Generalized Network Psychometrics: Combining Network and Latent Variable Models.

Area of Science:

• Psychometrics
• Network Science
• Statistical Modeling

Background:

• Traditional psychometric models explain item covariance via latent variables.
• Network models conceptualize covariance as direct pairwise interactions between observable variables.

Purpose of the Study:

• To introduce network models as a formal psychometric framework.
• To generalize network models to encompass latent variable structures within structural equation modeling (SEM).
• To present latent network modeling (LNM) and residual network modeling (RNM) as extensions.

Main Methods:

• Developed two generalizations of the network model: LNM and RNM.
• Integrated these models into a general SEM framework.
• Implemented the methodology in the free software package lvnet with exploratory search algorithms.

Main Results:

• LNM allows for explorative identification of conditional independence relationships between latent variables.
• RNM enables identifiable models where local independence is violated.
• Simulation studies confirmed the adequacy of search algorithms in identifying network structures.
• Empirical application demonstrated utility on a personality inventory dataset.

Conclusions:

• Network modeling provides a flexible and powerful framework for psychometric analysis.
• LNM and RNM extend SEM capabilities, allowing for nuanced modeling of complex variable relationships.
• The lvnet software facilitates the application of these advanced network psychometric methods.