Meta analysis for correlation where different direction has different meaning

[Therain]

Hi There,

I am doing a meta-analysis for correlation between sleep and mental health. The issue is in some studies higher score indicates better sleep, while in other of my studies lower score means better sleep. So the direction of r could mean different things. I am wondering how should I handle this? Can I just pick absolute value? For example, if r=-0.2, then I change it to 0.2. Or if r=0.45, then I change it to -0.45?

Thanks in advance!

[ballardw]

You would have to look at the definitions of how the original scores were assigned to see use of absolute value would be appropriate.

Then you may get to spend some time with algebra and equation fitting all of the scoring systems into equivalent measure. Consider an example of standardizing fuel efficiency between 3 studies, one scores on miles per gallon, one uses kilometers per liter (these two at least sound similar) and a third that uses passenger miles per ton. That last one is going to involve some digging and might find that the first two scales involved gasoline (or diesel or similar fuel) and the last involved coal. In some cases such as this third you might find that scoring systems derived from collecting different information to create the score really are not amenable to any direct comparison.

[Therain]

Got it. Thank you for the suggestion!