Tools for Those Who Summarize the Evidence Base
Hi, I am intersted to do a meta-analysis, have data set from about 50 experiments, mostly unpublished, some published in the form of reports, but non of them in any scientific journal, but each experiment was conducted using a RCBD (randomized complete block design). Data structure is quite complex.
can we do meta-analysis on such type of data?
can anyone suggest me, how can I assess the quality of this data, or what should be the statistical quality criteria for inclusion of each experiment into meta-analysis.
what to do the publication bias phenomenon here?
I'm sure you can do meta-analysis on these date, assuming their designs are highly similar. Though your mentioning that the data structure is "quite complex" is a concern and it might make a meta-analysis tougher.
You can certainly use routine tests for publication bias (e.g., Egger's test) even with most being unpublished. Though the fact that you have unpublished and published cases suggests that the best comparison will be published vs. unpublished. Of course, significant heterogeneity will mean that interpretations of these tests are more difficult.
There are lots of methodological quality scales out there (e.g., PEDro), and they are more and more routinely applied. Whether they actually gauge quality as you intend it to be gauged is the question you should have in mind as you decide whether to use such a thing. (Or as you modify them for your purposes.)
Hope that helps!