Abstract

Mediation models are widely used in various fields such as psychology, education, and business administration. Some traditional methods like Baron and Kenny’s approach and multivariate delta method were used before, but now bootstrap methods have become a standard procedure. Since the bias-corrected bootstrap (BC bootstrap) was proposed to correct for the potential bias in the confidence intervals (CIs) obtained by the percentile bootstrap, the BC method has been frequently used domestically based on some old research that the BC bootstrap provides more accurate CIs and higher power. However, recent studies have reported that the BC method may not be appropriate for testing mediating effects because it produces inflated Type I error rates. The goal of the present study is to enhance researchers’ understanding about testing methods for the mediation effect by examining the recent research trend of testing methods for mediating effects and by comparing the performance of the percentile bootstrap and the BC bootstrap through a series of simulations under various conditions. In Study 1, the results show that the power was high in the order of BC bootstrap, percentile bootstrap, and multivariate delta method. As the number of indicators, the effect size, the factor loading, and the sample size increased, the power also increased. Study 2 demonstrates that Type 1 error rates were also high in the order of BC bootstrap, percentile bootstrap, and multivariate delta method, confirming that the BC bootstrap also generates the highest Type 1 error in structural equation mediation models. Finally, the limitations of this study and guidelines based on the simulation results are discussed.