How Do I Conduct Marginal Effects at the Means Analysis Given a Zero-Inflated Beta Regression?

[MD Mubtasim-Fuad]

I am running a zero-inflated beta regression with the DV as anti-Francophone sentiment (from 0 to 1). Let's say I want to analyze the marginal impact on average anti-Francophone sentiment given a change in my social network dummy variable (0 = no close friends of visible minority status/no close friends, 1 = at least one close friend of visible minority status). I want this in proportion-wise terms (impact on the 0-1 proportion scale of the DV). Here is the distribution of my DV:

There are a lot of zero observations, and considering my sample is about 15% Francophone, that is probably one reason why, hence why I am using a zero-inflated beta model. Then why I try running this code:

mfx, predict(pr) at(soc_net_vis_status = 0)
marginsplot

I get the following error:

predict() expression pr unsuitable for marginal-effect calculation

or if I run this code:

margins, dydx(immigrant_status francophone_status region gender_status age_group education_status religious_status income_group urban_status party_id_status soc_net_vis_status ideology_index) atmeans

margins, at(soc_net_vis_status=(0/1))
marginsplot

I get the following error:

invalid dydx() option;
variable francophone_status may not be present in model as factor and continuous predictor

This process worked fine with my other regressions under a glm specification. Just as an example of one:

So why does it not work here? Here is my zoib regression code:

zoib franc_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, zeroinflate(francophone_status) vce(robust)

[Maarten Buis]

A discrete change at the means is probably not as meaningful when you have that many categorical variables. I would either use an average discrete change or a discrete change at specific realistic values of your other explanatory variables. Below is an example of an average discrete change, and it works.

use "http://fmwww.bc.edu/repec/bocode/k/k401.dta", clear

replace totemp = totemp/100

zoib prate mrate totemp age i.sole,       ///
     oneinflate( mrate totemp age i.sole)

margins, dydx(sole)