摘要: We show that the overall effect of an exposure on outcome, in presence a mediator with which may interact, can be decomposed into four components: (i) absence mediator, (ii) interactive when is left to what would exposure, (iii) mediated interaction and (iv) pure effect. These components respectively correspond portion due neither mediation nor interaction, just (but not mediation), both interaction). It shown this four-way decomposition unites methods attribute effects interactions assess mediation. Different combinations these measures for mediation, while other interaction. The carried out using standard statistical models software provided estimate each components. provides greatest insight how much mediated, together, neither.