Fig. 1. Values for 51 samples exclusive of the elephant were taken from Appendix 2 in the study by Savage and colleagues (Savage et al., 2004). (A) A straight line was fitted to log-transformed data by ordinary least squares. (B) Residuals from the analysis in A are plotted against predicted values for logBMR. The parabolic pattern to the display indicates that a straight line is not an adequate function for describing values in the bivariate plot and calls into question the utility of the transformation. Departures from linearity were noted by Dodds and colleagues (Dodds et al., 2001) and Kozlowski and Konarzewski (Kozlowski and Konarzewski, 2005) in similar analyses, but both groups of investigators nonetheless continued to work with values in the logarithmic domain. (C) A two-parameter power function was fitted to arithmetic data by non-linear regression. (D) Residuals from the non-linear regression are plotted against body mass on a log scale to illustrate how variance increases with body size.