[SciPy-User] [OT] Bayesian vs. frequentist

David Cournapeau cournape@gmail....
Thu Feb 16 02:16:43 CST 2012


Le 15 févr. 2012 09:57, "Nathaniel Smith" <njs@pobox.com> a écrit :
>
> I don't understand why this is always framed as a "versus" debate.
"Bayesian methods" are the math for figuring out what to believe;
"frequentist methods" are the math for figuring if you're fooling yourself.
It makes perfect sense for engineers and in house estimates to use the
former and the FDA and scientists the latter. Different methods answer
different questions.

I believe the vs is at least partially justified, in the sense that they
are fundamentally incompatible. This is the only mathematical field I am
aware of where you have to "competing" frameworks. But a same person can
certainly use either for a particular analysis.

To be contrarian, there is a somehow famous paper from Judea Pearl (the
father of Bayesian networks), "Bayesianism and causality, or, why I am only
half-bayesian". The conversations between Gelman and Wasserman are also
helpful to show strenghts/weaknesses of each approach.

My own conclusion is that I still don't understand much about statistics.

David
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.scipy.org/pipermail/scipy-user/attachments/20120216/f28d5b5c/attachment.html 


More information about the SciPy-User mailing list