xtivreg2, interaction of the endogenous variable. Not same results when split sample and when use the interaction

[Ernest Dautovic]

I am running xtivreg2 for a regression with interaction:

gen hhrlichild=hhreallabinc*child
gen bonuschild=bonus*child

xtivreg2 y (hhreallabinc hhrlichild = bonus bonuschild) child $xhh1 $xcity pxy* ind* countytrend*,  fe robust cluster(county)

Note that the dummy child is dropped by Stata because of collinearity. Here I get the beta on the interaction term of 0.26.

However, I don't get the same result when I split the sample by households having or not a child:

xtivreg2 y (hhreallabinc = bonus) $xhh1 $xcity pxy* ind* countytrend* if child==1,  fe robust cluster(county)

xtivreg2 y (hhreallabinc = bonus ) $xhh1 $xcity pxy* ind* countytrend* if child==0,  fe robust cluster(county)

Here I get the difference between the two beta coefficients is 1.2

Could it be that, because of too many dummy variables and trends (more than 600), I don't get the same results?

I did the same regression using

reghdfe

but get the same different results.