Missings in the stratifying variables

[pumilumi]

I have been using randtreat to assign treatment in RCTs (thanks!) and have a question on how randtreat handles observations that have missings in some of the variables listed in strata().

I bumped into this conundrum: if one of the stratifying variables has some missings, randtreat assigns treatment to those observations (it looks like it pools all observations that have a missing in any of the stratifying variables in one strata and randomizes within). The problem is, although I stratified by sex, I end up getting a strata of missings that contains both women and men. In other words, it assigned treatment to observations that had missing in some of the stratifying variables.

A minimal working example:

//The problem is, although I stratified by sex, I have a strata that contains both women and men.

sysuse bpwide, clear

//In this example, I want to create agegrp-sex strata. I force some observations to have missing in the agregrp or sex.

* I set missing observations in the two stratifying variables

replace agegrp = . if agegrp == 1 & bp_before < 152

replace sex = . if bp_before == 149 

*assign treatment
randtreat, generate(treatment) replace strata(sex agegrp)

*create a strata variable so I can tabulate the stratifying variables against the groups
egen strata = group(sex agegrp)

tab sex treatment if strata == .

*I observe that the missing strata contains and has assigned treatment to both men and women

Any insight is highly appreciated! Pumi.