Generalized equation for predicting body density of women from girth measurements

This study's purpose was to develop a generalized regression equation to predict body density in adult women. Subjects, 482 women, were hydrostatically weighed and circumference (girths) recorded for thigh, hips (buttocks), iliac, and abdomen (mean of abdomen 1 and abdomen 2). Age (range = 15-79 yr), weight (38.3-132.9 kg), and height (145.5-186.3 cm) were also recorded. Percent body fat ranged from 12.7 to 63.1%. Stepwise multiple regression was used to select the best set of predictors (from seven) of body density. Capitalization on chance was negligible due to the favorable subject to predictor ratio (57 subjects per predictor). The regression equation (N = 400) developed for predicting body density was: body density = 1.168297 - (0.002824 x abdomen) + (0.0000122098 x abdomen2) - (0.000733128 x hips) + (0.000510477 x height) - (0.000216161 x age) [SEE = 0.009486784 (4.2% body fat), R = 0.889, adjusted R2 = 0.787]. Using this equation on a cross-validation sample (N = 82) produced a predicted mean (+/- SD) of 1.016 +/- 0.017 (validation sample mean = 1.016 +/- 0.021) and a total error (SE) of 0.0082 (3.6% body fat). The use of three girth measurements, height, and age enabled us to develop regression equations to predict body density in women that are comparable in accuracy to those using skinfold calipers and, thus, are a viable alternative.