Regression analysis with controls that apply to only some observations

[Loc Tran]

Hi, all! I have browsed around but haven't found a direct answer to this question and hoped you could help. I'd like to run a regression using survey data on immigrants and native-born populations in the United States, with the native-born population as the reference group. I have information on the year that the immigrant respondents arrived in the country but, as native-born populations, the U.S.-born respondents do not have information there. Is there a way to run a regression that controls for "years since arrival" only for the immigrant respondents but not the U.S.-born ones?

[George Ford]

I think you just replace yearsincearrival with 0 if native. since always zero, it doesn't have a coefficient to mess things up.

clear all
set obs 1000
egen year = seq(), from(2000) to(2020) block(50)
g immigrant = runiform() > 0.8
g arrival = int(runiform(2000, 2020)) * immigrant
g yearsincearrival = year - arrival if immigrant
replace yearsincearrival = 0 if yearsincearrival<-1 | mi(yearsincearrival)
g x = rnormal()
g y = 1 + 1*x*(~immigrant) + 2*x*immigrant + 0.5*yearsincearrival + rnormal()

reg y x yearsincearrival if ~immigrant
reg y x yearsincearrival if immigrant
reg y x c.x#c.immigrant yearsincearrival