A regression slope of -0.75 between log(10) density and log(10) body m
ass is thought to express equivalence of energy conversion among speci
es' populations of similar taxonomic and trophic status. Using larger
sample sizes than the usual 1-3 density estimates per species, we esti
mated a regression slope of -0.71 for terrestrial mammalian carnivores
. We investigated The sampling variation in this estimate, and those d
erived from smaller intra-specific subsets, using 1000-iteration boots
trap analyses to obtain 90% confidence intervals. As expected, these w
idened, as random subsets were reduced in size, but always contained t
he postulated -0.75. However, log(10) density also declined as 3/4 of
the log(10) spatial extent of study area, and study area accounted for
virtually all of the variation in density that was previously thought
due to body mass. We removed the effect of study area by using the sp
ecies-specific regression models between density and study area to pre
dict density at a common scale of 400 km(2). These common-scale densit
ies regressed against body mass with a slope of -0.16, but separated i
nto body mass classes less than and greater than 11 kg, they produced
slopes that were not significantly different from zero. We show that t
he allometry of density could be a case of circular logic, whereby bod
y mass has influenced the investigator's choice of study area, and the
resulting scale-dependent densities are related back to body mass. To
test the allometry hypothesis, the effect of study area on density es
timates needs to be removed. This requires conducting larger-scale stu
dies of the smaller-bodied species so that all species compared are re
presented by an average study area that is near the common scale. Furt
hermore, study sites need to be selected and designed to represent, mo
re than the local detail in species' density.