I've used the code from here to create a forest plot, and on another question someone kindly gave me some advice on how to change the scale to a non-log one. Unfortunately, I now have a new problem - the size of the weight boxes on the graph doesn't correspond to the pre-computed weights I've listed. Instead, the code is incorrectly calculating new weights/box sizes based on the hedge's g (referred to as 'odds ratio' in the code):
--e.g. 3% has two different sizes, 14% is smaller than 4%, etc
I've tried modifying the code to avoid that, but either:
1. It calculates the correct box size, but not anchored in the correct place:
2. It anchors the box in the correct place, but makes all the boxes a uniform size:
Is it possible to amend to the code to anchor the boxes in the right place and make them the correct size?
I've been modifying these lines:
/* Compute marker width */
x1=oddsratio / (10 ** (weight/2));
x2=oddsratio * (10 ** (weight/2));
And this is the overall code I've been using:
data forest;
input Study $1-18 grp OddsRatio LowerCL UpperCL Weight;
format weight percent5. Q1 Q3 4.2 oddsratio lowercl uppercl 5.3;
ObsId=_N_;
OR='HG'; LCL='LCL'; UCL='UCL'; WT='Weight';
if grp=1 then do;
weight=weight*.01;
Q1=OddsRatio-OddsRatio*weight;
Q3=OddsRatio+OddsRatio*weight;
lcl2=lowercl;
ucl2=uppercl;
end;
else study2=study;
datalines;
Study1 1 -0.170 -0.901 0.560 7
Study2 1 0.354 -0.530 1.237 4
Study3 1 -0.204 -1.186 0.779 3
Study4 1 -0.848 -1.842 0.146 3
Study5 1 0.495 -0.179 1.169 8
Study6 1 -0.058 -0.568 0.453 14
Study7 1 0.000 -0.494 0.494 15
Study8 1 0.479 -0.410 1.368 4
Study9 1 -0.272 -1.335 0.792 3
Study10 1 0.227 -0.372 0.827 10
Overall 2 0.0459 -0.1762 0.222 .
;
run;
proc sort data=forest out=forest2;
by descending obsid;
run;
/* Add sequence numbers to each observation */
data forest3;
set forest2 end=last;
retain fmtname 'Study' type 'n';
studyvalue=_n_;
if study2='Overall' then study2value=1;
else study2value = .;
/* Output values and formatted strings to data set */
label=study;
start=studyvalue;
end=studyvalue;
output;
if last then do;
hlo='O';
label='Other';
end;
run;
/* Create the format from the data set */
proc format library=work cntlin=forest3;
run;
/* Apply the format to the study values and remove Overall from Study column. */
/* Compute the width of the box proportional to weight in log scale. */
data forest4;
format studyvalue study2value study.;
drop fmtname type label start end hlo pct;
set forest3 (where=(studyvalue > 0)) nobs=nobs;
if studyvalue=1 then studyvalue=.;
/* Compute marker width */
x1=oddsratio / (10 ** (weight/2));
x2=oddsratio * (10 ** (weight/2));
/* Compute top and bottom offsets */
if _n_ = nobs then do;
pct=0.75/nobs;
call symputx("pct", pct);
call symputx("pct2", 2*pct);
call symputx("count", nobs);
end;
run;
ods listing close;
ods html image_dpi=100 path="." file='sgplotforest.html';
ods graphics / reset width=600px height=400px imagename="Forest_Plot_Vector" imagefmt=gif;
title "Meta-analysis ";
title2 h=8pt 'Hedges G and 95% CI';
proc sgplot data=forest4 noautolegend;
scatter y=study2value x=oddsratio / markerattrs=graphdata2(symbol=diamondfilled size=10);
scatter y=studyvalue x=oddsratio / xerrorupper=ucl2 xerrorlower=lcl2 markerattrs=graphdata1(symbol=squarefilled size=0);
vector x=x2 y=studyvalue / xorigin=x1 yorigin=studyvalue lineattrs=graphdata1(thickness=8) noarrowheads;
scatter y=studyvalue x=or / markerchar=oddsratio x2axis;
scatter y=studyvalue x=lcl / markerchar=lowercl x2axis;
scatter y=studyvalue x=ucl / markerchar=uppercl x2axis;
scatter y=studyvalue x=wt / markerchar=weight x2axis;
refline 0 / axis=x lineattrs=(pattern=shortdash) transparency=0.5;
inset ' Favors Sham Treatment' / position=bottomleft;
inset 'Favors Active Treatment' / position=bottom;
xaxis offsetmin=0 offsetmax=0.35 min=-2 max=2 minor display=(nolabel) ;
x2axis offsetmin=0.7 display=(noticks nolabel);
yaxis display=(noticks nolabel) offsetmin=0.1 offsetmax=0.05 values=(1 to &count by 1);
run;
ods html close;
ods listing;
Many thanks for any help!
Hello @j4ne,
I think that's easy:
/* Compute marker width */
c=0.8; /* Factor to adjust absolute marker width */
x1=oddsratio - c*weight;
x2=oddsratio + c*weight;
As far as I see, only the relative marker widths have a meaning in the graph. Hence, we can introduce an arbitrary "scale factor" c to adjust the absolute marker width. I started with the "weight/2" from your code, i.e. c=0.5, but this seemed to be too small to distinguish between weights 3% and 4%. After removing "/2", i.e. with c=1, the markers were slightly longer than in your sample output with "correct box size". Therefore, I introduced variable c and set it to 0.8. But you can play around with this factor to obtain "optimal" box sizes.
Hello @j4ne,
I think that's easy:
/* Compute marker width */
c=0.8; /* Factor to adjust absolute marker width */
x1=oddsratio - c*weight;
x2=oddsratio + c*weight;
As far as I see, only the relative marker widths have a meaning in the graph. Hence, we can introduce an arbitrary "scale factor" c to adjust the absolute marker width. I started with the "weight/2" from your code, i.e. c=0.5, but this seemed to be too small to distinguish between weights 3% and 4%. After removing "/2", i.e. with c=1, the markers were slightly longer than in your sample output with "correct box size". Therefore, I introduced variable c and set it to 0.8. But you can play around with this factor to obtain "optimal" box sizes.
Join us for SAS Innovate 2025, our biggest and most exciting global event of the year, in Orlando, FL, from May 6-9. Sign up by March 14 for just $795.
Learn how use the CAT functions in SAS to join values from multiple variables into a single value.
Find more tutorials on the SAS Users YouTube channel.
Ready to level-up your skills? Choose your own adventure.