Hi All,

Using data from ~2000 hospitals, I have identified patients who underwent 1 of 3 operations (type A, B, C). I would like to explore whether hospitals with high mortality following operation A also have high mortality following operations B and C.

As an initial approach, I fit 3 multi-level, mixed effects models with mortality as my dependent variable and hospital identifier as the random effect. I then estimated the random effects using the "predict" postestimation command and assessed correlations between the random effects using the "corr" command. Following pairwise comparison, I found no association (r<0.1) between random effects of mortality after all three operations.

To assess the efficacy of my method, I split my sample of Type A operations and fit separate multi-level models to each sample. Although I would expect to see significant correlation in random effects across the two samples, I did not, leading me to believe that there is something wrong with my methodology.

I would appreciate any advice.

Code:
melogit died age i.female if operationtype == 1 || hospital: , or base
predict randomeffect1 if operationtype == 1, reffects

melogit died age i.female if operationtype == 2 || hospital: , or base
predict randomeffect2 if operationtype == 2, reffects

melogit died age i.female if operationtype == 3 || hospital: , or base
predict randomeffect3 if operationtype == 3, reffects

preserve
duplicates drop hospital, force
scatter randomeffect1 randomeffect2
corr randomeffect1 randomeffect2