Skip to content

tweaking forest plot (metafor package)

3 messages · eesteves at ualg.pt, Michael Dewey, Viechtbauer Wolfgang (STAT)

eesteves at ualg.pt
# 
Dear All,

I'm having trouble tweaking a forest plot made using the R  
meta-analysis package metafor. I did the analysis based upon the  
correlation coeff from studies and plotted the corresponding forest  
plot easily
I'd like to subdivide the "table" by the moderator 'grupo' - i.e.  
create the necessary spacings in the list of studies to accommodate  
the moderator statistics (cf.  
http://rgm2.lab.nig.ac.jp/RGM2/func.php?rd_id=metafor:addpoly.default).

My main issue is that I cannot really understand how the function  
(forest()) really uses space; specifically, how to use the parameters  
'xlim', 'ylim' and 'rows' in the function? how can I change the scale  
from 0-1 to 0.5-1?

Hopefully I've made myself (relatively) clear but... Thanks.

Eduardo Esteves
Michael Dewey
#
At 00:10 18/07/2012, eesteves at ualg.pt wrote:
Not sure why you use cbind(grupo) but it will not change much.
An example of what went wrong would help but I 
think you will find that the rows argument tells 
forest where to put the results of the individual 
trials so if you give it non-consecutive integers 
you will get gaps in the plot which you can fill 
with addpoly. So if rows does not contain 4 you 
can use addpoly to put something in row 4. As far 
as xlim and ylim are concerned my advice would be 
to try a whole range of values and see what 
happens as this is one of those cases where it is 
easier to experiment than to explain. If all else 
fails either (a) look at the code (b) wait for 
Wolfgang to ell you and me what really happens.
Michael Dewey
info at aghmed.fsnet.co.uk
http://www.aghmed.fsnet.co.uk/home.html
Viechtbauer Wolfgang (STAT)
# 
I think Michael already gave you some really good hints. To add a bit more details to this, first of all, I suggest to use:

q2 <- rma(yi, vi, mods=cbind(grupo), data=qim)
forest(q2, transf=transf.ztor, digits=3, alim=c(0,1), refline=.5, addfit=FALSE)

so that the fitted values for that categorical moderator are not actually shown in the plot yet. Then use:

par("usr")

after you create the forest plot. That will tell you what the xlim values are that were chosen by default (the first and second value). Start adjusting these values with the xlim argument to see how that expands and shrinks the horizontal space.

Next, try the following. Suppose you have k=20 observed correlations in total that will be shown in the forest plot. Then the default value for ylim (for a model that does includes a moderator, as you have) will be ylim=c(0.5, 23) (i.e., the second value is k+3). The first study will be placed in row 20 of the plot, the second in row 19, and so on, all the way down to row 1 for the 20th study. So, try this:

abline(h=20, lwd=3)
abline(h=1, lwd=3)

which will show you where that first and last study are being placed.

In row 21, the horizontal line is plotted. So try this:

abline(h=20, lwd=3)

which shows you where that line is being placed.

Rows 22 and 23 are for some extra white space at the top of the figure to add column headings (with the text() function). So try this:

text(par("usr")[1], 22, "Study", pos=4, font=2)
text(par("usr")[2], 22, "Cor [95% CI]", pos=2, font=2)

which will add column headings for the first and last column. Of course, you can adjust the value 22 to something slightly different (i.e., to adjust the vertical placement of the headings).

Now try using ylim=c(0.5, 30) in the call to forest(). You will see that a bunch of extra vertical space is added. The first study is still placed in row 20, going down again to row 1 for the 20th study. Confirm this with:

abline(h=20, lwd=3)
abline(h=1, lwd=3)

Now, you want to use the row argument to actually tell the forest() function where those studies should be placed. Try rows=c(25:21, 15:1) in the call to forest(), which will place study 1 in row 25, study 2 in row 24, going down to study 5 in row 21, and then study 6 in row 15, study 7 in row 14, going down to study 1 in row 1. Again, you can use something like:

abline(h=25, lwd=3)
abline(h=15, lwd=3)

to check on the vertical placement of the 1st and 6th study.

So now you have full control over the placement of the studies and you can add space as needed. Then you can use the addpoly() function to add summary polygons for the various levels of the "grupo" factor. I suggest carefully going through the last example in:

help(addpoly.default)

to see a detailed example of this.

Finally, to change the axis limits to .5 and 1, just use alim=c(0.5, 1) in the call to forest().

I hope this helps!

Best,

Wolfgang

--   
Wolfgang Viechtbauer, Ph.D., Statistician   
Department of Psychiatry and Psychology   
School for Mental Health and Neuroscience   
Faculty of Health, Medicine, and Life Sciences   
Maastricht University, P.O. Box 616 (VIJV1)   
6200 MD Maastricht, The Netherlands   
+31 (43) 388-4170 | http://www.wvbauer.com