I have a large binary logit model (cross-section) with many variables, mostly dummies. From the information on age, I have created 6 dummies: 17-24, 25-34, 35-44, 45-54 (reference group), 55-64 and 65+. From the information on migration background, I have created 3 dummies: NoMigrationBackground (reference group), EU-MigrationBackground and NonEU-MigrationBackground (EU as in European Union).
Now, I want to add the interaction effect NonEU-MigrationBackground*65+. My question is, how would I interpret the marginal effect (APE)? To which group do I compare it to?
Let's say it is 10%. Do I then say: People with a non-EU migration background who are at least 65 years old have an increased likelihood of Y=1 of on average 10% compared to people younger than 65 years old with either an EU migration background or no migration background, all else equal? Or is it only compared to the original reference groups, so ...compared to people with no migration background who are 45-54 years old?