Correcting the Bias Correction for the Bootstrap Confidence Interval in Mediation Analysis

• DOI: 10.31234/osf.io/pe4m2
• Publication Date: 2021-11-09T18:00:42Z
• Authors: Tristan Tibbe; Amanda Kay Montoya
• Keywords: Nominal level

The bias-corrected bootstrap confidence interval (BCBCI) was once the method of choice for conducting inference on indirect effect in mediation analysis due to its high power small samples, but now it is criticized by methodologists inflated type I error rates. In place, percentile (PBCI), which does not adjust bias, currently recommended inferential effects. This study proposes two alternative methods creating intervals around effect. Using a Monte Carlo simulation, these were compared BCBCI, PBCI, and introduced Chen Fritz (2021). results showed that perform continuum, where BCBCI has best balance (i.e., having closest an equal proportion CIs falling above below true effect), highest power, rate; PBCI worst balance, lowest fall between all three performance criteria. An extension original simulation after controlling rate inflation suggests increased might only be their higher Thus, if control over desired, still use with Future research should examine presence missing data, confounding variables, other real-world complications enhance generalizability results.