2
$\begingroup$

I recently came across a paper that uses the McNemar test to evaluate the effectiveness of an intervention to improve adherence to treatment. The study used a pre/post test design whereby the adherence behaviour for both the control and intervention group was measured using the same instrument and expressed as continuous data. It then compares the relative improvement (post / pre) between the two groups. I've never seen the McNemar test used for this before, but am not sufficiently familiar with statistics. Is it really an appropriate method? Wouldn't a (paired) t-test or something like that make more sense?

$\endgroup$
  • $\begingroup$ How exactly did they use McNemar's test? $\endgroup$ – Scortchi May 2 '13 at 11:25
  • $\begingroup$ It doesn't actually explain much so am not sure. The data they show are adherence in both intervention and control, expressed as a percentage of the number of participants, at pre- and post-test. Then a relative improvement is given, which is calculated as the ratio of the percentage post vs. pre test. Does that help? $\endgroup$ – amsbridge May 2 '13 at 11:30
  • $\begingroup$ If they're showing adherence as a percentage, doesn't that imply it's Yes/No - discrete rather than continuous as you stated? $\endgroup$ – Scortchi May 2 '13 at 11:49
  • $\begingroup$ They show both in fact, but you're right that the percentages -which is what they use in their comparison- are of course discrete. Apologies for the confusion. $\endgroup$ – amsbridge May 2 '13 at 11:54
  • $\begingroup$ So they used a cut-off to turn a continuous variable into a binary one before carrying out McNemar's test on the counts of values of the binary variable? $\endgroup$ – Scortchi May 2 '13 at 12:03
2
$\begingroup$

Within either the treatment or the control group the McNemar test seems perfectly appropriate to test marginal homogeneity of the pre/post vs adherent/non-adherent contingency table - it's a matched-pair design for dichotomous variables. I can't see how you'd use it to compare treatments vs controls, as they're unmatched, presumably. It's not clear from your description what tests were actually carried out in this study.

$\endgroup$
  • $\begingroup$ ah okay, that clarifies it already a lot. The study also does not elaborate any further on the tests carried out, so I'm afraid we'll have to keep guessing on that one. :) Thanks! $\endgroup$ – amsbridge May 2 '13 at 13:22

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.