1
$\begingroup$

I am performing a meta-analysis where some data sets I would like to include are reported as not being normally distributed. My understanding is that non-normally distributed data should not be processed in continuous approaches to data analysis. As such, should studies that are non-normally distributed be excluded from the meta-analysis?

In an attempt to use only normally distributed data, where a data set has been described as non-normally distributed, I have been trying to gain access to the raw data making up that data set which I can then log-transform, and if the data is then normally distributed, including this data set in the meta-analysis. This leads me to another question: is combining raw data (which we know is normally distributed) with log-transformed data (which has been normalised) a robust methodological approach in meta-analysis?

Thank you!

$\endgroup$
1
$\begingroup$

So that this does not appear unanswered here is my original comment amplified to an answer:

The underlying data in the primary studies and its distribution is not relevant to the meta-analysis. What is at issue is the quantity which is extracted from the primary studies and its distribution. To see this at its most extreme consider primary studies which use binary outcomes. Nobody would suggest that a binary variable is Gaussian but the quantity extracted might be the log odds ratio and it may well be reasonable to assume that is Gaussian.

$\endgroup$

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.