[R-meta] Different results for tests of funnel plot asymmetry using R package meta and metafor

Viechtbauer Wolfgang (SP) wolfgang.viechtbauer at maastrichtuniversity.nl
Thu Nov 30 12:21:44 CET 2017


Yeah, I'll add a note about that to the documentation.

As for consistency: For the classic Egger test, the distribution is t with df=k-2 under the null (under the assumptions of the multiplicative model). But for the additive model, there is no longer any exact statistical theory. Even the K&H method is just an approximation (albeit a very good one). Also, since the Thompson-Sharp approach is just meta-regression, metafor is being consistent in the sense that it uses the same inference method that was used in fitting the model to begin with.

But in the end, I would say that this is all pretty irrelevant for practice, as testing for funnel plot asymmetry / small study effects is hardly an exact science and only useful when k is not too small, at which point the difference between t and z becomes increasingly negligible.

Best,
Wolfgang

-----Original Message-----
From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces at r-project.org] On Behalf Of Guido Schwarzer
Sent: Thursday, 30 November, 2017 11:33
To: r-sig-meta-analysis at r-project.org
Subject: Re: [R-meta] Different results for tests of funnel plot asymmetry using R package meta and metafor

Hi Wolfgang,

Thank you for the explanations and the R code. I did not know that using 
argument /test/ in rma.uni() has any influence on the test for 
funnelplot asymmetry. AFAICS, the help pages of rma.uni() and regtest() 
do not describe this feature.

Concerning the use of a standard normal or t-distribution, I think the 
behavior of metabias() is more consistent as the classic Egger test 
(/method = "linreg"/) and the Thompson-Sharp test (/method = "mm"/) only 
differ in their weights: 1 / seTE^2 versus 1 / (seTE^2 + tau2).

This said, I think it would be interesting to evaluate the properties of 
the Knapp-Hartung variant of the regression test for funnel plot asymmetry.

Best wishes,
Guido



More information about the R-sig-meta-analysis mailing list