Document Type



The prediction of body composition variables from bioelectric impedance (BI) has considerable potential for use in surveys, because BI is reliable, and the equipment is portable (weight, 1.04 kg). The purpose of the present study was to determine if BI with selected anthropometric variables predicted %BF (percent body fat) accurately. Two groups of sub' jects were used from whom accurate anthropometric variables were ob­tained. The validation group of 148 healthy White adults (77 men; 71 women) aged 18 to 30 was used to formulate two parsimonious models for each sex to predict %BF from selected anthropometric variables, one with­out and one with stature2 divided by resistance (S2/R). The cross-validation group, aged 18 to 30 years (19 White men; 29 White women), was used to assess the stability of equations derived from S2/R and anthropometric vari­ables. Principal component analysis applied to 16 potential predictors showed five components explained most of the variation in %BF. All possi­ble subsets regression procedure was employed to select the best equation on the basis of: (1) five predictors at most, (2) minimum root mean square error and (3) 0.1 level of significance. The multiple R2 and r.m.s.e. were not changed by the inclusion of S2/R in men. However, the inclusion of S2/R changed the R2 from 0.73 to 0.81 and the r.m.s.e. from 3.83% to 3.22% in women. Cross-validation of the equations that included S2/R showed the accuracy of prediction (coefficient of variation, 0.23 for men; 0.16 for women) was approximately the same as for the validation group. These findings indicated that the addition of S2/R to selected anthropometric variables significantly improved the prediction of %BF for women, but not for men.