Purpose: Equations that estimate percentage body fat (PBF) from anthropometrics are widely used, although most were developed in small nonrepresentative samples. No study has examined the generalizability of these equations in a nationally representative population. This study evaluated the validity of 26 sets (2 for males and 3 for females only) of published equations for PBF estimation in American adults using data from the National Health and Nutrition Examination Survey 1999–2004.
Methods: Data were from 9934 adults ages ≥20 yr. Stratified by sex, equations were evaluated against dual-energy x-ray absorptiometry using R2, root mean square error, and mean signed difference. Differential bias was evaluated by the absolute value of the discrepancy between the mean signed difference values in normal weight and obese adults.
Results: In subgroups matched to the range of age and race/ethnicity in which equations were derived, most equations had R2 values between 0.5 and 0.7 and root mean square error estimates between 3.0 and 4.0 percentage points for males and between 3.5 and 4.5 percentage points for females. Analyses in sample stratified by age, obesity status, or race/ethnicity showed that 15 of the 23 equations for males and 20 of the 24 equations for females had important differential bias of more than two percentage points. Equations that included WC performed the best in males, and those that included body mass index performed best in females. Equations using skinfold thickness performed less well in older adults.
Conclusions: Published PBF equations had moderately strong R2 values in a representative sample of American men and women, but both nondifferential and differential biases were substantial for most equations.