[SciPy-User] the skellam distribution

Ernest Adrogué eadrogue@gmx....
Tue Nov 10 16:38:23 CST 2009


 9/11/09 @ 14:36 (-0500), thus spake josef.pktd@gmail.com:
> I'm not sure the example is correct. You are simulating two
> independent poisson variables, so the difference skellam_var should be
> distributed as skellam with
> mu1 = lam1
> mu2 = lam2
> and theoretical rho should be zero. Or am I missing something?

Yes, rho should be zero, but in practice, there may be a
certain amount of correlation even though the variables are
theoretically independent.

If you set rho=0 when it's not actually zero, this will
result in an artificially worse fit, which sort of defeats
the purpose of this test. It would make sense to set rho=0
if we were testing whether the to Poisson variates are
independent, though.

Bye the way, I have opened a ticket here:
http://projects.scipy.org/scipy/ticket/1050

Bye.
Ernest



More information about the SciPy-User mailing list