KHB can be used for several different types of non-linear regression models. This include logistic (logit), ordinal (ologit), and multinominal (mlogit) – but not Poisson or Cox). What it does is that it compares the effect of x on y in a model without any covariates (“reduced model”) with a model with on or more covariates (“full model”).    

The method is not part of standard Stata, so unless you already have done so, install the KHB package: 

ssc install khb

Basic command

khb modeltype yvar xvar || zvar(s)

Useful options

khb modeltype yvar xvar || zvar(s), disentangle  khb modeltype yvar xvar || zvar(s), summary  khb modeltype yvar xvar || zvar(s), or

Explanations
yvar	Insert the name of the y-variable that you want to use.
xvar	Insert the name of the x-variable that you want to use.
zvar(s)	Insert the name of the mediator as well as any confounder(s) that you want to include.
modeltype	Specify what kind of regression model you want to perform (e.g. logit, ologit, or mlogit).
disentangle	Disentangle the contribution of each z-variable.
summary	Summary of decomposition.
or	Display odds ratios.

More information help khb