Dear all,

I have a question about interpretation of my model's results.

I am running a probit model about how parties' policies affect their support across respondents belonging to different occupational groups. My model measures an effect of interaction between a party's position on a policy dimension with respondents' occupational status (such as white-collar, blue-collar etc.). The expectation is that the effect differs across occupational groups.

Individual vote for specific party is a binary variable (a respondent either voted for a given [1] or not [0]), and threfore I used probit model to analyze it.

Party_votet = i.Occupational_status + Party_policy_(t-1) + Occupational_status * Party_policy_(t-1) + age + gender + i.education + FEcountry + FEyear

The probit model returns the results that are consistent with the theory - the interaction between policy and occupational status is significant for some occupational groups but not others (the sign of the coefficients changes as well).

However, when I estimate the marginal effects of the interaction (to check the size of the effect), the marginal effects of policy are negative and significant (go in te same direction) for almost all of the occupational categories. This might go against the original expectation that the effect of party policy varies across different occupational groups.

Could you please specify how to interpret the descrepancy between the results of the main probit model estimates as compared to the marginal effects? I have read very opposing takes on this issue.

I would be very grateful for any advice.