im trying to measure the impact of child's health (height for age) on their school enrollment (0/1)
schoolenroll= male + heightforage + male*heightforage + othervariables

For the whole sample, heightforage was positive indicating + relationship.
but if i check the marginal effects specifically for boys and girls:

margins,dydx( HeightZscoreP) at (male=(1 0))

Average marginal effects Number of obs = 527
Model VCE : Linearized

Expression : Pr(schoolenroll3), predict()
dy/dx w.r.t. : HeightZscoreP

1._at : male = 1

2._at : male = 0

-------------------------------------------------------------------------------
| Delta-method
| dy/dx Std. Err. z P>|z| [95% Conf. Interval]
--------------+----------------------------------------------------------------
HeightZscoreP |
_at |
1 | .0113025 .0091748 1.23 0.218 -.0066798 .0292848
2 | .0688765 .0109976 6.26 0.000 .0473216 .0904314
-------------------------------------------------------------------------------

the relationship is insignificant for boys.

my supervisor told me to get rid of the interaction term, it will make interpretation easier
so i regressed separately for boys first:
preserve
keep if male==1
svy: prob schoolenroll3 c.HeightZscoreP motherage mothedu householdexp bananadef_3 sugardef_3 oniondef_3 potatoesdef_3 schooldist d
> istrdummy1-distrdummy12 distrdummy14- distrdummy18
(running probit on estimation sample)

-------------------------------------------------------------------------------
| Linearized
schoolenroll3 | Coef. Std. Err. t P>|t| [95% Conf. Interval]
--------------+----------------------------------------------------------------
HeightZscoreP | .2029357 .0609489 3.33 0.004 .0737297 .3321417

here the coefficient has significant relationship with boys.

can someone explain why the results oppose? and which one should i do? marginal effects or run separately for boys and girls