Making Apples-to-Apples Comparisons Between Internal and External Studies Using Propensity Scores (3)
Examples using SAS:
SAS code 1:
title 'PSMATCH: ATT Weighting';
proc psmatch data=t1 region=treated;
class trt genotype;
psmodel trt(Treated='study treatment')= b_age genotype b_var;
where vsn=0;
assess lps var=(b_age genotype b_var)/ weight=attwgt;
output out(OBS=region)=Outex;
run;
SAS code 2:
title 'PSMATCH: ATT Weighting';
proc psmatch data=t1 region=allobs(psmin=0.05 psmax=0.95);
class trt genotype;
psmodel trt(Treated='study treatment')= b_age genotype b_var;
where vsn=0;
assess lps var=(b_age genotype b_var)/ weight=attwgt;
output out(OBS=region)=Outex1;
| Aspect | region=treated |
region=allobs(psmin=0.05 psmax=0.95) |
|---|---|---|
| Who gets dropped | Controls outside treated PS range | Both groups outside [0.05,0.95] |
| Treated retention | All treated kept | Some treated may be dropped |
| Target estimand | ATT in full treated population | ATT in trimmed treated population |
| Weight stability / ESS | Potentially lower ESS (more extreme weights) | Typically higher ESS (fewer extremes) |
| Bias–variance trade‑off | Better generalizability to all treated; more variance | Better precision; less generalizable |
When to use which
-
Use
region=treatedif your primary estimand is ATT in all treated and you prefer not to discard any treated subjects. Be prepared to assess weight extremes and ESS. -
Use
region=allobs(psmin=…, psmax=…)if you want to stabilize the analysis (trim tails), accepting a narrower target population for better precision.
Comments
Post a Comment