I am doing a case-control study with 80 disease cases matched 1:3 to non-diseased controls and examining whether they had a binary exposure prior to developing the disease. I am using multivariable conditional logistic regression for the main analysis. My question is, for the "descriptive" or "univariate" table where we try to investigate simple associations of each (non-matched) variable with the outcome, do I just use something like a chi-square test and ignore the matching, or do I use main effects conditional logistic regression (or something else)? Any help would be greatly appreciated.


I would not use the word "descriptive" statistics here (I would limit "descriptive" and "univariate" to summary or detailed descriptions of single-variables—e.g. histograms, mean and standard deviation, median and IQR, etc.). I would instead use language like "bivariate tests for association."

To test whether prevalence of exposure in cases differed from prevalence of exposure in (3x) controls, I might recommend reading Miettinen's generalization of McNemar's test (McNemar's test is a $\chi^{2}$ test that accounts for the individual matching of the case data with the control data):

Miettinen, O. S. (1969). Individual matching with multiple controls in the case of all-or-nothing responses. Biometrics, 25(2):339–355.

You might also consider Cochran's Q test (which is a nonparametric analog of repeated measures/blocked design ANOVA for nominal data).

Cochran, W. G. (1950). The comparison of percentages. Biometrika, 37(3/4):256–266.

  • $\begingroup$ Thanks, both of those a good suggestions (and you are right of course about the bivariate/univariate thing). $\endgroup$ – user2414716 May 14 '14 at 15:00
  • $\begingroup$ Feel free to click the check mark if you feel this is an acceptable answer. $\endgroup$ – Alexis May 14 '14 at 19:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.