omit the main effect in a regression model with an interaction between two dummies when FE are added

[Simona Gamba]

Dear Statalist,
in the following link: http://www.ats.ucla.edu/stat/stata/f...ain_effect.htm (suggested me by Dave Airey) it is explained how results change when a main effect of a model with an interaction term is omitted.
In my model, I want to evaluate the effect of a given policy for both developed and developing countries.
One of the two main effect (being a developing country) is omitted because it is constant over time and my model includes country FE (therefore, it is omitted because of collinearity). Therefore, I just have the dummy variable for the other main effect (having the policy), the interaction term (interaction between having the policy and being a developing country) and the country FE.
Do this alter my results and the way I have to interprete them?
Thank you for your help,
Simona

[Jeff Wooldridge]

Nothing changes in interpreting the coefficients. Your model does have the main effects and one of them gets swept away by the inclusion of country fixed effects. That just means you cannot estimate a "developing country" effect in the absence of the policy. Subject to the usual problems of inferring causality, you can estimate the effect of the policy on non-developing countries (as the coefficient on the policy variable) and whether there is a differential effect for developing countries (as the coefficient on the interaction term).

[Simona Gamba]

Thank you very much for your answer!