Model Specification Searches in Structural Equation Modeling Using Bee Swarm Optimization.

Area of Science:

• Psychological Measurement
• Computational Psychology
• Optimization Algorithms

Background:

• Metaheuristics are valuable for complex combinatorial problems in psychological research, particularly in scale construction and model specification.
• Existing methods may not efficiently explore the intricate structures of psychological measurement instruments.

Purpose of the Study:

• To introduce a novel Bee Swarm Optimization (BSO) algorithm for exploring the underlying structure of psychological measurement instruments.
• To simultaneously assign items to an unknown number of nested factors within a confirmatory bifactor model and select items for the final scale.

Main Methods:

• The BSO algorithm mimics bees' foraging behavior, with scout bees performing broad exploration (e.g., adding/removing factors) and onlooker bees conducting local exploitation (e.g., item assignment/swapping).
• This division of labor balances diversification (exploration) and intensification (exploitation) for robust model specification.
• The algorithm was tested on two empirical datasets: Holzinger-Swineford and the Short Dark Triad Questionnaire (SDQ3).

Main Results:

• The BSO algorithm successfully identified the underlying structure in both the Holzinger-Swineford and SDQ3 datasets.
• The study illustrated the impact of key hyperparameters, including colony size, scout-to-onlooker ratio, and the number of elite solutions.

Conclusions:

• The proposed BSO algorithm offers an effective approach for exploring complex structures in psychological measurement.
• It provides a valuable tool for scale construction and confirmatory bifactor analysis, balancing exploration and exploitation.
• Future research can explore further applications and refine the algorithm's parameters.