The estimation of body density in men: are general equations general?

Regression equations for estimating body density and fatness from anthropometric measurements are usually specific to the populations from which they are derived, owing to methodological and biological factors. Some equations are, however, regarded as having general validity for subjects of different ages, levels of fatness and fitness. In this study, five such general equations were tested by comparing estimated body density (Dest) with measured body density (Dmeas) in 138 European men employed in an Italian shipyard. Body density was measured by underwater weighing with the simultaneous determination of lung residual volume. Measurements were also made of height, weight, eight skinfolds, four circumferences and seven diameters by standard techniques. Dest from the five general equations were significantly different from Dmeas (P less than 0.001). New prediction equations for estimating D were drawn up by simple and multiple regression of various combinations of skinfolds expressed as individual skinfolds, sums, log sums and the quadratic form. Equations were drawn up on a validation sample (N = 92) and tested on a cross-validation sample (N = 46). The form of skinfolds and the combination selected made little difference to the SEEs and R2 but the inclusion of age resulted in lower SEEs. Stepwise multiple regression produced an equation incorporating thorax and triceps skinfolds and age with an SEE of 6 kg/m3 and R2 of 79%. The derived equations were successfully cross-validated, Dmeas and Dest were not significantly different, ryy' and SEE were similar to those of the validation sample. When the cross-validation sample was divided into two according to age, density or VO2 max, the derived equations were applicable to the upper and lower groups for age or physical fitness but not to the groups differing in fatness. These equations should be validated further on other samples by studies performed with due regard to the methodological factors, both technical and statistical, that may result in specificity. However, the poor performance of the general equations tested and the specificity for fatness of the derived equations suggest that biological factors may preclude the development of general prediction equations for body density.