Re: Logistic regression and center effect

You can read about conditional logistic regression in SAS Online DOC
under proc logistic/Details.

As for the problem, aren't ANOVA like analysis suitable to determine
what has the effect on outcome, controlling for something else?

The solution with dummy coding centers is perfect, although it won't
be enough to check if the influence of characteristics is the same in
every center. But probably there is no theoretical background to worry
about it.

On Nov 7, 10:39 pm, Ray Koopman <koop...@xxxxxx> wrote:
On Nov 7, 11:20 am, kchambe...@xxxxxxxxx wrote:

Dummy-code the centers: Choose one center arbitrarily as a baseline
center or reference center. Then generate eight dichotomous indicator
variables, one for each of the other centers. Patients from the
reference center will get eight 0's; patients from other centers will
get seven 0's and a single 1. Use the indicators as predictors in the
regression. (Note: the fact that you're doing logistic regression
plays no role in all this.)

Where are the a-priori means comparisons one just uses such codes?
Why not use orthogonal contrasts?

Fine, if you have contrasts that you're particularly interested
in and you know what you're doing. Otherwise, simple dummy coding
is both easy to implement correctly and easy to understand -- the
coefficient of each dummy variable represents a comparison of the
correponding group with the reference group.


.

Relevant Pages

  • Logistic regression and rare events
    ... G King and L Zeng, "Logistic regression in rare events data," ... In this paper it is stated (without reference) that: ... Other references dealing with the influence on rare events in logistic ...
    (sci.stat.consult)
  • Re: Logistic regression and center effect
    ... center or reference center. ... Then generate eight dichotomous indicator ... reference center will get eight 0's; patients from other centers will ... the fact that you're doing logistic regression ...
    (sci.stat.consult)
  • Re: Logistic regression reference
    ... > Does anyone have a good reference for logistic regression. ... > the course with Dr. Lemeshow. ... > references seem to focus on predictive modeling. ...
    (sci.stat.edu)