2
$\begingroup$

The treatment groups I would like to combine appear to be virtually the same, at least by generic label for each group, e.g., groups with the same treatment but implemented at different sites or stratified into subgroups that can be recombined into a sensible whole. Lipsey and Wilson (2001) recommend that if this is the case, you can combine the treatment groups into a composite whole if appropriate statistics are available. Which statistics are they referring to? How can I go about appropriately combining these groups?

$\endgroup$
  • $\begingroup$ Can you be a little more specific about the application and perhaps provide a more elaborate example otherwise you risk an answer that states: "It depends on your situation" $\endgroup$ – ctbrown Nov 14 '14 at 1:18
  • $\begingroup$ Trials with multiple sites are often combined using the Mantel Haenszel method (sphweb.bumc.bu.edu/otlt/MPH-Modules/BS/…) rather than just simply adding the sums together. $\endgroup$ – abousetta Nov 16 '14 at 0:06
2
$\begingroup$

I am trying to answer despite this being an old question.

The key premise is whether the study you are talking about is a randomized trial, and these are separate randomization arms. If this is the case, then you can safely pool the groups (remember though that this must be clinically sound).

If the study is not randomized, or the groups are not generated through randomization, then it depends on the comparison/effect estimate of interest. You can use stratification (eg the CMH approach recommended by abousetta), or meta-regression. However, inferences will be weak and the main scope of your work will be hypothesis generating.

$\endgroup$

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.