TEst Margins

[Sara Reiter]

Dear statalist users

My aim:
I´m running a logistic regression with an interaction term (EducationLevel##EthnicGroup).
Dependent variable is participation in training. I want to compare marginal effects of participation between the two ethnic groups by educational level (low, middle, high).

logistic i.EducationLevel##i.EthnicGroup
.margins, dydx(EthnicGroup)at(EducationLevel=(1(1)3))

.

My Sample:
My problem is that one of the ethnic groups (EthnicGroup1) is rather large (n=6212) so that all educational show an even amount of events (see table group1).
The second group (EthnicGroup2) is small (n=315) and an in lower educational level there are only 13 events (see table group2).

. tab Education Participation

                            Participation in training
Education Level
(group1)                    0              1        Total
------------------+----------------------+----------
          low                836           304        1140
          middle         2041         1278        3319
          high              895            865        1760
------------------+----------------------+----------
           Total           3772        2447         6212



. tab Education Participation

                              Participation in training
Education Level  
(group2)                   0              1         Total
------------------+----------------------+----------
          low                91             13        104
          middle         112             25        137
          high              47              27         74
------------------+----------------------+----------
           Total            250            65        315

My Question:
Should I consider using estimation for rare data (i.e. Firth Method)?

Thank you for your help
Sara