The Pratt-Arrow measure of absolute risk aversion, as defined by r(x} = u"{x}/u' (x}, is well known to be "invariant to linear transformations." However, this invariance property applies with respect to transformations of u and not with respect to arbitrary rescalings of x. The effects of this misunderstanding has led to ambiguity as to what actually constitutes behavior tnat is "slightly risk averse," "very risk averse," etc. The use of the coefficient in "per acre" analyses is particularly addressed.