Conditional regression or other methods?

[Staffan Lindberg]

Dear list!

 

I am reviewing a very old material of 420 female alcoholic patients. Each patient has up to 2-5 female controls (mostly 5) matched on age at admission, place of residence (Stockholm county) at admission, level of education and socioeconomic group. The common variables of interest are a mixture of nominal, ordinal and scale variables (mostly health/ sickness related spanning over an observation period of several decades).  I would like to test the differences between the patients and controls in these variables taking into account the more stable estimates of the controls. Will this type of testing be different for nominal, ordinal and scale variables?

 

I remember asking this to the list several years ago and , if I remember correctly, getting an answer from Martha Garcia-Granero suggesting “conditional regression” Since then I was swamped in other work and the material receded into oblivion. Now  I would like to tackle this again. I have never heard of “conditional regression”. Can anyone explain? Is this a relevant method or are there better methods? Any hints will be much appreciated.

 

best

 

Staffan Lindberg

Sweden

[Bruce Weaver]

I suspect Marta suggested conditional logistic regression.  Alternatively, you can model repeated (or correlated) measures in logistic regression via GENLIN using generalized estimating equations (GEE).  I asked for help with that a few years ago, and Rick Oliver (IIRC) gave some useful advice on how to get the syntax right.  That should give you the search terms you need to find the old thread in the archives.  (I'd search myself, but am on holidays, and running on battery power right now.)  Good luck.

Staffan Lindberg wrote

Dear list!



I am reviewing a very old material of 420 female alcoholic patients. Each
patient has up to 2-5 female controls (mostly 5) matched on age at
admission, place of residence (Stockholm county) at admission, level of
education and socioeconomic group. The common variables of interest are a
mixture of nominal, ordinal and scale variables (mostly health/ sickness
related spanning over an observation period of several decades).  I would
like to test the differences between the patients and controls in these
variables taking into account the more stable estimates of the controls.
Will this type of testing be different for nominal, ordinal and scale
variables?



I remember asking this to the list several years ago and , if I remember
correctly, getting an answer from Martha Garcia-Granero suggesting
"conditional regression" Since then I was swamped in other work and the
material receded into oblivion. Now  I would like to tackle this again. I
have never heard of "conditional regression". Can anyone explain? Is this a
relevant method or are there better methods? Any hints will be much
appreciated.



best



Staffan Lindberg

Sweden