Assessing the interchangeability of linked scores in multivariable statistical analyses.

Area of Science:

• Psychometrics
• Statistical Modeling

Background:

• Classical test theory is often used to assess measure generalizability.
• Existing literature commonly compares established subpopulations to evaluate data linkage.
• Multivariable analyses require a nuanced understanding of data linkage generalizability.

Purpose of the Study:

• To examine data linkage generalizability for multivariable analyses using classical test theory.
• To introduce a structural equation modeling (SEM) based statistical method for evaluating data linkage suitability.
• To assess linkage appropriateness beyond simple correlation, considering external variables.

Main Methods:

• Development of an SEM-based statistical methodology.
• Application of the method to PROMIS® Parent Proxy and Early Childhood Global Health measures.
• Evaluation of linkage suitability for continuous and categorical external variables.

Main Results:

• A high correlation (r = .829) between measures suggested general suitability.
• Detailed analysis revealed significant differences in content and measurement structure.
• These differences can compromise data interchangeability in specific use cases.

Conclusions:

• Statistical quality of a linkage is insufficient on its own.
• Users must evaluate the appropriateness of a linkage for specific research questions and multivariable analyses.
• Consideration of content and structure is crucial for reliable data linkage.