2
$\begingroup$

I am performing a meta-analysis of proportions, I have data about the number of cases and the total sample sizes.

I will be using the Freeman-Tukey double arcsine (Freeman & Tukey, 1950; Miller, 1978) to adjust the proportions so that they are between 0.2 and 0.8.

But when I plot data with confidence interval, the confidence intervals are so small and the I square is 100%? what is the problem?

Note: am using R

$\endgroup$
3
$\begingroup$

What you are seeing is entirely consistent. If you have very narrow confidence intervals you will see very high values for $I^2$. In a paper entitles "Undue reliance on $I^2$ in assessing heterogeneity may mislead" available here Rücker and colleagues explain why this is. Basically it is because you are comparing within study and between study variability. If the within is very small then, all else being equal, $I^2$ is bound to be large. Their article is open access by the way.

It is fine to continue with the analysis but you would want to report on the variability between studies which makes the idea of a single summary value slightly problematic.

$\endgroup$
  • $\begingroup$ and for the paper its very useful, so does that mean that I cant pool those studies in a meta-analysis? $\endgroup$ – basma Aug 21 '18 at 20:13
  • $\begingroup$ Not necessarily. It means that you need to think about the studies being pooled to assess whether they are appropriate to combine. Are things like study design, population of interest, treatment and control interventions, outcomes, follow up times, blinding, etc are all similar enough. There is no metric that will tell you if this is appropriate, only subject matter expertise. $\endgroup$ – prince_of_pears Aug 22 '18 at 11:22

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.