

validity of DXA for predicting %fat.

Menstrual status

Additional sources of error in estimating %fat in women

can be attributed to fluctuations of TBW through various

stages of the menstrual cycle (Bunt et al. 1989). Due to

potential increases in error, this fluctuation is particularly

important in the use of laboratory methods that assume a

constant FFM hydration, such as HW, ADP, and DXA

(Bunt et al. 1989). Bunt et al. (1989) determined that

errors can be reduced if subjects are measured outside of

peak BM periods. Furthermore, Bunt et al. (1989) stated

that the day of the menstrual cycle associated with peak

BM was not the same for all females. This understanding

may prove to be relevant when tracking changes or

assessing individual %fat. However, conducting body

composition assessments during non-peak BM days of the

month is not realistic in populations such as college ath-

letes. Therefore, menstruation status was not considered in

the current investigation, which could have increased the

error associated with HW and ADP. Nonetheless, these

methods provided valid estimations of %fat in this popu-

lation, regardless of menstrual cycle phase. More

importantly, an understanding of the limits of agreement

regarding 2C models to estimate %fat could aid in

reaching correct interpretations of %fat values estimated

by these methods. Specifically, HW-Brozek could under-

estimate %fat by 4.88%fat and overestimate %fat by as

much as 3.65%fat, while ADP-Brozek could underesti-

mate %fat by 5.19%fat and overestimate %fat by as much

as 5.38%fat.

Conclusion

The Wang-4C and Siri-3C models resulted in the lowest

TE values compared a criterion 5C model and, therefore,

are recommended for estimating %fat in Caucasian female

athletes and can be considered criterion methods. Although

the Lohman-3C equation also provided valid estimates of

%fat, the added accuracy of this equation by incorporating

BMC was negligible compared to the 2C models. There-

fore, because of the expense and training required to obtain

BMC, this model is not recommended over a 2C model that

uses either HW or ADP to determine BV.

In the current study, both 2C models produced low TE

values when compared to the 5C model of Wang et al.

(2002). However, due to the large limits of agreement, the

use of 2C models may not be appropriate when attempting

to identify individual assessments of %fat in athletic pop-

ulations due to variations in TBW and FFM hydration

(Clasey et al. 1999; Wang et al. 1998; Withers et al. 1998).

Therefore, whenever possible, multi-compartment models

should be used for assessing individual estimates of %fat or

to track changes in body composition following diet and

training interventions. Because the Wang-4C and Siri-3C

equations can be considered criterion methods, these

equations may be used in place of the 5C model. More

notably, the current results do not support the use of the

Lunar Prodigy Advance DXA with the enCORETM

2006

software version v.10.50.086 for use in estimating percent

body fat in Caucasian Division I NCAA female athletes.

However, utilizing the suggested regression equation

(%fat = 0.609(DXA derived %fat) ? 7.49) may produce

valid estimates of %fat in this population when utilizing

DXA alone.

Acknowledgments We would like to thank all of the athletes who

participated in this investigation and Vincent J. Dalbo for his efforts

128 Eur J Appl Physiol (2009) 105:119–130

123

during data collection. Additionally, we would like to thank Sarah

Cahill for her dedication and support as a strength coach.

References

Arngrimsson S, Evans EM, Saunders MJ, Ogburn CL, Lewis RD,

Cureton KJ (2000) Validation of body composition estimates in

male and female distance runners using estimates from a four-

component model. Am J Hum Biol 12:301–314. doi:10.1002/

(SICI)1520-6300(200005/06)12:3\301::AID-AJHB1[3.0.CO;

2-J

Ballard TP, Fafara L, Vukovich MD (2004) Comparison of BOD

POD and DXA in female collegiate athletes. Med Sci Sports

Exerc 36:731–735. doi:10.1249/01.MSS.0000121943.02489.2B

Baumgartner RN, Heymsfield SB, Lichtman S, Wang J, Pierson RN Jr

(1991) Body composition in elderly people: effect of criterion

estimates on predictive equations. Am J Clin Nutr 53:1345–1353

Bland JM, Altman DG (1986) Statistical methods for assessing

agreement between two methods of clinical measurement.

