Ordinal state-trait regression for intensive longitudinal data.

Area of Science:

• Psychological Science
• Quantitative Psychology
• Behavioral Research Methods

Background:

• Experience sampling studies repeatedly measure mental states, enabling analysis of between- and within-person effects.
• Traditional regression models are limited to continuous variables, while mental states are often ordinal.
• Existing methods struggle to accurately disaggregate effects for ordinal mental state data.

Purpose of the Study:

• To develop a novel statistical model for disaggregating between- and within-person effects between two ordinal variables.
• To address the limitations of current regression models when applied to ordinal mental state data.
• To provide a robust method for analyzing complex psychological associations in repeated-measures designs.

Main Methods:

• Developed a regression model for ordinal variables, positing a continuous latent response.
• Incorporated a novel penalty to balance linear and nonlinear model fits on the latent scale.
• Validated the method through a simulation study comparing its performance to linear and nonlinear models.

Main Results:

• The proposed penalization approach effectively balances model complexity, avoiding both over-restriction and over-fitting.
• Simulation results demonstrate the method's accuracy in disaggregating between- and within-person effects for ordinal data.
• The model successfully handles nonlinear relationships on the latent scale while maintaining interpretability.

Conclusions:

• The new ordinal regression model provides a powerful tool for psychological researchers using experience sampling or similar designs.
• This method enhances the ability to distinguish stable trait-like influences from transient state-like fluctuations in mental states.
• The approach offers a more nuanced understanding of psychological associations by accommodating the ordinal nature of mental state measures.