[R-meta] check meta-analysis done correctly

Viechtbauer Wolfgang (SP) wolfgang.viechtbauer at maastrichtuniversity.nl
Thu Nov 16 22:17:56 CET 2017

Dear Nadine,

1) You wrote nlu (with an 'el'), but it should be n1i (with a 'one').

2) I think you switched around what is pci and pti. pci is for the control group (the one whose information is being reused).

3) Even with this fixed, you run into division by zero in computing V because pci is zero in the multiple treatment groups. Just setting those NA values to 0 is not a proper way of dealing with that. You might consider using a +1/2 adjustment in computing pti and pci (which is also applied in escalc() when the log ORs are being computed). Something like:

dat$pci <-with(dat, (lameness_cases_ctrl + 1/2)/(total_ctrl + 1))
dat$pti <-with(dat, (lameness_cases_vax + 1/2)/(total_vax + 1))
dat$n1i <-with(dat, total_ctrl + 1)

4) Your model is still not correct. It should be:

dat$id <- 1:nrow(dat)
res <-rma.mv(yi, V, random= list(~1|Publication_cluster, ~1|id), data = dat)

5) BUT: 

Most of these studies are rather small in the first place. It is questionable whether using a 'normal-normal model' (assuming normal sampling distributions and normal random effects) is sensible here. I personally think this calls for using a logistic mixed-effects model. See, for example:

Simmonds, M. C., & Higgins, J. P. T. (2016). A general framework for the use of logistic regression models in meta-analysis. Statistical Methods in Medical Research, 25(6), 2858-2877.


>-----Original Message-----
>From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces at r-
>project.org] On Behalf Of Nadine Vogt
>Sent: Thursday, 16 November, 2017 20:30
>To: r-sig-meta-analysis at r-project.org
>Subject: [R-meta] check meta-analysis done correctly
>ATTACHMENT(S) REMOVED: Anorexia MV analysis.R | Metaanalysis in R.R |
>Hi Wolfgang,
>Many thanks for your previous email specifying that I needed to include
>two random effects instead of just one, and to calculate a covariance
>matrix due to correlation between OR estimates due to the use of the same
>control group for multiple treatment groups.
>I would be grateful if you could check my current models. I have attached
>code for two different models---the lameness one included studies with
>multiple treatment arms, and the anorexia one did not.
>Thanks in advance,
>Nadine Vogt

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