I am new to meta-analysis. I am looking at the psychological symptoms that develop in healthy people when given a drug versus placebo. Different measurement scales are used -so I need to use SMD. There is a mixture of change data and absolute measurement data. Mostly absolute measurement data. I want do a metaregression for participant age etc I also have the problem that there is paired data and unpaired data- most of which is paired. What is the best way of calculating the effect size combining all this data. Or would it be better to not include change data and independent group data. The other thing to say is that in some of the studies have 0 as the s.d for the placebo group because well people dont have symptoms with placebo - so this data looks like change data.

  • $\begingroup$ I think mvmeta in R or Stata can do the trick. $\endgroup$ – Joe_74 Jan 29 '19 at 9:10
  • $\begingroup$ Honest answer is you need to work closely with a statistician. You shouldn't combine change scores with final values using SMD (see Cochrane Handbook). You also shouldn't combine paired and independent values. Each carry their own assumptions. $\endgroup$ – abousetta Jan 30 '19 at 17:26
  • $\begingroup$ Also, patients with placebo do have symptoms (or else you're assuming there is no placebo effect) even if not directly linked to placebo. So an SD of 0 is implausible. If that's what's reported then you will have to add a fudge factor to allow the calculations to run. So, back to your question... talk to a statistician because if your assumptions are not correct then neither will the results. $\endgroup$ – abousetta Jan 30 '19 at 17:27

Your Answer

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

Browse other questions tagged or ask your own question.