[R-meta] Publication Bias
Monfil Herrera, Laura
lmonfil at iconcologia.net
Fri Nov 10 11:22:48 CET 2017
Thank you so much for your quick and accurate answer.
De: Viechtbauer Wolfgang (SP) [mailto:wolfgang.viechtbauer at maastrichtuniversity.nl]
Enviado el: Friday, November 10, 2017 11:17 AM
Para: Monfil Herrera, Laura; r-sig-meta-analysis at r-project.org
Asunto: RE: Publication Bias
That is indeed quite a difference in the p-values of the two models.
Just a small note: The null hypothesis is that the plot is *symmetric* (or more precisely, that there is no relationship between the outcomes and the predictor, which by default is the standard error of the outcomes).
It is impossible to say which model is the more appropriate one in this particular case. However, with large heterogeneity, I think the simulation studies indicate that the 'rma' model approach has better control of the Type I error rate. Gerta Rücker and Guido Schwarzer have done research on these issues, so I hope they could chime in here with their insights. In fact, this article is quite relevant:
Rücker, G., Schwarzer, G., & Carpenter, J. (2008). Arcsine test for publication bias in meta-analyses with binary outcomes. Statistics in Medicine, 27(5), 746-763.
It focuses on arcsine square root transformed risk differences instead of double arcsine transformed proportions, but I think the conclusions would be very similar. To quote:
"We found that the AS-Thompson test performed better than most tests, maintaining the size comparable to that of the test by Peters et al. (unless sample sizes are small and there is no heterogeneity, where it is conservative), while having greater power than most tests including the Peters test (especially when both effect size and heterogeneity are relatively large)."
Note that "AS-Thompson" refers to using the 'rma' model.
>From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces at r-
>project.org] On Behalf Of Monfil Herrera, Laura
>Sent: Friday, 10 November, 2017 10:07
>To: r-sig-meta-analysis at r-project.org
>Subject: [R-meta] Publication Bias
>I have a doubt regardin publication bias and the metaphor function
>I want to check the publication bias from a random effect meta-analysis
>(proportions with double arcsine transformation and big amount of
>heterogeneity I^2: 95.77%) and perform a symmetry test. My concern is
>about the model that I have to use in the regtest function "lm" or
>"rma". I could see in specification notes that "lm" corresponds to the
>classical Egger test and "rma" to their extension. I also have checked
>"Sterne, J. A. C., & Egger, M. (2005). Regression methods to detect
>publication and other bias in meta-analysis" but I can't decide which
>is the appropriate model to use.
>Running both options, the final conclusion is couldn't reject the null
>hypothesis for funnel plot asymmetry but both p_values are very
>differents: lm: 0.822 and rma: 0.064.
>Could you give some advice which could be the most appropriate model?
>Thanks in advance.
More information about the R-sig-meta-analysis