Advances in mediation analysis can facilitate nursing research

Two recent advances in the statistical methods for testing hypotheses about mediation effects are important for nursing science. First, bootstrap sampling distributions provide more accurate tests of hypotheses about mediated effects. Second, methods for testing statistical hypotheses about subgroup differences in mediation models (moderated mediation) are now well developed.

The aims of this study were to demonstrate the use and relatively simple computation of bootstrap sampling distributions in tests of mediation effects and to demonstrate a recently refined method for testing hypotheses about moderated mediation.

Using hypothetical data, a step-by-step demonstration was provided of the construction of a bootstrap sampling distribution for a correlation coefficient. Then, tests of mediation and moderated mediation were demonstrated using data from a clinical trial of an intervention for caregivers of patients with Parkinson's disease or Alzheimer's disease. In a model hypothesizing that mutuality between caregiver and care recipient mediates the effect of objective on subjective levels of caregiver burden, the bootstrap sampling distribution was calculated of the mediation effect and, from that, two types of 95% confidence intervals for it. Then the hypothesis was tested that the mediating effect of mutuality was stronger for caregivers of patients with Parkinson's disease than for caregivers of patients with Alzheimer's disease.

Statistical hypothesis testing should never dictate all conclusions. However, the statistical advances in mediation analysis described here will facilitate nursing research as both nurse scientists and methodologists understand their assumptions and logic.