Stratify or include interaaction terms when assessing the associations for boys and girls between several exposure variables and outcome

[Kjell Weyde]

Dear all,
I have a model with 6 exposure variables and some other covariates. I would like to analyse whether the associations between each of the 6 exposures and the outcome are different for boys and girls. However, I am not sure what the best way to do this is. Should I include an interaction term between sex and each of the exposures:

regress outcome (c.X1 c.X2 c.X3 c.X4 c.X5 c.X6)##i.sex Maternal_age i.Maternal_edu i.Parity i.SMOKING_X

and say that there are statistically significant sex differences if any of the c.X#i.sex terms have a p<.05?
Or are there other options in Stata 15 that could be better, for example the -suest-? For example.

regress outcome X1 X2 X3 X4 X5 X6 Maternal_age i.Maternal_edu i.Parity i.SMOKING_X if sex==1
est store m1
regress outcome X1 X2 X3 X4 X5 X6 Maternal_age i.Maternal_edu i.Parity i.SMOKING_X if sex==2
est store m2
suest m1 m2
test [m1_mean]X1=[m2_mean]X1
test [m1_mean]X2=[m2_mean]X2
test [m1_mean]X3=[m2_mean]X3
test [m1_mean]X4=[m2_mean]X4
test [m1_mean]X5=[m2_mean]X5
test [m1_mean]X6=[m2_mean]X6

Best,
Kjell Weyde

[Carlo Lazzaro]

Kjell:
I would add an interaction term and then rely on -margins- and -marginsplot-.

[Phil Bromiley]

While I agree with Carlo that interaction is the way to go, it assumes the error variance is the same for both sexes. You might want to allow different error variances by sex (see the vce option in regress).