Meat analysis using metacor() form the meta package
557 views
Skip to first unread message
Ayala Allon
Apr 12, 2017, 3:11:10 AM4/12/17
to Israel R User Group
Hi everyone,
I'm trying to do a meta analysis (random effects estimates with correlations) using the metaocr() from the meta package. I already did it using the Comprehensive Meta Analysis Software, but the free trial there is over and I want to be able to do this in R (I need to add more studies now, but first I wanna make sure I can replicate the analysis I already did using R).
This is my data (also attached as a text file):
1. The Experiment column is the label for each study.
2. The Corr column is the effect size column.
3. The N represents the number of cases in each study.
> cor
Experiment Corr N
1 Experiment1 -0.3173759 16
2 Experiment2 0.2253440 16
3 Experiment3 -0.3209169 16
4 Experiment4 -0.7070982 16
5 Experiment5 -0.4100372 16
6 Experiment6 -0.5502341 16
I tried running metacor() with the following arguments. However, I get an error and I can't figure it out. I tried google it and everything but couldn't find anything:
> library(meta)
> testcor <- metacor(cor$Corr, cor$N, cor$Experiment,
+ data=NULL, subset=NULL,
+ sm=gs("ZCOR"),
+ level=gs(0.95), level.comb=gs(0.95),
+ comb.fixed=gs(FALSE), comb.random=gs(TRUE),
+ hakn=gs("hakn"),
+ method.tau=gs("method.tau"), tau.preset=NULL, TE.tau=NULL,
+ tau.common=gs("tau.common"),
+ prediction=gs(FALSE), level.predict=gs("level.predict"),
+ null.effect=0,
+ method.bias=gs("method.bias"),
+ backtransf=gs("backtransf"),
+ title=gs("title"), complab=gs("complab"), outclab="",
+ byvar, bylab, print.byvar=gs("print.byvar"),
+ byseparator = gs("byseparator"),
+ keepdata=gs("keepdata")
+ )
Error in chkchar(x) : Argument 'x' must be a character string.
I'm also attaching the results of the meta analysis using the Comprehensive Meta Analysis Software.
Does anyone know why I get this error? Any help would be greatly appreciated.
Thanks,
Ayala
Avraham N Kluger
Apr 13, 2017, 4:43:26 AM4/13/17
to Israel R User Group
Try this code:
r <- c(-0.3173759, 0.2253440, -0.3209169, -0.7070982, -0.4100372, -0.5502341)
N <- rep (16, 6)
Study <- paste0("Experiment", 1:6)
x <- as.data.frame(cbind (r, N, Study))
x
install.packages("meta") #Run once for each computer, or when you want to update the package.
library(meta) #load the package for use, each time you start R, or after detaching it.
metacor(r, N, Study) #Run function metacor from "meta" to meta analyze correlations.
fit <- metacor(r, N, Study) #Run function metacor from "meta" to meta analyze correlations and
#insert output into an object named "fit".
forest(fit)
funnel(fit)
Avraham N Kluger
Apr 13, 2017, 4:58:23 AM4/13/17
to Israel R User Group
Meta analysis
Avi Kluger
April 13, 2017
Ayala’s data with forest and funnel plots
In this example, I fixed the issue of the data frame, so you have an example that will also work when you read the data from your text file.
######The code below just replicate your data #######################################
r <- c(-0.3173759, 0.2253440, -0.3209169, -0.7070982, -0.4100372, -0.5502341)
N <- rep (16, 6)
x <- as.data.frame(cbind (r, N))
x$Study <- paste0("Experiment", 1:6)
x## r N Study
## 1 -0.3173759 16 Experiment1
## 2 0.2253440 16 Experiment2
## 3 -0.3209169 16 Experiment3
## 4 -0.7070982 16 Experiment4
## 5 -0.4100372 16 Experiment5
## 6 -0.5502341 16 Experiment6#install.packages("meta") #Run once for each computer, or when you want to update the package.
library(meta) #load the package for use, each time you start R, or after detaching it. ## Loading 'meta' package (version 4.8-1).
