Signal-to-Noise Ratio in Estimating and Testing the Mediation Effect: Structural Equation Modeling versus Path...

Area of Science:

• Social and behavioral sciences
• Statistical modeling

Background:

• Mediation analysis is crucial for understanding causal processes.
• Previous comparisons of SEM and PAWC for mediation analysis may lack validity due to issues with latent variable scaling.
• Measurement errors can bias parameter estimates in composite scores.

Purpose of the Study:

• To compare the accuracy and precision of structural equation modeling (SEM) and path analysis with weighted composites (PAWC) in mediation analysis.
• To introduce signal-to-noise ratio (SNR) as a metric independent of latent variable scales for parameter estimation.
• To evaluate the statistical efficiency and power of PAWC versus SEM.

Main Methods:

• Comparison of SEM and PAWC using signal-to-noise ratio (SNR) for parameter estimates.
• Analysis of indirect effect estimation in mediation analysis.
• Investigation of conditions affecting SEM's SNR and PAWC's performance.

Main Results:

• PAWC consistently yields greater SNRs than SEM in mediation analysis, even with measurement errors.
• Path analysis via factor scores demonstrates significantly higher SNRs than SEM.
• Mediation analysis with equally weighted composites (EWCs) also shows higher SNRs compared to SEM.
• PAWC's advantage is more pronounced with strong predictor-mediator relationships, but prediction error in the mediator can negatively impact PAWC.

Conclusions:

• PAWC is statistically more efficient and powerful than SEM for mediation analysis in empirical research.
• The findings challenge previous conclusions comparing SEM and PAWC, highlighting the utility of SNR.
• Understanding the influence of predictor-mediator relationships and prediction error is key for optimal methodology selection.