Let's say I have two DVs, Anti-Muslim Sentiment (ranging continuous 0 to 1 inclusive) and Anti-Aboriginal Sentiment (ranging continuous 0 to 1 inclusive). I run a GLM estimation on both each containing the same covariates. I am trying to compare the magnitude of marginal effects for a given covariate subject to some variation in that covariate. What I am doing now is using the margins command on both regressions, subject to variation in the same covariate for each (say immigrant status - a dummy variable), and I am comparing the baseline likelihood (on the reference category for the covariate for both) and the change in probability when the variation is enacted (0 to 1 change), and looking at the magnitude of the change for both. I rerun this margins command for both regressions, iterating over each covariate. This is, as you can imagine, not only time-consuming but possibly inappropriate for comparison. Would I need to use a "seemingly unrelated regression" to compare both effects? How so?
For context, here is what I have now for my regressions:
Here is a series of margins commands I run on a stored regression - the Anti Muslim Sentiment Index (yes I know this could have simplified with a for loop):
The above I rerun for the Anti Aboriginal Sentiment Index as well. Note I have one other DV that is similar but measures Anti-Racial Minority Sentiment, which I also ran through the same process, though I excluded that in the question for the sake of simplicity. I run those stored margins through a coefplot. Now here is the SUR regression I run on all three:
sureg (muslim_support_index immigrant_status region francophone_status gender_status age_group education_status religious_status income_group urban_status party_id_status soc_net_vis_status ideology_index)(aboriginal_support_index immigrant_status region francophone_status gender_status age_group education_status religious_status income_group urban_status party_id_status soc_net_vis_status ideology_index)(racial_support_index immigrant_status region francophone_status gender_status age_group education_status religious_status income_group urban_status party_id_status soc_net_vis_status ideology_index)
The covariates are the same for each DV (of which there are three) - made a macro that included all controls/covariates, to make it less wordy, but that gave me an error for some reason when using it in the sureg command. Anyways, what does the SUR regression output actually say? How does it help me address the problems of:
1) If the two DVs (or three, if you want to also include the racial minority index) can indeed even be compared in terms of the direction and magnitude of their marginal effects, and if not, if a SUR regression can make them comparable
2) How to actually compare the marginal effects given the SUR regression
For context, here is what I have now for my regressions:
Code:
glm muslim_support_index i.immigrant_status i.francophone_status i.region i.gender_status i.age_group i.education_status i.religious_status i.income_group i.urban_status i.party_id_status i.soc_net_vis_status c.ideology_index, family(binomial) vce(robust)
Code:
glm aboriginal_support_index i.immigrant_status i.francophone_status i.region i.gender_status i.age_group i.education_status i.religious_status i.income_group i.urban_status i.party_id_status i.soc_net_vis_status c.ideology_index, family(binomial) vce(robust)
Code:
margins, dydx(immigrant_status) post estimates store MargMusImm est restore Reg1 margins, dydx(francophone_status) post estimates store MargMusFranc est restore Reg1 margins, dydx(region) post estimates store MargMusReg est restore Reg1 margins, dydx(gender_status) post estimates store MargMusGender est restore Reg1 margins, dydx(age_group) post estimates store MargMusAge est restore Reg1 margins, dydx(education_status) post estimates store MargMusEd est restore Reg1 margins, dydx(religious_status) post estimates store MargMusRelgion est restore Reg1 margins, dydx(income_group) post estimates store MargMusInc est restore Reg1 margins, dydx(urban_status) post estimates store MargMusUrb est restore Reg1 margins, dydx(party_id_status) post estimates store MargMusPID est restore Reg1 margins, dydx(soc_net_vis_status) post estimates store MargMusSocNet est restore Reg1 margins, dydx(ideology_index) post estimates store MargMusIdeo est restore Reg1
sureg (muslim_support_index immigrant_status region francophone_status gender_status age_group education_status religious_status income_group urban_status party_id_status soc_net_vis_status ideology_index)(aboriginal_support_index immigrant_status region francophone_status gender_status age_group education_status religious_status income_group urban_status party_id_status soc_net_vis_status ideology_index)(racial_support_index immigrant_status region francophone_status gender_status age_group education_status religious_status income_group urban_status party_id_status soc_net_vis_status ideology_index)
The covariates are the same for each DV (of which there are three) - made a macro that included all controls/covariates, to make it less wordy, but that gave me an error for some reason when using it in the sureg command. Anyways, what does the SUR regression output actually say? How does it help me address the problems of:
1) If the two DVs (or three, if you want to also include the racial minority index) can indeed even be compared in terms of the direction and magnitude of their marginal effects, and if not, if a SUR regression can make them comparable
2) How to actually compare the marginal effects given the SUR regression
