Indirect effects in mediation analyses should not be tested for statistical significance

Mediation analysis is used to gain insight into the mechanisms of exposure-outcome effects by dividing this effect into a direct and an indirect effect. One of the problems of mediation analysis is that in many situations, the standard error of the indirect effect is much lower than the standard errors of the total and direct effect. Because this problem is ignored in the epidemiological literature, the purpose of this paper was to illustrate this problem and to provide an advice regarding the statistical testing of indirect effects in mediation analysis.

To illustrate the problem of the estimation of the standard error of the indirect effect two real life datasets and several simulations are used.

The paper shows that the problem of estimating the standard error of the indirect effect was most pronounced when the relationship between exposure and mediator and the relationship between mediator and outcome were equally strong. Furthermore, the magnitude of the estimation problem is different for different strengths of the mediation effect.

The indirect effect in mediation analysis should not be tested for statistical significance but the importance of mediation should be evaluated by its clinical relevance.