Modeling individualized coefficient alpha to measure quality of test score data.

Area of Science:

• Psychometrics
• Statistical modeling
• Health literacy research

Background:

• Traditional reliability measures like Cronbach's alpha assume homogeneity.
• Heterogeneity in subjects and items can bias reliability estimates.
• Need for a subject- and item-specific reliability measure.

Purpose of the Study:

• To define and estimate individualized coefficient alpha.
• To measure individualized internal consistency of responses.
• To develop a robust statistical framework for reliability analysis.

Main Methods:

• Development of a regression model using three sets of generalized estimating equations.
• Modeling expectation and variance of responses.
• Estimation of individualized coefficient alpha using a novel third set of equations.
• Extension of methods to handle missing data.

Main Results:

• The proposed method effectively estimates individualized coefficient alpha.
• The statistical framework allows for valid inference on reliability.
• Simulation studies and real-world data analysis demonstrate method performance.
• Application to health literacy data from China.

Conclusions:

• Individualized coefficient alpha offers a more accurate reliability assessment in heterogeneous data.
• The generalized estimating equation approach provides a flexible and robust methodology.
• This approach enhances the quality of psychometric analysis in various research fields.