Statistical properties of four effect-size measures for mediation models

Area of Science:

• Psychometrics
• Statistical modeling
• Quantitative psychology

Background:

• Assessing mediation effects is crucial in statistical analysis.
• Existing effect size measures for mediation may suffer from bias and inefficiency.
• Understanding the performance of different estimators is vital for accurate interpretation.

Purpose of the Study:

• To compare classical and Bayesian estimators for four mediation effect size measures.
• To evaluate estimator performance in single-mediator and two-mediator models.
• To identify the most reliable effect size measures for quantifying indirect effects.

Main Methods:

• Examined classical and Bayesian estimators for four effect size measures.
• Analyzed performance in single-mediator and two-mediator models.
• Assessed estimators based on bias, efficiency, power, Type I error, coverage, imbalance, and interval width.

Main Results:

• Standardized mediation effect sizes were relatively unbiased and efficient across models.
• Bootstrap interval estimates for standardized measures outperformed others in the single-mediator model.
• Bayesian summaries reduced relative bias for standardized effect sizes under specific conditions.

Conclusions:

• Standardized effect-size measures are recommended for quantifying mediated effects.
• Both classical and Bayesian approaches support the superiority of standardized measures.
• These findings enhance the accuracy and reliability of mediation analysis.