OK, so your situation is much more complicated than it originally seemed.
Given that you have interactions of migration with both age group and farm size, there is no such thing as "the effect of migration for age group 15-25." So, no, you cannot estimate what does not exist. There are different effects of migration for age group 15-25, one for each category of farm size. Now, there are various ways you can calculate some sort of summary of these effects, of which the commonest would be to average them, weighting the average by the proportion of study participants in each farm size category. This would be known as "the average marginal effect." This is also a statistic that -margins- could give you if you were able to use it.
I think the best you can do is to calculate an average of appropriate coefficients here. In terms of your model equation what you want is (beta1 + beta6 + beta7 + beta8)/4. Or better, calculate the proportions of the estimation sample in all size categories, let's call them w1, w2, w3, and w4, where w1+w2+w3+w4 = 1. A better estimator is w1*beta1 + w2*beta6 + w3*beta7 + wr*beta8. You can calculate that with -lincom-. Of course, to use -lincom- you will have to replace the beta's with the actual names of the various _b[] coefficients output by your regression.
Given that you have interactions of migration with both age group and farm size, there is no such thing as "the effect of migration for age group 15-25." So, no, you cannot estimate what does not exist. There are different effects of migration for age group 15-25, one for each category of farm size. Now, there are various ways you can calculate some sort of summary of these effects, of which the commonest would be to average them, weighting the average by the proportion of study participants in each farm size category. This would be known as "the average marginal effect." This is also a statistic that -margins- could give you if you were able to use it.
I think the best you can do is to calculate an average of appropriate coefficients here. In terms of your model equation what you want is (beta1 + beta6 + beta7 + beta8)/4. Or better, calculate the proportions of the estimation sample in all size categories, let's call them w1, w2, w3, and w4, where w1+w2+w3+w4 = 1. A better estimator is w1*beta1 + w2*beta6 + w3*beta7 + wr*beta8. You can calculate that with -lincom-. Of course, to use -lincom- you will have to replace the beta's with the actual names of the various _b[] coefficients output by your regression.
Comment