A mediator is a variable that is influenced by the x-variable and influences the y-variable. In other words, some (it could be a little or a lot) of the effect of x on y is mediated through z. For example, let us say that we are interested in the association between parents’ educational attainment (x) and children’s success on the labour market (y). It could be reasonable to assume that the educational attainment of the parents (x) influences children’s own educational attainment (z), which in turn affects their following success on the labour market (y).

Pathways and mechanisms

In data analysis, we often talk about “explaining” an association by the inclusion of certain mediating variables. Particularly when one has a data material that consists of information collected across several points in time (i.e. longitudinal or life course data), it is common to talk about mediation as “pathways” or “mechanisms”.

Mediation analysis

Traditionally, mediators have been treated similarly to confounders in multiple regression analysis. This means that one includes one or more mediators in the model and see how much is explained of the association that we are interested in. This approach has been heavily criticised in the context of non-linear regression models (for statistical reasons that we will not discuss here). There are some specific types of mediation analysis that can be used; one of them is the KHB method, which will be explored in Mediation analysis.