[R] metafor/ meta-regression

[Viechtbauer Wolfgang (STAT)]

Dear Fernanda,

Currently, there is no option in metafor that will automatically give you standardized coefficients for meta-regression models. The question is also how you would like to standardize those coefficients. In the usual regression models, the standardized coefficients are those that you would obtain if both the dependent and the independent variables are all rescalled to have a mean of 0 and a standard deviation of 1.

For meta-regression models, I don't see how rescalling the dependent variable (i.e., the outcome) would give you something sensible. There are some additional issues with the fact that you would also have to rescale the sampling variances accordingly.

One possibility may be to just rescale the predictors. If this is what you want, then you can always just rescale the variables manually (the scale() function will help you do that).

For example:

data(dat.bcg)

dat <- escalc(measure="RR", ai=tpos, bi=tneg, ci=cpos, di=cneg, data=dat.bcg, append=TRUE)
res <- rma(yi, vi, mods=cbind(ablat, year), data=dat, method="REML")
res

dat$ablat.std <- c(scale(dat$ablat))
dat$year.std  <- c(scale(dat$year))

res <- rma(yi, vi, mods=cbind(ablat.std, year.std), data=dat, method="REML")
res

Part of the output:

           estimate      se     zval    pval    ci.lb    ci.ub
intrcpt     -0.7321  0.1238  -5.9137  <.0001  -0.9748  -0.4895  ***
ablat.std   -0.4046  0.1478  -2.7371  0.0062  -0.6943  -0.1149   **
year.std     0.0209  0.1605   0.1299  0.8966  -0.2938   0.3355

So, the results indicate that, for a one standard deviation increase in absolute latitude, the relative risk decreases on average by .41 points. And for a one standard deviation increase in year, the relative risk increases on average by .02 points.

Whether one can determine the relative importance of predictors this way (this at least seems to be the reason why people often ask for standardized coefficients) is another issue.

Best,

--
Wolfgang Viechtbauer
Department of Psychiatry and Neuropsychology
School for Mental Health and Neuroscience
Maastricht University, P.O. Box 616
6200 MD Maastricht, The Netherlands
Tel: +31 (43) 368-5248
Fax: +31 (43) 368-8689
Web: http://www.wvbauer.com



----Original Message----
From: r-help-bounces at r-project.org
[mailto:r-help-bounces at r-project.org] On Behalf Of Fernanda Melo
Carneiro Sent: Wednesday, January 12, 2011 20:18 To:
r-help at stat.math.ethz.ch Subject: [R] metafor/ meta-regression

> Hi I have tryed to do the meta-regression in metafor package, but I
> would like to get the standardized coefficients for each variable,
> however in command:
>
> res<-rma.uni (yi, vi, method="REML",
> mods=~cota+DL+uso+gadiente+idade, data=turbidez)
>
> I just have the coefficients no standardized (estimate) of the
> multiple regression.
>
> What I need to do?
>
> Thanks
>
> Fernanda Melo Carneiro contato: (62) 3521-1480 e 8121-7374
> www.ecoevol.ufg.br Laboratório de Ecologia Teórica e Síntese (UFG)