Does mediation tell you effect of predictor above that of the mediator?

Question

I am currently conducting a set of analyses examining the relationship between two predictors and an outcome. For example, the relationship between motivation (predictor 1), revision (predictor 2), and performance in an exam (outcome).

I have reason to believe that predictor 2 (e.g. revision) may mediate the relationship between predictor 1 (motivation) and the outcome (performance on the exam). I have therefore ran a mediation model and find evidence of full mediation after controlling for covariates.

I am also interested in whether a model containing the predictor (e.g. motivation) and the mediator (e.g. revision) is more predictor of the outcome than the mediator alone. Can I obtain this from the mediation model, or would I need to conduct additional analyses to examine this (e.g. separate regression analyses including only the mediator (model 1) and then the mediator and the predictor (model 2), and then compare these models)?

Answer 1

The test of the direct effect is a test of whether adding the predictor to a model with just the mediator (and covariates) explains more variability in the outcome than the model with just the mediator (and covariates). The test statistic for the direct effect is the same test statistic for the change in $R^2$ or ANOVA test you would run to compare the two models. Given this, you already know your answer; if you have full mediation, that suggests the direct effect is negligible.

If you want to claim full mediation, though, you should run an equivalence to test to test that the direct effect isn't negligible. Just because the direct effect is nonsignificant, doesn't mean it is equal to zero. A wide confidence interval for the direct effect could include meaningful nonzero values that would oppose the full mediation interpretation.