Level-specific reliability coefficients from the perspective of latent state-trait theory

Area of Science:

• Psychological research methods
• Quantitative psychology
• Statistical modeling

Background:

• Ecological momentary assessment (EMA) is increasingly used in psychological research.
• Intensive longitudinal data (ILD) requires advanced statistical models.
• Multilevel latent state-trait (ML-LST) models, based on LST-R theory, are a potential approach for analyzing ILD.

Purpose of the Study:

• To clarify the interpretation and identification of Level 1 (within-subject) and Level 2 (between-subject) reliability coefficients within LST-R theory.
• To extend the understanding of reliability beyond traditional LST-R coefficients (reliability, consistency, occasion-specificity).
• To provide a theoretical framework for newly defined level-specific coefficients in ML-LST models.

Main Methods:

• The study discusses the interpretation of level-specific coefficients using multilevel versions of the Multistate-Singletrait (MSST), Multistate-Indicator-specific trait (MSIT), and Multistate-Singletrait model with M-1 correlated method factors (MSST-M-1).
• Focuses on the MSST-M-1 model to define between-subject coefficients as measures of indicator-unspecificity or scale unidimensionality.
• Highlights distinctions between occasion-specificity and within-subject reliability.

Main Results:

• In the MSST-M-1 model, the between-subject coefficient quantifies indicator-unspecificity (shared between-level variance) or scale unidimensionality.
• The study clarifies the theoretical meaning of within-subject reliability and occasion-specificity.
• The performance of the multilevel MSST-M-1 model is demonstrated with empirical data.

Conclusions:

• The interpretation of within-subject and between-subject reliability in ML-LST models is clarified, enhancing LST-R theory.
• The MSST-M-1 model provides a framework for understanding item properties and scale structure in ILD.
• These findings support the robust application of ML-LST models in psychological research using EMA data.