A robust alternative estimator for small to moderate sample SEM: Bias-corrected factor score path analysis.

Area of Science:

• Psychology
• Statistics
• Addiction Research

Background:

• Structural equation modeling (SEM) with full information maximum likelihood (FIML) is standard for addiction research but suffers from bias in small samples.
• Limited information estimators, like bias-corrected factor score path analysis, offer unbiased estimates in small to moderate sample sizes.
• Underutilization of limited information methods stems from unfamiliarity, lack of practical guidance, and insufficient comparative studies.

Purpose of the Study:

• To delineate a bias-corrected limited information estimator for SEM in addiction research.
• To provide practical guidance and R code for applying this method.
• To compare the performance of limited versus full information estimators in small to moderate samples.

Main Methods:

• Step-by-step analysis of a sequential mediation case study on internet addiction.
• Implementation using R code with the lavaan package.
• A simulation study to assess estimator divergence and performance.

Main Results:

• The limited information estimator demonstrated superior performance over FIML in small to moderate sample sizes.
• Key advantages included reduced bias, increased efficiency, and enhanced statistical power.
• Differences between estimators were consistent, highlighting the practical implications for addiction research.

Conclusions:

• The bias-corrected limited information estimator is a valuable alternative to FIML for addiction researchers with small to moderate sample sizes.
• This method provides more accurate and reliable results, addressing a critical limitation in current practice.
• Accessible guidance and software facilitate the adoption of this improved analytical approach.