Table 2.
Results of mediation analyses
| Model | β | 95% CI | P | Expected power |
|---|---|---|---|---|
| Dominance; aggression; testosterone | ||||
| Average mediation effect1 | −0.309 | [−0.791, −0.010] | 0.039 | 0.323 |
| Average direct effect2 | 0.471 | [0.041, 1.00] | 0.031 | 0.412 |
| Total effect3 | 0.162 | [−0.229, 0.510] | 0.402 | 0.147 |
| Proportion mediated4 | −1.90 | [−21.9, 22.5] | 0.437 | |
| Dominance; creatinine; testosterone | ||||
| Outlier included Average mediation effect | 0.152 | [0.009, 0.360] | 0.037 | 0.478 |
| Average direct effect | 0.471 | [0.033, 1.01] | 0.033 | 0.876 |
| Total effect | 0.623 | [0.182, 1.15] | 0.004 | 0.992 |
| Proportion mediated | 0.244 | [0.014, 0.830] | 0.039 | |
| Outlier excluded Average mediation effect | 0.175 | [0.001, 0.410] | 0.049 | 0.606 |
| Average direct effect | 0.446 | [0.007, 0.970] | 0.048 | 0.877 |
| Total effect | 0.621 | [0.180, 1.140] | 0.003 | 0.991 |
| Proportion mediated | 0.282 | [0.001, 0.950] | 0.050 |
For each analysis, variables are listed in the following order: main predictor; mediator; outcome. Testosterone and creatinine values were individual averages correcting for time of day of sample collection and the sine and cosine of sample collection date. Each value was weighted by the number of urine samples used to generate averages. All mediation models included chimpanzee age and aggression rate as control predictors. Nonparametric bootstrap confidence intervals (CIs) were generated using the percentile method with 10 000 replicates. ‘Expected power’ refers to the expected statistical power for detecting a mediation effect based on beta coefficients derived from multivariate regression; power was calculated using the ‘MedPower’ app (Kenny, 2017). Values in bold and italic fonts indicate P values <0.05 and <0.10, respectively.
‘Average mediation effect’, or average indirect effect, indicates the proportion of the main predictor’s association with the outcome variable attributable to the mediator or its effect on the outcome variable attributable to the mediator.
‘Average direct effect’ indicates the extent to which the main predictor directly predicts the outcome variable (i.e. not through the mediator).
‘Total effect’ refers to the combined direct and mediated/indirect effects.
‘Proportion mediated’ indicates the proportion of the total effect attributable to the mediation effect.