For this example, we want to see if grade point average (z) mediates the association between exposure to bullying (x) and educational level (y). 

Dataset

StataData1.dta

Variable name	educ
Variable label	Educational level (Age 40, Year 2010)
Value labels	1=Compulsory 2=Upper secondary 3=University

Variable name	gpa
Variable label	Grade point average (Age 15, Year 1985)
Value labels	N/A

Variable name	bullied
Variable label	Exposure to bullying (Age 15, Year 1985)
Value labels	0=No 1=Yes

Define the analytical sample 

We start by defining the analytical sample: 

gen pop_mediate2=1 if educ!=. & gpa!=. & bullied!=.

Let us have a quick look at the variables: 

sum educ gpa bullied if pop_mediate2==1

Regression models

Now, we can run the regression model with the khb command.  

khb ologit educ bullied || gpa if pop_mediate2==1, summary disentangle or

The model without any z-variables (the “reduced” model) shows that there is a negative (OR=0.67) and statistically significant association (95 % CI=0.58 to 0.77) between bullied and educ. This means that individuals who were exposed to bullying at age 15 have lower odds of attaining a high level of education as adults, in comparison to those who were not exposed to bullying. In the model where the z-variable gpa is included (the “full” model), the association is still negative but very weak and stastistically non-significant (OR=0.95, 95% CI=0.83 to 1.10).  

In the table called Summary of confounding, we can see that the amount of the association explained by the z-variables (in this case, we only included gpa), is 88%. This is also shown specified further in the table called Components of Difference.