[R-meta] Obtaining study-level effect size and sampling variance through robust variance models
Viechtbauer, Wolfgang (SP)
wo||g@ng@v|echtb@uer @end|ng |rom m@@@tr|chtun|ver@|ty@n|
Mon Mar 11 10:00:55 CET 2019
I just simulated some data to illustrate the plotting. There was no particular reason I used a uniform for simulating the sampling variances. As for the outcomes, the standard models assume that the sampling distributions are normal, so that is why I simulated 'yi' from a normal distribution.
From: Mutlu Umaroglu [mailto:mutlu.umaroglu using hacettepe.edu.tr]
Sent: Monday, 11 March, 2019 8:28
To: Viechtbauer, Wolfgang (SP); r-sig-meta-analysis using r-project.org
Subject: Re: [R-meta] Obtaining study-level effect size and sampling variance through robust variance models
Dear Viechtbauer and all,
In this scenarios, vi follows uniform distribution and yi follows normal
distribution. vi~U(.01,1) and yi~N(0,vi+.5=T^2).
Why vi follows uniform distribution? Is uniform distribution more robust
Other words, what is the recommended distribution (and value) of Yi and
Vi in a typical simulation scenario for meta analysis? And Why?
10-03-2019 14:01, Viechtbauer, Wolfgang (SP) yazmış:
> Plus addpoly(), to add a summary polygon to the forest plot that shows
> the results from robu(). For example:
> dat <- data.frame(vi = runif(20, .01, 1))
> dat$yi <- rnorm(20, 0, sqrt(dat$vi + .5))
> dat$study <- sort(sample(1:10, 20, replace=TRUE))
> res <- robu(yi ~ 1, var.eff.size = vi, studynum = study, data=dat)
> forest(dat$yi, dat$vi, ylim=c(-1.5,res$M+3), slab=paste("Study",
> dat$study), cex=1)
> addpoly(res$reg_table$b.r, ci.lb=res$reg_table$CI.L,
> ci.ub=res$reg_table$CI.U, cex=1)
> -----Original Message-----
> From: R-sig-meta-analysis
> [mailto:r-sig-meta-analysis-bounces using r-project.org] On Behalf Of
> Michael Dewey
> Sent: Saturday, 02 March, 2019 14:18
> To: Mufan Luo; r-sig-meta-analysis using r-project.org
> Subject: Re: [R-meta] Obtaining study-level effect size and sampling
> variance through robust variance models
> Dear Mufan
> You do not need to fit a model with rma.uni to use forest.
> On 01/03/2019 18:16, Mufan Luo wrote:
>> Dear meta-analysists,
>> Hope this email finds you well.
>> I am conducting a meta-analysis using robust variance model. To create
>> forest plot for each study, I’d like to obtain mean effect size and
>> sampling variance for each study.
>> I decided to use forest function in metafor to create the forest
>> Since the forest function only accepts rma file, I am trying to fit a
>> rma model (rather than rma.uni) that produces the same coefficient,
>> 95% CI and p-value as the robu model.
>> For example, below is my robu model,
>> run.anxiety <- robu(formula = Fisher.s.Z ~ 1,
>> var.eff.size = Fisher_var,
>> data = anxiety,
>> studynum = Study,
>> modelweights = "CORR")
>> According to prior discussion about converting robu to rma.uni in this
>> mail list, I also calculated the number of studies k in cluster j,
>> average of sampling variance Vbar, and tau square.
>> tau_sq_robu_anx <- as.numeric(run.anxiety$mod_info$tau.sq)
>> anxiety$k <- with(anxiety, table(Study)[Study])
>> anxiety$Vbar <- with(anxiety, tapply(Fisher_var, Study, mean)[Study])
>> I am trying to get the weight and plug it into the following model,
>> rma(yi = weightedES, vi = ??, data = weighted)
>> However, I am not sure if the correct calculation is
>> anxiety$Vnew <- with(anxiety, as.numeric(Vbar + tau_sq_robu_anx)
>> Thank you so much for our attention.
More information about the R-sig-meta-analysis