Regression equations were developed for the estimation of percent body fat from less direct anthropometric measures. The selection of the best regressors from 17 independent variables was performed by the maximum R2 improvement method which is superior to conventional stepwise selection techniques. Due to the nonconstant variance among random subgroups of the sample for percent body fat, a weighted least squares analysis with iterations was performed. The final estimates of the regression coefficients and their associated standard errors were obtained by the jackknife method. Four sets of regression equations for estimating percent body fat, with their pure errors, are presented for boys and girls (age 18 years) and for men and women (age > 18 years). These equations can be used also to estimate total body fat and body density by applying Siri’s equation. The calculated pure errors for these are provided. These equations were applied to another set of data and the prediction root mean squared errors for percent body fat, total body fat and body density were compared with the pure errors. There was good overall concordance. There were significant positive correlations between serial values for percent body fat. This suggested the occurrence of “tracking. ” An alternative analysis, based on fitting polynomials to serial data for each individual, revealed evidence of “tracking” in boys but not in girls.
Mukherjee, Debabrata and Roche, Alex F.
"The Estimation of Percent Body Fat, Body Density and Total Body Fat by Maximum R2 Regression Equations,"
1, Article 8.
Available at: https://digitalcommons.wayne.edu/humbiol/vol56/iss1/8