[R-meta] Results interpretation publication bias
||@t@ @end|ng |rom dewey@myzen@co@uk
Mon Apr 3 18:16:21 CEST 2023
Please keep the list copied in as other may be able to help better than me.
Those funnel plots look very odd. I would normally have expected to see
the pseudo confidence bands plotted about the estimated value, not zero.
Apart from a few studies of low precision everything else is in an
almost rectangular block which does not suggest to me evidence of small
I am attaching your plot in case someone else has ideas.
On 03/04/2023 15:47, Wilma Charlott Theilig wrote:
> Dear Michael,
> those were some really helpful tips, thanks!
> I have attached the funnel plot for the trim and fill method with and
> without outlier as a PDF.
> Greetings Wilma
> *Von:* Michael Dewey <lists using dewey.myzen.co.uk>
> *Gesendet:* Montag, 3. April 2023 14:55:06
> *An:* R Special Interest Group for Meta-Analysis
> *Cc:* Theilig, Wilma Charlott
> *Betreff:* Re: [R-meta] Results interpretation publication bias
> Dear Wilma
> Comments in-line
> On 03/04/2023 10:14, Wilma Charlott Theilig via R-sig-meta-analysis wrote:
>> Hey all,
>> I am currently conducting my first meta-analysis using a random effects model. I wanted to investigate whether there is publication bias in my studies using different small-study effects methods. Egger's test and rank correlation test indicate funnel plot asymmetry. I then used the trim and fill method twice.
> If you are interested in publication bias, as opposed to small-study
> effects more generally, then
> suggests that a selection model would be preferable to trim and fill.
>> Once with all studies and since I had a very high I2 value (98.4%) and then again but with excluded outliers (here I used the find.outlier function from dmetar to identify 39 of 79 studies as outliers) as a sensitivity analysis.
> A method which identifies the majority of your data as outliers is not
> detecting outliers in the normal meaning of the word. Is it telling you
> you have a mixture distribution of effect sizes? When you look at them
> do you see any common feature about the primary studies? That is where I
> would look first to try to identify possible moderators.
>> Both values are higher (r= .57 and r= .38 without outliers) than than the original overall pooled effect size (r= .36). So my interpretation would be: there seems to be "missing" mainly large effect sizes with small standard errors (i.e. large studies reporting large effect sizes, at least according to the contour enhanced funnel
> Although this list deprecates attachments actually seeing the funnel
> plot might help here. I think the list might accept pdf attachments.
> But maybe it is also due to high heterogeneity? How can I interpret
> this result? I would greatly appreciate an answer a
>> nd some help.
>> Best regards
>> [[alternative HTML version deleted]]
>> R-sig-meta-analysis mailing list @ R-sig-meta-analysis using r-project.org
>> To manage your subscription to this mailing list, go to:
> <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient> Virus-free.www.avg.com <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient>
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 673071 bytes
Desc: not available
More information about the R-sig-meta-analysis