Analysis of multivariate mixed longitudinal data: a flexible latent process approach

Area of Science:

• Psychology
• Statistics
• Longitudinal Data Analysis

Background:

• Multivariate longitudinal data of varying types (binary, ordinal, quantitative) are common in psychology.
• Analyzing change over time in the underlying latent construct requires sophisticated statistical approaches.

Purpose of the Study:

• To propose a unified approach for modeling the latent process underlying multiple longitudinal outcomes of different types.
• To handle individually varying and outcome-specific measurement times in longitudinal studies.

Main Methods:

• Utilizing random-effect models, including linear mixed models for latent trajectory.
• Employing outcome-specific threshold models for discrete data and flexible non-linear transformations for quantitative data.
• Developing likelihood and information criteria for discrete data to compare continuous versus discrete distribution models.

Main Results:

• The proposed approach effectively models the latent process trajectory across diverse data types.
• Parameterized non-linear transformations demonstrated utility in modeling sum scores.
• Longitudinal item response models were applied to describe latent construct change over time.

Conclusions:

• The developed method provides a flexible framework for analyzing complex longitudinal psychological data.
• It enhances the understanding of latent construct dynamics by integrating multiple outcome types and measurement schedules.
• The approach offers improved model comparison tools for discrete longitudinal data.