Since mechanisms of behavior change are not always evaluated in physical activity interventions, current interventions are limited until these mechanisms are better understood (Bauman, Sallis, Dzewaltowski, & Owen, 2002). Therefore, studies are needed that examine mediating variables, derived from theory, in the design, implementation, and evaluation of interventions. A mediator is a variable that must be included in an intervention in order for a specific change in a dependent variable or outcome to occur (MacKinnon, 2008). MacKinnon (2008) describes several methods of identifying mediators using statistical procedures, including the causal inference approach, difference in coefficients, product of coefficients, structural equation modeling and bootstrap estimates of the mediated effect.
Ornes, Lynne L. and Ransdell, Lynda B.. (2010). "A Pilot Study Examining Exercise Self-Efficacy as a Mediator for Walking Behavior in College-Age Women". Perceptual and Motor Skills, 110(3), 1098-1104. http://dx.doi.org/10.2466/pms.110.C.1098-1104