Agreement Between Two Quantitative Measurement Methods When the Underlying Latent Trait Is Not Constant

Area of Science:

• Biostatistics
• Psychometrics
• Measurement Theory

Background:

• Standard statistical methods for agreement assessment often assume a constant latent trait.
• This assumption is problematic when the 'individual' is a person, whose trait may change over time.

Purpose of the Study:

• To extend the general measurement error model to accommodate a time-varying individual latent trait.
• To evaluate statistical methods for estimating bias in measurement agreement when individual traits change.

Main Methods:

• Investigated four settings: constant trait, variable trait without trend, linear trend, and approximate linear trend.
• Assessed two methods: Generalized Least Squares (GLS) estimator (Sprent) and the two-stage method (Taffé).

Main Results:

• The two-stage method generally outperformed the GLS estimator in estimating bias.
• The two-stage method is applicable even with a single measurement per individual, unlike the GLS method.

Conclusions:

• The two-stage method provides a more robust approach for assessing measurement agreement in human subjects.
• This extended model and method are crucial for accurate bias estimation in longitudinal studies involving people.