std. err in regression results and marginal effects

[Meng Yu]

I noticed when calculating marginal effects, std. err changes from that in the regression results. However, the significance level seems to remain the same, i.e. if it is insignificant in the regression results, so is in the marginal effects calculation. Can anyone explain why this is so? I appreciate your help.

[Joseph Coveney]

You have an illustration of this to show what you're talking about?

[Meng Yu]

Regression results:

	Observed	Bstrap *
	Coef.	Std. Err.	z	P>z	[95% Conf.	Interval]

participation#sex
not participate #male	0	(empty)
not participate #female	0	(empty)
participate #male	0	(empty)
participate #female	0.097995	0.113563	0.86	0.388	-0.12458	0.320574

marginal effects:

		Delta-method
	dy/dx	Std. Err.	z	P>z	[95% Conf.	Interval]
1.particpate
_at
male	0.024421	0.086667	0.28	0.778	-0.14544	0.194284
female	0.122415	0.081307	1.51	0.132	-0.03694	0.281773

I wonder how the marginal effect of female is calculated, including the std. err.

[Phil Bromiley]

If you're getting all these non-estimates, I would change the model to only include what you can estimate. Your model should also include main effects. You don't need participate#male and participate#female - just include the main effect which will be for males (or females) and the one interaction (which is the difference from the main effect). This difference (.097) is the difference between the two effects in the margins command.

Sometimes, margins won't even work if you have variables in the model that can not be estimated.

[Meng Yu]

Thank you, Phil, for your reply. I have included the main effects in the model. What is shown is only the interaction coefficients. Both variables are dichotomous variables.
I wonder how the std. err. of the marginal effect of male/female is calculated. If participate#female is not significant, does it mean the marginal effect of male/female are both insignificant?