Commentary: Are Three Waves of Data Sufficient for Assessing Mediation?

Area of Science:

• Statistics
• Psychometrics
• Social Sciences

Background:

• Structural equation models (SEMs) are widely used for analyzing complex relationships between variables.
• Previous research (Maxwell, Cole, & Mitchell, 2011) identified biases in SEMs using cross-sectional data for mediated effects.
• The limitations of SEMs with longitudinal data for estimating mediated effects remain an area for investigation.

Purpose of the Study:

• To extend previous findings on biased mediated effect estimates in SEMs.
• To investigate the impact of simple autoregressive SEMs on effect estimation using longitudinal data.
• To determine if simple SEMs can erroneously imply direct effects when only indirect effects exist, and vice versa.

Main Methods:

• Analysis of simple structural equation models (SEMs).
• Utilizing longitudinal data to assess model performance.
• Examination of autoregressive SEMs to evaluate direct and indirect effect estimations.

Main Results:

• Simple SEMs, even with longitudinal data, can produce biased estimates of mediated effects.
• Autoregressive SEMs may incorrectly suggest the presence of indirect effects when only direct effects are present.
• Conversely, simple SEMs can also imply direct effects exist when the true relationship is indirect.

Conclusions:

• The application of simple SEMs with longitudinal data requires careful consideration due to potential biases.
• Researchers should be cautious about interpreting direct and indirect effects from simple autoregressive models.
• Findings underscore the need for more robust modeling techniques to accurately capture mediation in longitudinal studies.