[R-meta] Using MAd agg function

[A C Del Re]

Hi Emily,

Technically, yes, you have accounted for dependent effect sizes (ES) and
could use this aggregate ES in further analyses. However, I think it
depends on the actual outcomes and if it makes sense to aggregate. For
example, 3 measures of psychological symptomology would probably make sense
to combine (well depending on your research questions) but combining, say
measures of intelligence, psychological traits, and physical endurance,
might not.

Regarding the estimated correlation, there are a few possibilities to
consider: (1) check previous literature in your substantive area about the
correlation among the outcomes and use that value as the estimate; (2) if
there are enough studies providing each of these outcomes, check the
correlation between them in your data and use that; (3) conduct sensitivity
analyses at a few different correlation values and see if there are any
substantial differences.

Bill Hoyt and I just published an article examining this issue (and other
issues with effect sizes) within psychotherapy studies. You can get it here:

www.acdelre.com/pubs

Hope this helps,

AC

-- 
AC Del Re, PhD
acdelre.com


On Sun, Dec 17, 2017 at 3:20 AM, Emily Russell <emilyrussell99 at outlook.com>
wrote:

> Dear All
>
>
> I wondered if anyone could help with an issue about aggregating effect
> sizes?  I have 3 effect sizes from one study that come from one sample and
> so I want to aggregate them into one.  The result of escalc is this
>
>
> Study       yi           vi
> Arnold  -0.2386 0.2014
> Arnold   0.4556 0.2052
> Arnold   0.0563 0.2001
>
>
> so then I used the agg function from MAd
>
>
> ag<-agg(id=Study, es=yi, var=vi,  cor=.5, method="BHHR", mod=NULL, data=ma)
>
> which gives the following which I think I can then use as one study in my
> meta-analysis
>
>       id         es              var
> 1 Arnold 0.09111254 0.1348174
>
> Does this seem reasonable?  Also as I don't know the correlation between
> the three effect sizes I have just used the default of 0.5.  I suspect this
> is low, but that is just a guess - is that reasonable?
>
> Thanks
>
> Emily
>
>
>
>         [[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
>

	[[alternative HTML version deleted]]