The dataset has three exposure variables, e.g. smoking at time 1, smoking at time 2, and smoking at time 3. I generated 4th exposure variable: smoking at any time. All exposure variables are binary variables (0 vs 1).
Code:
gen smoking_any=0 if smoking1==0 & smoking2==0 & smoking3==0 replace smoking_any=1 if smoking1==1 | smoking2==1 | smoking3==1
smoking1 | smoking2 | smoking3 | smoking_any |
1 | 0 | . | 1 |
0 | 0 | . | . |
0 | 1 | . | 1 |
. | . | 1 | 1 |
0 | 0 | 0 | 0 |
0 | 0 | 0 | 0 |
0 | 0 | 0 | 0 |
. | 0 | 0 | . |
0 | 0 | . | . |
When I include all four exposure variables in the imputation model, it always says ".... predicts data perfectly" or "convergence not achieved".
If I impute "smoking_any" separately from the other three variables, it looks like the prevalence of smoking at any time would be overestimated.
Can I use passive imputation approach after I impute smoking1, smoking2, and smoking3? But it is said that "this method is actually a misspecification of your imputation model and will lead to biased parameter estimates in your analytic model".
Thank you very much.
0 Response to Multiple imputation for correlated exposure variables
Post a Comment