Multiobjective Genetic Programming Can Improve the Explanatory Capabilities of Mechanism-Based Models of Social...

Area of Science:

• Social Science
• Complex Systems Modeling

Background:

• Generative models in social science are crucial for realist explanations but offer limited insights into alternative mechanisms.
• Existing models, while validated, represent only one set of potential entities and mechanisms.

Purpose of the Study:

• To propose a new model discovery framework that fully supports realist explanation in social science.
• To address the limitations of generative models by exploring a plurality of candidate structures.

Main Methods:

• Exploiting the ontology of existing generative models to propose new candidate structures.
• Utilizing genetic programming for automated search and a multi-objective approach for evaluating models.
• Applying the framework to a case study of US societal alcohol use patterns (1980-2010).

Main Results:

• The framework successfully identified three competing explanations for societal alcohol use patterns.
• Novel integrations of social role theory were discovered, not previously considered by the modeler.
• Demonstrated the framework's ability to uncover alternative causal mechanisms.

Conclusions:

• Model discovery enhances the explanatory utility of generative approaches in realist social science.
• The proposed framework provides a more comprehensive tool for understanding social phenomena.
• Encourages practitioners to adopt model discovery for improved scientific explanation.