The docstring for varioussizes arguments states that size is given as symbol area in square points. Checking withmatplotlib.Collections code, this argument doesn't actually calculate the symbol area, but instead is a squared scale factor. This might lead to a wrong interpretation especially if irregularly-shaped symbols are used, but it also not correct for circles. On a more general basis, is it not easier to use linear scale factor as an argument, instead of its square? I just don't see applications where the latter is more useful than the former.