A modified approach to fitting relative importance networks

Area of Science:

• Psychometrics
• Network Analysis
• Statistical Modeling

Background:

• Relative importance networks are commonly estimated using general dominance for multiple regression.
• This method offers desirable properties like R² contributions and replicability.
• Existing methods have limitations that can be addressed through improved predictor selection.

Purpose of the Study:

• To introduce a modified approach for estimating relative importance networks using best-subsets regression.
• To enhance the utility, specificity, and computational efficiency of network psychometrics.
• To explore the potential of best-subsets regression for Gaussian graphical models and Ising models.

Main Methods:

• A modified approach employing best-subsets regression prior to network estimation.
• Selection of optimal predictor subsets for each item in the network.
• Evaluation and demonstration of the proposed method's performance and benefits.

Main Results:

• The modified approach offers significant computation time savings for larger networks.
• Principled edge selection enhances network specificity.
• The method allows for signed networks and potential generalization to other statistical models.

Conclusions:

• The enhanced relative importance network approach provides a valuable advancement in network psychometrics.
• Best-subsets regression offers a principled and efficient method for predictor selection.
• This modification expands the applicability of network analysis in statistical research.