I have developed a case-control study, matched 1:7. I addressed the conditional logistic regression analysis with COXREG (SPSS).
By using univariate analysis (McNamer/Wilcoxon), I found that there is no statistically difference between case and control groups (checked for one of my dichotomous independent variable).
When I used COXREG (in order to create consitional log regression in SPSS) and analyse the same independent variable (analyzed alone in regression) - I realized there was differenet between case and control groups.
To sum up - I found a different statistically significant while using univariate and COXREG (conditional log regression) analysis (for the same variable).
What could be my mistake?
People who do not use SPSS may be wondering why Daniel Gabbai is talking about the COXREG command when he is estimating a conditional logit model. The reason is that SPSS still does not have a conditional logistic regression command (believe it or not). For years, the recommendation has been to cajole the COXREG command into estimating the model. Details can be found in these notes on the IBM website:
- https://www.ibm.com/support/pages/conditional-logistic-regression-using-coxreg
- https://www.ibm.com/support/pages/11-matched-case-control-studies-conditional-logistic-regression
Daniel, bear in mind that there are other ways to account for the correlated nature of the data within match groups. You could use GENLIN with generalized estimating equations (GEE), for example; or you could use GENLINMIXED with patients clustered within match groups. Something to consider if you do not have to use conditional logistic regression. And FWIW, when I used GENLIN with GEE to reanalyze some data that I looked at with conditional logistic regression (using Stata's clogit command), I found that the results were very similar using the two approaches.
HTH.
Bruce Weaver One more reason to move to R. Thank you Bruce. David
BTW here's the details on how R does that calculation. Thanks again Bruce.
David
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression