9.4 Reporting Mediation Results

We analyse a path model as a series of regression models, so the general rules for reporting mediation are the same as for reporting regression analyses (see Section 7.2). If you summarize results in a table, make sure that the table includes:

  1. The unstandardized regression coefficients for all direct and indirect effects tested in the regression models.

  2. The confidence intervals and significance levels of the unstandardized effects.

  3. The F test and measure of model fit (\(R^2\)) for each regression model.

Table 9.2: Unstandardized effects in a model regressing newspaper reading time on age with one mediator (News Site Use) and two covariates (Education, Political Interest). Theoretical approximation for direct effects, bootstrap results for indirect effects, using 5,000 bootstraps.
B 95% CI
Outcome: News Site Use
constant 6.62 *** [5.92; 7.31]
age -0.93 *** [-0.97; -0.88]
education 0.06 * [0.01; 0.11]
pol.interest 0.12 *** [0.06; 0.17]
R2 0.86
F (3, 308) 617.40 ***
Outcome: Newspaper Reading Time
constant 13.59 ** [5.26; 21.93]
age 4.54 *** [3.62; 5.47]
education 0.06 [-0.34; 0.46]
pol.interest 0.52 * [0.07; 0.96]
newssite -1.55 ** [-2.47; -0.64]
R2 0.79
F (4, 307) 290.85 ***
Indirect Effect
Age > News Site Use > Reading Time 1.44 [0.61; 2.17]
Note. * p < .05. ** p < .01. *** p < .001.

A path model may yield a lot of direct effects, so it is good practice to present results as a path diagram with the values of the standardized or unstandardized regression coefficients as labels to the arrows. A path model conveniently summarizes the results for the reader (Figure 9.11). Remember that we don’t use standardized regression coefficients if the predictor or a covariate is dichotomous variable or a set of dummy variables (see Section 6.1.2).

Figure 9.11: Unstandardized direct effects for a path model with one mediator and two covariates.

If effect mediation is central to your report, focus your presentation and interpretation on the indirect effects and compare them to the direct effects. Report the size and confidence interval of each indirect effect. If possible, add both the direct and indirect effect to a diagram such as Figure 9.11.

Interpret an unstandardized indirect effect just like any unstandardized regression effect, namely, as the predicted difference in the outcome for a one unit difference in the predictor. It is usually interesting to compare the sizes of the direct and indirect effects. Is the effect predominantly mediated in the model or is only a minor part of the effect mediated in the model?

Inform the reader that you bootstrapped the indirect effect and report the number of bootstrap samples and the method used for the confidence intervals (see Section 9.5). For a more elaborate discussion of reporting mediation, see Hayes (2013: 198-202).