Fig. 2
From: Required sample size to detect mediation in 3-level implementation studies
Multilevel mediation model (3-2-1). Note: The diagram presents the 3-2-1 mediation design for which statistical power was calculated in this study. The boxes signify each construct in the design and show the levels at which the construct exhibits variance: X = independent variable which varies only at level 3; M = mediator which resides at level 2 but exhibits variance at levels 2 and 3 (due to clustering); Y = outcome which resides at level 1 but exhibits variance at levels 1, 2, and 3 (due to clustering). The variance of M and Y at the higher levels of analysis are represented by ICC values. Arrows indicate effects that can be estimated through conventional multilevel regression (MVM) [32] or through multilevel structure equation modeling (MSEM) [36]. The paths that make up the indirect effect (i.e., mediation at level 3) are a3*b3. The c’3 path represents the direct effect. The b2 path is typically not of substantive interest; it represents the relationship between the within-organization component of M and within-organization component of Y