[R-meta] R Bayesian Meta-regression, how to test covariates?
Mark Pilling
ma_pilling at yahoo.co.uk
Wed Feb 14 11:47:27 CET 2018
0 down vote favorite
I'm using the bmeta package in R to conduct a Bayesian Meta-regression.The function allows covariates to be entered by including them as "X" variables in a data.list object.
My question (possibly stupid), is where are the regression covariates in the output?
Data from p12 of bmeta_manual.pdf, with a made-up covariate added:
Study study year y var prec Covar_bin
1 Hwang 2004 1.3 3.0888167 0.3237486 0
2 Ashrafi 2005 -4.79 4.2111455 0.2374651 0
3 Ensieh 2010 -2.49 3.0201244 0.3311122 0
4 Agha 2010 -3.39 1.5519691 0.6443427 1
5 Kim 2012 -0.5 0.6469829 1.5456359 1
6 Bulent 2012 -1 1.6380733 0.6104733 1
Reproducible example
data<-read.csv("Book1.csv")
Meta-analysis
data.list1<-list(y=data$y, prec = data$prec)
x1<-bmeta(data=data.list1, outcome ="ctns", model="std.mv", type="ran")
x1
mu=-1.613
Meta-regression
data.list2<-list(y=data$y, prec = data$prec, X=cbind(data$Covar_bin))
x2<-bmeta(data=data.list2, outcome ="ctns", model="reg.mv", type="ran")
x2
mu=-1.717
So the overall effect estimate (mu) has changed, but why is there no output for the regression covariate to determine if it should be included?
[[alternative HTML version deleted]]
More information about the R-sig-meta-analysis
mailing list