How to interpret "marginals" in case of the zero-weighted binomial model?

[Shashank Shukla]

How to interpret "marginals" in case of the zero-weighted binomial model?

It may be noted that in the case of ordered probit, one gets a set of numbers as marginals as there is multiple ordered category of the dependent variable. Binomial also may have values between 0 and N. Is the default marginal computed by STATA for binomial regression (say with atmeans) is true for the average or expected value of the dependent variable as well as all independent variables. For example, if Y is our dependent variable, and X is an independent variable and the marginal of X is given as 0.12, can I conclude that one unit increase of X at its mean will lead to 0.12 units of increase in E(Y)?

[Maarten Buis]

There is no such thing as a zero-weighted binomial. Do you mean zero-inflated negative binomial (Stata command zinb)?

It is easiest for us if you gave us exactly what you typed in Stata, both the model and the subsequent margins command. That way we can be sure what you mean. Otherwise we often have to guess at what you might mean (and often guess wrong).

[Shashank Shukla]

It was my error. I meant Zero-inflated binomial regression
The command I typed is as below:

zib FIH Avg_Yrs_School_Female_Village Per_Capita_Cons_Vill [weight= Freq_Females_Village], n(N) inflate(Prop_PPLReport_Violence_Village Prop_PPL_FacedViolence_Village Per_Capita_Cons_Vill) nolog vce(robust)

For marginals I gave the command
margins , dydx(*) atmeans

[Shashank Shukla]

The question for which I need guidance is as below:
Is the default marginal computed by STATA for binomial regression (say with atmeans) is true for the average or expected value of the dependent variable as well as all independent variables?

For example, if Y is our dependent variable, and X is an independent variable and the marginal of X is given as 0.12, can I conclude that one unit increase of X at its mean will lead to 0.12 units of increase in E(Y)?

[Maarten Buis]

The default for margins is the default for predict for that command. All estimation commands have their own default prediction. This is typically documented in help <command_name>_postestimation. This typically includes a list of links to allowed post-estimation commands, one of which is predict. You click on that, which will show a list of possible predictions, and that list indicates which one is the default. So in your case, the default is the expected number of successes. Since you ask for dydx(), it is the change in the expected number of successes for a unit change in x.

[Shashank Shukla]

Makes sense. Many thanks Prof. Maarten.