Mediation analysis revisited: Practical suggestions for addressing common deficiencies

Abstract Four issues that can affect statistical conclusions from mediation analysis are presented here: The implications of omitting mediators; not conducting reverse mediation analysis; using inappropriate measures; and not considering a wider array of experiment-based methods. Suggestions for addressing each of these are advanced. Previous issues of AMJ, JMR and JCR are then examined to gauge the extent to which these suggestions were used. Less than half of the published papers inspected (44.4% of the total) endeavored to address at least three of the four issues raised above. AMJ authors will realize higher statistical as well as theoretical rigor if they consider these suggestions.