[R-meta] Direction of the effect sizes

[Michael Dewey]

Dear Miguel

I think the answer is going to depend on exactly what analysis you are 
doing. If you analyse the studies in several separate m-a then I do not 
see any problem as the differences come out in the interpretation. If 
you are going to do a multi-level m-a with both outcomes and both 
interventions included then the problem will be taken care of by the 
effect of the moderators and possibly their interaction.

I can see that things may in some cases be made easier by reflecting the 
effect sizes (multiplying by -1) for some groups of studies but I find 
it hard to think of a problem which would be solved by using absolute 
values. I suppose if you were trying to detect a form of bias which 
meant that too few studies near the null were available it might be 
necessary.

Michael

On 05/12/2017 10:32, Miguel García Kml wrote:
> Hello everyone!
> 
> A doubt has arisen that is perhaps quite basic/absurd, but which takes me
> into my head for a while. I will try to explain my doubt in the simplest
> way.
> 
> Imagine that we have two styles of music (let's say classical and rock) and
> different psychological measures (scores in a memory task and levels of
> anxiety). Suppose that classical music is tremendously relaxing and
> therefore reduces anxiety levels, but it will worsen the results in memory
> tasks. On the other hand, rock music will increase anxiety levels, but will
> improve memorization. In this way, studies would display positive and
> negative effect sizes in memory scores and anxiety, which indicates
> effectiveness, no matter what the direction is (strange example, but it is
> easier to understand that way).
> 
> Two doubts arise regarding this:
> 
> 1) Should I use the absolute values of the effect size? since, although
> some ES are negative, it indicates effectiveness as a whole (my
> meta-analysis is about effectiveness). For instance, classical music
> exposure might display relaxing effects (+ES) that are incompatible with
> memorization (-ES), but as a whole, it shows effectiveness regardless of
> the direction of the effect size.
> 
> 2) If I should use the absolute values, should I use them in all the
> analyzes? (i.e., meta-regression, funnel plot, influence diagnosis, etc.)
> Or perhaps just in the regression analysis to be able to explain the
> heterogeneity in terms of effectiveness? I am confused.
> 
> I hope that I have explained my question correctly, and I apologize if it's
> too obvious. Thank you very much!
> 
> Best,
> 
> Miguel
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-meta-analysis mailing list
> R-sig-meta-analysis at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis
> 

-- 
Michael
http://www.dewey.myzen.co.uk/home.html