Bias and limits of agreement between hydrodensitometry, bioelectrical impedance and skinfold calipers measures of percentage body fat

Previous research has often used correlations as a statistical method to show agreement; however, this is not a valid use of the statistic. The purpose of this study was to investigate the bias and limits of agreement for three methods of estimating percentage body fat for 117 male and 114 female university athletes: hydrodensitometry (HYD), bioelectrical impedance (BIA) and skinfold calipers (SKF). The mean (SD) percentage body fat for males as assessed by HYD, BIA and SKF methods, respectively, were 13.2 (3.3)%, 14.1 (3.3)% and 13.0 (3.2)%. Female body fat measurements were 22.5 (3.9)%, 23.7 (4.3)% and 23.8 (4.2)%, respectively. Pearson product moment correlations for male and female body fat percentages between the three methods were high, ranging from 0.81 to 0.86 (P < 0.05). However, compared to the criterion measure of body fat percentage (HYD), the magnitude of agreement BIA and SKF revealed a different pattern. The mean absolute difference between HYD and BIA measurements of body fat for males was −0.8 (2.0)% fat, and between HYD and SKF was it was 0.2 (1.7)% fat. The mean absolute difference for females between HYD and BIA was −1.2 (2.5)%; for HYD and SKF it was −1.4 (2.2)%. Compared to the HYD measures for males and females, the BIA and SKF measures were as much as a 3.8% underestimation and a 6.2% overestimation of body fat. This study provides evidence that the strength of a correlation does not indicate agreement between two methods. In future, reliability and validity studies should examine the absolute differences between two variables and calculate limits of agreement around which a practitioner can appreciate the precision of the methodologies. Accepted: 26 August 1997