Could anyone explain in plain English what the difference is between Scott's and Silverman's rules of thumb for bandwidth selection? Specifically, when is one better than the other? Is it related to the underlying distribution? Number of samples?
P.S. I am referring to the code in SciPy.