How to concatenate parameters from two -nlcom- commands into one estimation result using -estadd- or... in a program that I will bootstrap?

[Alex Woersching]

I am using nlcom to compute adjusted relative risks and adjusted risk differences following -margins- after a logistic regression GEE model. One nlcom postestimation has several factor variables. The second postestimation computes an effect for age, which I evaluate at age 45 and 55, so must run in a separate margins command from the factor variables, which I do not want evaluated at particular ages. It is easy enough to make two nlcom commands, but I am bootstrapping and want to bootstrap all the parameters at once rather than have to bootstrap twice. I am trying to figure out how to combine or concatenate the nlcom results into one estimation result.

I have tried different approaches with -estadd- with no success including specifying the estimates from one of the nlcoms as a matrix or scalar or writing a new estadd subcommand program (like estadd_mysubcommand at the end of the help file). Here, I simplify the code with only one factor variable for brevity.


capture program drop bootstrap
program define bootstrap, eclass
xtset new_cluster
xtgee y i.b age_10 , ///
corr(exch) family(binomial) link(logit) vce(robust)
est sto boots
margins b, post
nlcom (rr_b:_b[2.b]/_b[1.b]) (rd_b:_b[2.b]-_b[1.b]) , post
est sto boots_fx
est res boots
margins , at(age_10=(4.5 5.5)) post coeflegend
nlcom (rr_age5545:_b[2._at]/_b[1._at]) (rd_age5545:_b[2._at]-_b[1._at]), post coeflegend
est sto boots_age
estadd :boots_fx *I know this doesn't work
end

bootstrap (rr_b:_b[rr_b]) (rr_age5545:_b[rr_age5545]) (rd_b:_b[rd_b]) (rd_age5545:_b[rd_age5545]), reps(10) seed(12345) cluster(cluster) ///
idcluster(new_cluster) group(studyid) : bootstrap
estat bootstrap, all

Thanks!

[Alex Woersching]

I figured it out using matrices, key changes bolded.

capture program drop bootstrap
program define bootstrap, eclass
xtset new_cluster
xtgee y i.b age_10 , ///
corr(exch) family(binomial) link(logit) vce(robust)
est sto boots
margins b, post
nlcom (rr_b:_b[2.b]/_b[1.b]) (rd_b:_b[2.b]-_b[1.b]) , post
matrix boots_fx = e(b)
est res boots
margins , at(age_10=(4.5 5.5)) post coeflegend
nlcom (rr_age5545:_b[2._at]/_b[1._at]) (rd_age5545:_b[2._at]-_b[1._at]), post coeflegend
matrix boots_age = e(b)
matrix boots = (boots_fx,boots_age)
ereturn post boots
end