1
$\begingroup$

I'm interested in conducting a meta-analysis of mental health in the first year after an event. I have a set of prospective studies that assess mental health at baseline (i.e., immediately after the event) and then at a variety of follow-up time points among the people that experienced the event (e.g., one month, six months). Some studies also include a comparison group, although I'm mostly interested in changes within the group that experienced the event. Ultimately, I would like to be able to speak to trajectories of change in various disorders after this event- for example, do rates of depression decrease significantly by 3 months and level out? Does anxiety decrease immediately?

My dilemma is, some studies assess mental health as a binary variable (i.e., number of people who have the disorder or do not have the disorder), while others assess it as a continuous variable (i.e., mean scale score/SD). I'm familiar with the single-group, pretest–posttest change/raw score effect sizes to look at changes over time for mean data (Morris & DeShon, 2002) but I'm not sure how to handle this for the dichotomous data. Ideally, I'd like to be able to find a common effect size statistic so I can analyze both types of data together. Is this possible? If so, what effect size can I use? If I have to analyze the binary data separately, can I just use an odds ratio?

$\endgroup$
1
$\begingroup$

If some studies give you an odds ratio and some give you a standardised effect size then it certainly is possible, given some assumptions, to convert between them, Since you work in the health field a useful reference might be Sue Chinn's article in Stats in Medicine from 2000 available here. If you find it is behind a paywall I think there are copies elsewhere on the web. The same method has been discovered independently by other people in other domains.

$\endgroup$
  • $\begingroup$ My question is not whether I can combine two types of effect sizes- I have done so in past meta-analyses and know that that is possible in theory. However, since this an analysis of longitudinal data, and the effect size to represent mean change over time within one group appears to be different than the effect sizes that represent mean differences between two independent groups (unless I'm wrong, but it's outlined in the article I cited), I wasn't sure whether the same is true for binary data and whether such ESs can be combined. Apologies if my question was unclear. $\endgroup$ – Emily Sep 25 '16 at 14:07

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.