Latent variable GIMME using model implied instrumental variables (MIIVs).

Area of Science:

• Psychology
• Psychometrics
• Computational Statistics

Background:

• Psychological research increasingly requires personalized and generalizable dynamic models.
• Measurement error is common, necessitating the use of multiple indicators for latent constructs.
• Existing data-driven model searches often exclude latent variables.

Purpose of the Study:

• To introduce a novel approach, latent variable GIMME (LV-GIMME), for data-driven model searches that incorporates latent variables.
• To extend the capabilities of group iterative multiple model estimation (GIMME) by integrating latent variable analysis.
• To allow for dynamic models where latent variable relationships can vary across individuals.

Main Methods:

• LV-GIMME builds upon GIMME, the idiographic filter, and model implied instrumental variables with two-stage least squares estimation (MIIV-2SLS).
• LV-GIMME performs data-driven searches for relationships among latent variables.
• MIIV-2SLS is used for estimation, relaxing the assumption of constant latent variable models across individuals.

Main Results:

• Simulated data studies confirmed LV-GIMME's ability to reliably detect relationships among latent constructs.
• Latent constructs demonstrated greater power in detecting effects compared to using observed variables directly.
• Empirical data from functional MRI and daily self-report studies were used as examples.

Conclusions:

• LV-GIMME offers a powerful new tool for psychological researchers seeking to model dynamic individual processes.
• The inclusion of latent variables enhances the precision and power of data-driven model searches.
• This approach accommodates individual differences in the structure of latent variable relationships.