Suppose we have 5 studies of our choice with percentage in results. Can we perform meta analysis based on percentage, as mean, sd or other values are not reported in any of those studies.
5
-
$\begingroup$ Usually for a meta-analysis, you need an effect size, and a standard error (or variance) for each sample being pooled. Sometimes you get this from the study, sometimes the paper provides information that lets you calculate it. Usually in a meta-analysis you are combining results from studies that compare two groups. What information do you have? If you have two groups, and for each you have a percentage, this is straightforward if you also have a sample size. $\endgroup$– Jeremy MilesOct 16, 2015 at 5:05
-
$\begingroup$ Thanks @JeremyMiles. As I said, all these studies are not reporting everything. I am new to Meta Analysis, so was preparing all the important information. 2-3 studies have reported enough information to calculate effect size as you said, but others have not provided. For others, only percentage and mean is given. I was bit confused whether to work on these 5 studies (the topic of my choice and interest), as every study has not followed the same pattern and methodology that we usually see in meta analysis (If I am correct). Over that one is case-control and other 4 are before-after. $\endgroup$– Yogesh SharmaOct 17, 2015 at 15:56
-
$\begingroup$ Out of the 4 studies, 2 have matched groups and 2 have unmatched, means before-after on same group in 2 studies and before-after on 2 different unmatched groups. If this work is not feasible then I won't be able to proceed. I had asked some fellows about the methodology and feasibility, if the meta analysis can be performed in these condition, but have not received any clear view, so asked here. @JeremyMiles. $\endgroup$– Yogesh SharmaOct 17, 2015 at 16:02
-
$\begingroup$ For before and after mean, you can calculate the change (it's the difference in the means) but for a standard error you need to know something like the sd of the difference, or the correlation. Unmatched groups it's just the regular correlation. It's never super clear whether you can do a meta-analysis. In my experience I often end up saying that I've done the meta-analysis, but the studies were a bit crappy, or didn't provide the right data, so I don't have massive confidence in the results. $\endgroup$– Jeremy MilesOct 17, 2015 at 16:36
-
1$\begingroup$ Thanks @JeremyMiles for the suggestions. I would rather try to find the solution or drop the idea of this. $\endgroup$– Yogesh SharmaOct 18, 2015 at 13:33
Add a comment
|