Some myths concerning statistical hypothesis testing

Sturla Molden sturla at molden_dot_net.invalid
Fri Nov 8 07:42:08 EST 2002


On Fri, 08 Nov 2002 13:24:11 +0100, Glen M. Sizemore wrote:

> "Curve fitting" attempts to do something similar, does it not? Because
> one sets the null hypothesis equal to one's quantitative hypothesis and
> then tries to reject it. No?

No. 

Classical regression analysis try to obtain the probability of getting
at least as large regression coefficients, given that that the true 
coefficients are zero.

It is just another flavour of the p-value.

There is a Bayesian alternative. In some cases (presence of multicol-
linearity) classical regression fails toatlly and classical statisticans 
try to patch this by using something they call "ridge regression".
"Ridge regression" is actually a simple case of Bayesian regression,
which always work by the way.   


Sturla Molden



More information about the Neur-sci mailing list