Generalized Linear Factor Score Regression: A Comparison of Four Methods.

Area of Science:

• Statistics
• Psychometrics
• Econometrics

Background:

• Factor score regression (FSR) is an emerging alternative to structural equation modeling (SEM).
• Existing literature primarily focuses on FSR with normally distributed outcomes, leaving guidance limited for other distributions.
• Generalized linear factor score regression (GLFSR) extends FSR to non-normal outcomes, but the performance of different factor scoring methods within GLFSR is not well understood.

Purpose of the Study:

• To compare the performance of different factor scoring methods when estimating regression coefficients in generalized linear factor score regression (GLFSR).
• To evaluate how scoring method choice impacts research conclusions under various non-normal outcome distributions.

Main Methods:

• A simulation study was conducted to assess four factor scoring methods: regression method, correlation-preserving method, and two sum score methods.
• These methods were evaluated within the context of ordinary, logistic, and Poisson factor score regression models.
• Performance was assessed based on coefficient bias, standard error bias, accuracy, and empirical Type I error rates.

Main Results:

• Factor scoring method performance varied significantly across the evaluated GLFSR models.
• The regression method demonstrated superior performance in terms of coefficient and standard error bias, accuracy, and Type I error rates.
• Both the regression method and the correlation-preserving method generally outperformed the sum score methods.

Conclusions:

• The choice of factor scoring method has a substantial impact on research findings in GLFSR.
• The regression method is recommended for its robust performance across different non-normal outcome distributions in GLFSR.
• Researchers should carefully consider the scoring method when applying FSR to non-normally distributed data.