Lancet 1:307–310

Brozek J, Grande F, Anderson JT, Keys A (1963) Densitometric

analysis of body composition: revision of some quantitative

assumptions. Ann N Y Acad Sci 110:113–140. doi:10.1111/

j.1749-6632.1963.tb17079.x

Bunt JC, Lohman TG, Boileau RA (1989) Impact of total body water

fluctuations on estimation of body fat from body density. Med

Sci Sports Exerc 21:96–100. doi:10.1249/00005768-198902000-

00017

Clasey JL, Kanaley JA, Wideman L, Heymsfield SB, Teates CD,

Gutgesell ME, Thorner MO, Hartman ML, Weltman A (1999)

Validity of methods of body composition assessment in young

and older men and women. J Appl Physiol 86:1728–1738

Economos CD, Nelson ME, Fiatarone MA, Dallal GE, Heymsfield

SB, Wang J, Yasumara S, Ma R, Vaswani AN, Russell-Aulet M,

Pierson RN (1997) A multi-center comparison of dual energy

X-ray absorptiometers: in vivo and in vitro soft tissue measure-

ment. Eur J Clin Nutr 51:312–317. doi:10.1038/sj.ejcn.1600400

Ellis KJ (2000) Human body composition: in vivo methods. Physiol

Rev 80:649–680

Fahey TD, Akka L, Rolph R (1975) Body composition and VO2 max

of exceptional weight-trained athletes. J Appl Physiol 39:559–

561

Fields DA, Wilson GD, Gladden LB, Hunter GR, Pascoe DD, Goran

MI (2001) Comparison of the BOD POD with the four-

compartment model in adult females. Med Sci Sports Exerc

33:1605–1610. doi:10.1097/00005768-200109000-00026

Friedl KE, DeLuca JP, Marchitelli LJ, Vogel JA (1992) Reliability of

body-fat estimations from a four-compartment model by using

density, body water, and bone mineral measurements. Am J Clin

Nutr 55:764–770

Genton L, Hans D, Kyle UG, Pichard C (2002) Dual-energy X-ray

absorptiometry and body composition: differences between

devices and comparison with reference methods. Nutrition

(Burbank. Los Angeles Cty Calif 18:66–70

Heymsfield SB, Wang J, Heshka S, Kehayias JJ, Pierson RN (1989)

Dual-photon absorptiometry: comparison of bone mineral and

soft tissue mass measurements in vivo with established methods.

Am J Clin Nutr 49:1283–1289

Heymsfield SB, Lohman TG, Wang Z, Going SB (1996) Multicom-

ponent molecular level models of body composition analysis. In:

Roche AF, Heymsfield SB, Lohman TG (eds) Human body

composition. Human Kinetics, Champaign, pp 129–148

Heyward VH, Wagner DR (2004) Applied Body Composition

Assessments. Human Kinetics, Champaign

Keppel G, Wickens TD (2004) Design and Analysis: A Researchers

Handbook. Prentice Hall, Upper Saddle River

Lohman TG (1986) Applicability of body composition techniques and

constants for children and youths. Exerc Sport Sci Rev 14:325–

357. doi:10.1249/00003677-198600140-00014

Lohman TG, Harris M, Teixeira PJ, Weiss L (2000) Assessing body

composition and changes in body composition. Another look at

dual-energy X-ray absorptiometry. Ann N Y Acad Sci 904:45–

54

Matthie J, Zarowitz B, De Lorenzo A, Andreoli A, Katzarski K, Pan

G, Withers P (1998) Analytic assessment of the various

bioimpedance methods used to estimate body water. J Appl

Physiol 84:1801–1816

Minderico CS, Silva AM, Keller K, Branco TL, Martins SS, Palmeira

AL, Barata JT, Carnero EA, Rocha PM, Teixeira PJ, Sardinha

LB (2007) Usefulness of different techniques for measuring

body composition changes during weight loss in overweight and

obese women. Br J Nutr 1–10

Moon JR, Hull HR, Tobkin SE, Teramoto M, Karabulut M, Roberts

MD, Ryan ED, Kim SJ, Dalbo VJ, Walter AA, Smith AE,

Cramer JT, Stout JR (2007) Percent body fat estimations in

college women using field and laboratory methods: a three-

compartment model approach. J Int Soc Sports Nutr 4:16. doi:

10.1186/1550-2783-4-16

Moon JR, Tobkin SE, Roberts MD, Dalbo VJ, Kerksick CM, Bemben

MG, Cramer JT, Stout JR (2008) Total body water estimations in

healthy men and women using bioimpedance spectroscopy: A

deuterium oxide comparison. Nutr Metab (Lond) 5:7. doi:

