Resampling and Distribution of the Product Methods for Testing Indirect Effects in Complex Models

Area of Science:

• Psychometrics
• Statistical modeling

Background:

• Traditional mediation testing relies on the z test, which has limitations.
• Previous research focused on simple mediation models with single mediators.

Purpose of the Study:

• Evaluate advanced mediation testing methods in complex path models.
• Compare resampling methods and distribution-based tests against the traditional z test.
• Assess methods for testing contrasts of effects.

Main Methods:

• A simulation study was conducted with a path model including 1 independent variable, 3 mediators, and 2 outcomes.
• Varied factors included sample size, number of paths (2- and 3-path), statistical test, effect sizes, and contrast values.
• Confidence intervals were used to assess power, Type I error rate, coverage, and bias.

Main Results:

• The bias-corrected bootstrap demonstrated the least biased confidence intervals and highest power.
• This method also showed the most accurate Type I error rate.
• All tested methods showed reduced power and increased Type I error for 3-path effects compared to 2-path effects.

Conclusions:

• The bias-corrected bootstrap is recommended for mediation analysis due to its superior performance in complex models.
• Resampling approaches offer greater power and flexibility for testing contrasts.
• Confidence intervals for mediated effects remain biased, consistent with prior research.