[R-meta] Question regarding non-linear effects in meta-regression

[Michael Dewey]

Dear Jens

Comments in line

On 24/04/2018 12:38, Jens Schüler wrote:
> Hi all,
> 
> I am currently investigating the possibility of assessing non-linear 
> effects in a meta-regression.
> 
> For example, I am interested in the relationship between risk-taking and 
> performance, and I want to assess whether the mean score of risk taking 
> has a non-linear moderating effect on that relationship.
> 
> I had a look at Wolfgang’s previous examples on adding restricted cubic 
> splines to mixed effects models via the rms package e.g.
> 
> https://stats.stackexchange.com/questions/279668/meta-regression-using-restricted-cubic-splines-with-rma-from-the-metafor-pack?rq=1
> 
> https://stat.ethz.ch/pipermail/r-sig-meta-analysis/2018-February/000551.html 
> 
> 
> and looked a bit into Frank Harrell’s book concerning the adequate 
> number and location of nodes.
> 
> My question now is, how do I determine/test whether the plotted spline 
> is actually significant/meaningful?
> 

I would suggest that looking at the plot as suggested in the examples to 
which you link is the best way to visualise and explain to others what 
exactly is going on with your moderator.

If you want a formal test then comparing a model with and without the 
spline using anova() would seem the way to go.
> I am a bit unsure what I should make of the mixed effects model results 
> in that regard or if I have to perform additional tests.
> 
> Best
> 
> Jens
> 
> 
> 
> _______________________________________________
> 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