Document Type



To consider the effects of multiple sources of error, a gener- alizability (G) study was completed on the hydrostatic weighing (HW) and skinfold (SF) method of measuring body composition, and the results were related to measurement error in prediction equations. For HW ten measure­ments were made twice on two days by two investigators. Five SFs were measured at the triceps, subscapula, suprailiac, abdomen, thigh, and calf on two days, by two investigators using two calipers. Each site was analyzed separately using a completely crossed design where all facets but calipers were random. A multiple linear regression (MLR) prediction equation was derived with the calf and abdomen as predictors of body density. The Phi- coefficient over all facets for HW was high (.972) and if only the last four repetitions were considered the Phi-coefficient was .979. When restricted to one set of ten or to the last four repetitions, Phi-coefficients were .909 and .937, respectively. SF Phi-coefficients ranged from .953 at the calf to .991 at the triceps. Restricted to three repetitions by one investigator on one day, the Phi-coefficients ranged from .885 to .954.For equations by Katch and McArdle, Sloan, and for the MLR equa­tion generated from the present study, a multiple R, corrected for attenua­tion, was computed using error terms for all measurements and restricted to one set of repetitions. Only modest increases (< .084) in multiple Rs were observed. It was concluded that HW & SF measurements are reliable, and that the limit for predictive validity of prediction equations has been approached.