Hi,

Please, I wanted to know the appropriateness of deriving marginal effects from a multilevel model. I ran conditional and adjusted marginal effects after multilevel models to derive my predicted probabilities for the first and second differences of my average marginal effects and my marginsplot. I was guide by Mize's (2019) paper and analysis on average marginal effects for my analyses. However, one of my Professor wanted me to ask if the average marginal effects I generated applies or recognizes the error structure of the mixed models, which has two levels in my case, for its analyses. I felt that it does but I am not 100% sure. My Professor feels that the marginal effects analyses I conducted does not recognize the two-level error structure of my data. He feels that the margins effects are only based on the individual level data. My commands are:

meglm protest1 c.female##c.demcons employed urban age edu p_interest volun_act volun_inact log_gdp [pw=wght1] || country:female, family(bernoulli) link(logit)

margins, at (demcons=(1(10)41 47) female=(-0.5 0.5)) post
qui mlincom 2-1, add rowname(Males: 1 to 11 yrs) stat(est se p)
qui mlincom 3-1, add rowname(Males: 1 to 21 yrs) stat(est se p)
qui mlincom 4-1, add rowname(Males: 1 to 31 yrs) stat(est se p)
qui mlincom 5-1, add rowname(Males: 1 to 41 yrs) stat(est se p)
mlincom 6-1, add rowname(Males: 1 to 47 yrs) stat(est se p)

qui mlincom 8-7, add rowname(Females: 1 to 11 yrs) stat(est se p)
qui mlincom 9-7, add rowname(Females: 1 to 21 yrs) stat(est se p)
qui mlincom 10-7, add rowname(Females: 1 to 31 yrs) stat(est se p)
qui mlincom 11-7, add rowname(Females: 1 to 41 yrs) stat(est se p)
mlincom 12-7, add rowname(Females: 1 to 47 yrs) stat(est se p)

qui mlincom (8-7) - (2-1), add rowname(Difference: 1 to 11 yrs) stat(est se p)
qui mlincom (9-7) - (3-1), add rowname(Difference: 1 to 21 yrs) stat(est se p)
qui mlincom (10-7) - (4-1), add rowname(Difference: 1 to 31 yrs) stat(est se p)
qui mlincom (11-7) - (5-1), add rowname(Difference: 1 to 41 yrs) stat(est se p)
mlincom (12-7) - (6-1), add rowname(Difference: 1 to 47 yrs) stat(est se p)

Thus, please, do you think the margins effects commands I have in my command recognizes or applies the two-level error structure of the multilevel model?

Thank you and looking forward to your reply.

Eugene Emeka Dim.