Marginal Effects after running -glm- estimator

[Drishta Tengur]

I want to calculate the marginal effects of my control variables after regressing my dependent variable Y, which is bounded between 0 and 1. The following is my regression:
- glm Y a b c d e f i.country, fa(b) link(logit) vce(robust) -, where a and b are dummy variables and c and d are interaction terms between the dummy variables and the variable e.

I am not sure if I should use:
- margins a b c d e f, - or - margins a b c d e f, dydx -

Can somebody please help me with this query?

[Maarten Buis]

Both are wrong. You need to use factor variables for categorical variables and for interactions. Without that Stata does not know what the relationship between the variables is, and will give you wrong answers.

Here are various ways of using margins after such a model. All of them (and more) are correct, but give you a different view on the same problem. So it is worthwhile to look at all of them. Once you see how they are related you will really start to understand your model.

use http://fmwww.bc.edu/repec/bocode/c/citybudget.dta, clear
gen polit = minoritylef + 2*noleft
label define polit 0 "majority left" ///
                   1 "minority left" ///
                   2 "no left"
label value polit polit
label variable polit "political orientation city government"    
               
glm governing i.polit##c.houseval popdens, ///
    link(logit) family(binomial) vce(robust)

// look at a graph of predicted proportions    
margins, at(houseval=(.75(.25)3.5) polit=(0/2) popdens=.5)
marginsplot

// marginal effects
margins, dydx(*)

// different marginal effects for different values of polit
margins, dydx(houseval) over(polit)