## Type 'help("meta-package")' for a brief overview.metacor(r, N, Study, data = x) #Run function metacor from "meta" to meta analyze correlations.## COR 95%-CI %W(fixed) %W(random)
## Experiment1 -0.3174 [-0.7026; 0.2116] 16.7 16.7
## Experiment2 0.2253 [-0.3044; 0.6486] 16.7 16.7
## Experiment3 -0.3209 [-0.7045; 0.2079] 16.7 16.7
## Experiment4 -0.7071 [-0.8906; -0.3255] 16.7 16.7
## Experiment5 -0.4100 [-0.7527; 0.1075] 16.7 16.7
## Experiment6 -0.5502 [-0.8218; -0.0750] 16.7 16.7
##
## Number of studies combined: k = 6
##
## COR 95%-CI z p-value
## Fixed effect model -0.3754 [-0.5487; -0.1710] -3.49 0.0005
## Random effects model -0.3754 [-0.5987; -0.0979] -2.61 0.0091
##
## Quantifying heterogeneity:
## tau^2 = 0.0603; H = 1.34 [1.00; 2.12]; I^2 = 44.0% [0.0%; 77.8%]
##
## Test of heterogeneity:
## Q d.f. p-value
## 8.92 5 0.1122
##
## Details on meta-analytical method:
## - Inverse variance method
## - DerSimonian-Laird estimator for tau^2
## - Fisher's z transformation of correlationsfit <- metacor(r, N, Study, data = x) #Run function metacor from "meta" to meta analyze correlations and
#insert output into an object named "fit".
forest(fit)funnel(fit)On Wednesday, April 12, 2017 at 10:11:10 AM UTC+3, Ayala Allon wrote:
Ayala Allon
Apr 13, 2017, 6:03:11 AM4/13/17
to Israel R User Group
Hi Avi,
Wow, many tnx!!! After I saw what you did I tried running the metacor() with the following arguments and it worked. I got exactly what you got.
testcor <- metacor(cor$Corr, cor$N, cor$Experiment)
I wonder why I got the error yesterday when trying to specify some of the other arguments.
As for generating the forest plot I used the forestplot package (see below). I will definitely check out the forest plot meta offers
library(forestplot)
cochrane_from_rmeta <-
structure(list(
mean = c(NA, cor$Corr, -.375354256640877, NA),
lower = c(NA, -0.702552236, -0.304359568, -0.704543333, -0.890627584, -0.752741977, -0.821792317, -0.598671803696715, -1),
upper = c(NA, 0.211623028, 0.648598183, 0.207853851, -0.325476408, 0.107523011, -0.0749798348470502, -0.097892973605829, 1)),
.Names = c("mean", "lower", "upper"),
row.names = c(NA, -9L),
class = "data.frame")
tabletext <- cbind(c("Experiment", paste("Experiment", 1:6), "Meta Analysis",""))
forestplot(tabletext,
cochrane_from_rmeta,new_page = TRUE,
is.summary=c(TRUE, rep(FALSE, 6), TRUE, FALSE),
clip = c(-1.0, 1.0),
xlog = FALSE,
col = fpColors(box = "black", line = "black", summary="black"),
hrzl_lines = gpar(col="black"),
txt_gp = fpTxtGp(cex = 1.25, ticks = gpar(cex = 1, fontface = "bold"),
xlab = gpar(cex = 1.25)),
xticks = c(-1.0, -0.5, 0, 0.5, 1),
xticks.digits = 2,
grid = TRUE, title = "Correlation and 95% CI",
xlab = "\nMeta Analysis: r = -0.375 (95% CI: -0.599 - -0.098), Z = -2.609, p = 0.009")
Thank you very much again for the help! After replicating the analysis in R I'm ready to add more studies to the analysis :)
Best,
Ayala
Reply all
Reply to author
Forward
0 new messages