The purpose of this study was to compare the accuracy of four methods
to assess body composition of women. Seventy-seven Caucasian women [me
an (+/- SD) age: 31.8 +/- 8.6 years; mass: 59.5 +/- 9.1 kg; stature: 1
62.4 +/- 6.9 cm; Quetelet Index: 22.5 +/- 3.1 kg/m2] were tested for p
ercent body fat (%BF) with hydrostatic weighing (HW), near-infrared sp
ectrophotometry (NIR), bioelectrical impedance analysis (BIA), and sev
en-site skinfolds (7-SFs). Compared to %BF with HW (24.9 +/- 6.5%), an
analysis of variance revealed no mean differences (P greater-than-or-
equal-to 0.05) among %BF with NIR (26.0 +/- 5.5%), BIA (25.7 +/- 5.8%)
and 7-SFs (24.0 +/- 6.0%). The correlations between %BF with HW and N
IR, BIA and 7-SF were r = 0.47, r = 0.77, and r = 0.79, respectively (
P less-than-or-equal-to 0.05), and prediction errors (SEE) were 5.8%,
4.2%, and 4.1%. Comparisons of %BF obtained from machine readings vers
us those computed from the manufacturer's equation indicated significa
nt differences for the BIA (machine 25.7%, equation 27.8%) and NIR (ma
chine 26.0%, equation 21.5%) methods. We concluded that although the m
ean %BF differences were small among the four methods, the large SEE v
alues may allow the use of BIA and 7-SFs but not NIR. Inaccuracies of
machine readings versus equation-computed %BF indicate that BIA and NI
R variables and/or constants in the equations supplied are not identic
al to those used in machine-generated calculations.