I am working with a meglm model, and need to produce margins out of it. Based on the [ME] meglm postestimation example, let's look at

Code:
webuse bangladesh, clear
meglm c_use ibn.urban age i.children, nocons || district: ibn.urban, nocons fam(bin) link(logit) nofvlabel
margins , over(district)
So far so good. However now I want to figure out if the standard errors just look at the point estimates of the random effects (conditional => smaller standard errors), or integrate over their posterior distribution (larger standard errors). I was expecting to be able to feed this into -margins- with the appropriate options of -predict-, but Stata throws r(198) with no further explanations (Stata 16.1 up to date, born 18 Feb 2020):

Code:
margins , predict(pr , marginal ) over(district)
margins , predict(pr marginal ) over(district)
margins , predict(pr conditional(ebmodes) )
margins , predict(pr conditional(ebmeans) )
margins , predict(pr reffects conditional(ebmeans) )