Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: Interpreting conditional logistic regression equations using 2 similar types of matching.


From   "Donald Spady" <dspady@ualberta.ca>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: Interpreting conditional logistic regression equations using 2 similar types of matching.
Date   Tue, 6 Feb 2007 16:40:09 -0700

Dear all
This may be a second posting.  The first one did not seem to go through and
on reading the FAQ I realized that I might have sent non-plain text. (I
copied the output from Stata using the copy table command rather than the
copy text command.  Hopefully this one will work.)

I am looking at the effect of iron-deficiency anemia on the likelihood of a
child having a seizure. I am using clogit but have a problem in that I have
matched my case/control "group" variable (age) as either very tightly
matched (same month:  "tight") or more loosely matched (up to 3 months
difference: "loose").  The whole age range of difference is 30 months, so 3
months is still fairly closely matched.  When I do the 2 clogit equations I
get different Odds ratios (which seems OK), but don't know which one is the
'right' one to use.  Any help would be appreciated.

If I use the same cases I get different odds ratios.
clogit patient Anem, group(tight) or
note: 40 groups (40 obs) dropped due to all positive or
      all negative outcomes.

Iteration 0:   log likelihood = -197.63842  
Iteration 1:   log likelihood = -197.44509  
Iteration 2:   log likelihood = -197.44481  
Iteration 3:   log likelihood = -197.44481  

Conditional (fixed-effects) logistic regression   Number of obs   =
574
                                                  LR chi2(1)      =
2.98
                                                  Prob > chi2     =
0.0845
Log likelihood = -197.44481                       Pseudo R2       =
0.0075

----------------------------------------------------------------------------
--
     patient | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf.
Interval]
-------------+--------------------------------------------------------------
--
        Anem |   2.142857   .9808665     1.67   0.096     .8737077
5.255575
----------------------------------------------------------------------------
--
r; t=0.22 16:31:34

. clogit patient Anem, group(loose) or
note: 112 groups (112 obs) dropped due to all positive or
      all negative outcomes.

Iteration 0:   log likelihood = -173.23535  
Iteration 1:   log likelihood = -173.23185  
Iteration 2:   log likelihood = -173.23185  

Conditional (fixed-effects) logistic regression   Number of obs   =
502
                                                  LR chi2(1)      =
1.50
                                                  Prob > chi2     =
0.2213
Log likelihood = -173.23185                       Pseudo R2       =
0.0043

----------------------------------------------------------------------------
--
     patient | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf.
Interval]
-------------+--------------------------------------------------------------
--
        Anem |   1.538462   .5480964     1.21   0.227     .7653064
3.092701
----------------------------------------------------------------------------
--
r; t=0.16 16:31:43

In both cases the cases used are the same; the cases and controls are paired
on age but using slightly different criteria.  One equation drops 40 groups
and the other drops 112 groups.  The total number of groups is the same for
both equations.
My questions are: 
1. Is it reasonable to see such a 'big' difference in OR between the two
equations?  
2. Which is more likely to give the most reliable OR; the equation where the
matching was rigid, or the equation where the matching is more relaxed?

Many thanks

Donald W Spady MD, MSc
Departments of Pediatrics and Public Health Sciences
8226B, Aberhart Centre 1
University of Alberta
Edmonton, Alberta,
CANADA
T6G 2J3

780-407-1244: Office
780-407-7136: FAX

Nature has no reset button!


*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index