When to Use Different Tests for Power Analysis and Data Analysis for Mediation
Several options exist for conducting inference on indirect effects in mediation analysis. While methods which use bootstrapping are the preferred inferential approach for testing mediation, they are time consuming when the test must be performed many times for a power analysis. Alternatives which are more computationally efficient are not as robust, meaning accuracy of the inferences from these methods are more affected by nonnormal and heteroskedastic data (Biesanz et al., 2010). While previous research focused on how different sample sizes would be needed to achieve the same amount of power for different inferential approaches (Fritz & MacKinnon, 2007), we explore how similar power estimates are at the same sample size. We compare the power estimates from six tests using a Monte Carlo simulation study, varying the path coefficients and tests of the indirect effect. If tests produce similar power estimates, the more computationally efficient test could be used for power analysis and the more intensive test involving resampling can be used for data analysis. We found that when the assumptions of linear regression are met, three tests consistently perform similarly: the joint significance test, the Monte Carlo confidence interval, and the percentile bootstrap confidence interval. Based on these results, we recommend using the more computationally efficient joint significance test for power analysis then using the percentile bootstrap confidence interval for the data analysis.