10.1186/1743-7075-5-7

Oldroyd B, Smith AH, Truscott JG (2003) Cross-calibration of GE/

Lunar pencil and fan-beam dual energy densitometers–bone

mineral density and body composition studies. Eur J Clin Nutr

57:977–987. doi:10.1038/sj.ejcn.1601633

Prior BM, Modlesky CM, Evans EM, Sloniger MA, Saunders MJ,

Lewis RD, Cureton KJ (2001) Muscularity and the density of the

fat-free mass in athletes. J Appl Physiol 90:1523–1531

Sardinha LB, Silva AM, Teixeira PJ (2003) Usefulness of age-

adjusted equations to estimate body fat with air displacement

Plethysmography in male adolescent athletes. Acta Diabetol

40:51–55. doi:10.1007/s00592-003-0026-x

Silva AM, Minderico CS, Teixeira PJ, Pietrobelli A, Sardinha LB

(2006) Body fat measurement in adolescent athletes: multicom-partment molecular model comparison. Eur J Clin Nutr 60:955–

964. doi:10.1038/sj.ejcn.1602405

Siri WE (1961) Body composition from fluid spaces and density.

Analysis of methods. In: Brozek J, Henschel A (eds) Techniques

for Measuring Body Composition. National Academy of

Sciences, Washington, DC, pp 223–244

Tylavsky F, Lohman T, Blunt BA, Schoeller DA, Fuerst T, Cauley

JA, Nevitt MC, Visser M, Harris TB (2003) QDR 4500A DXA

overestimates fat-free mass compared with criterion methods.

J Appl Physiol 94:959–965

Van Der Ploeg GE, Withers RT, Laforgia J (2003) Percent body fat

via DEXA: comparison with a four-compartment model. J Appl

Physiol 94:499–506

Van Loan MD, Withers P, Matthie J, Mayclin PL (1993) Use of

bioimpedance spectroscopy to determine extracellular fluid,

intracellular fluid, total body water, and fat-free mass. Basic Life

Sci 60:67–70

Vescovi JD, Hildebrandt L, Miller W, Hammer R, Spiller A (2002)

Evaluation of the BOD POD for estimating percent fat in female

college athletes. J Strength Cond Res 16:599–605. doi:10.1519/

1533-4287(2002)016\0599:EOTBPF[2.0.CO;2

Wang ZM, Deurenberg P, Guo SS, Pietrobelli A, Wang J, Pierson RN

Jr, Heymsfield SB (1998) Six-compartment body composition

model: inter-method comparisons of total body fat measurement.

Eur J Appl Physiol (2009) 105:119–130 129

123

Int J Obes Relat Metab Disord 22:329–337. doi:10.1038/sj.ijo.

0800590

Wang Z, Pi-Sunyer FX, Kotler DP, Wielopolski L, Withers RT,

Pierson RN Jr, Heymsfield SB (2002) Multicomponent methods:

evaluation of new and traditional soft tissue mineral models by

in vivo neutron activation analysis. Am J Clin Nutr 76:968–974

Wang Z, Shen W, Withers RT, Heymsfield SB (2005) Multicompo-

nent molecular-level models of body composition analysis. In:

Heymsfield SB, Lohman TG, Wang Z, Going SB (eds) Human

body composition. Human Kinetics, Champaign, pp 163–176

Williams JE, Wells JC, Wilson CM, Haroun D, Lucas A, Fewtrell MS

(2006) Evaluation of Lunar Prodigy dual-energy X-ray

absorptiometry for assessing body composition in healthy

persons and patients by comparison with the criterion 4-compo-

nent model. Am J Clin Nutr 83:1047–1054

Wilmore JH, Vodak PA, Parr RB, Girandola RN, Billing JE (1980)

Further simplification of a method for determination of residual

lung volume. Med Sci Sports Exerc 12:216–218

Withers RT, LaForgia J, Pillans RK, Shipp NJ, Chatterton BE,

Schultz CG, Leaney F (1998) Comparisons of two-, three-, and

four-compartment models of body composition analysis in men

and women. J Appl Physiol 85:238–245

130 Eur J Appl Physiol (2009) 105:119–130

123