stsplit on different dates in an imputed dataset

[Timo Mo]

Dear Statalists,

Thank you for seeing the post!

I'm curious whether we can stsplit on different dates in an imputed dataset.

I have a dataset of about 98000 non-smokers, 7000 smokers who answered the second questionnaire, and 5000 smokers did not provide detailed age when started smoking.
I first imputed the age when started smoking (age_start), then I calculated the date first started smoking (date_start) . Now I want to stsplit the dataset before Cox regression.

However, it seems that in each imputation (from 1/10), the obs created are different thus not allowed for the MI structure.
In the MI strucre, within each imputation, _mi_id is the unique id for each line. After mi stsplit, the _mi_id will be updataed automatically to remain unique. I tried to manully update the _mi_id after each stsplit, but it didn't work.

Below are the codes I used. Thanks!

mi impute chained ///
(ologit) education income ///
(mlogit, augment) job ///
(regress) age_start ///
= i.sex i.cov1 i.cov2, add(10) rseed(123)

mi passive: gen date_start = dob + (age_start*365.25) if smoke==1 

mi stset date_exit, failure(fail) origin(dob) enter(date_bl) scale(365.25) id(pid)
mi stsplit split, after(date_start) at(0)

And the error :

performing stsplit on m=0;
4002 obs created performing stsplit on m=1;
7408 obs created results could not be duplicated performing stsplit on m=0 resulted in 115147 obs.
​​​​​​​perfroming the same operation on m=1 resulted in 118553 obs. Results should have been the same.

[Timo Mo]

I think the only possible way is to mi extract each imputation, do the stsplit and stcox normally, and then find out a way to pool all the results, but the last step seems to be difficult.