Dear all,
I am analysing a case-control study with conditional logistic regression. I
am fitting different transformations of the continuous exposure variable
(derived by different weighting schemes). I want to compare the different
weighting schemes.
Now, it would be *incorrect* to choose the transformation that yields the
highest odds ratio. Is it correct to choose the transformation that yields
the best log likelihood? At first I thought so, but now I am no longer sure.
Is "giving the best log likelihood" not basically equivalent with "highest
odds ratio estimate", since the case-control status is the dependent
variable in the model?
Thanks for your help.
Angelika
******************************************************
Angelika Schaffrath Rosario
Statistician
GSF - Research Center for Environment and Health GmbH
Institute for Epidemiology
Postfach 1129, D-85758 Neuherberg near Munich, Germany
Phone ++49 89 3187-4577 Fax -4499
e-mail [log in to unmask]
|