Combining Diverse Expert Opinions in Risk Analysis Using Relative Causal Knowledge.

Area of Science:

• Causal inference
• Risk analysis
• Decision theory

Background:

• Expert disagreement is prevalent in risk analysis due to varied data, causal assumptions, and abstraction levels.
• Traditional consensus methods can obscure critical differences in expert understanding.
• There is a need for principled methods to manage expert disagreements in scientific and regulatory contexts.

Purpose of the Study:

• To introduce a novel knowledge-based framework, Relative Causal Knowledge for Expert Resolution (RCKER), for managing expert disagreements.
• To provide tools for translating, comparing, diagnosing, and reconciling differences in expert causal models.
• To evaluate the reconcilability of expert models without assuming consensus is always possible.

Main Methods:

• Utilizing causal inference and category theory to analyze expert causal models.
• Treating each expert model as a partial perspective.
• Identifying structural and interventional consistency across models.
• Evaluating reconcilability at different levels of abstraction.

Main Results:

• RCKER demonstrated conditional compatibility between expert models on PM2.5 health effects after abstraction and confounder alignment.
• RCKER identified irreconcilable structural and interventional differences between models on formaldehyde and leukemia.
• The framework successfully managed expert disagreement without forcing a misleading consensus.

Conclusions:

• RCKER offers a principled approach to understanding and managing expert disagreement in risk analysis.
• The framework preserves substantively distinct expert views, avoiding the pitfalls of forced consensus.
• Findings have implications for risk communication, regulatory policy, and AI-assisted review tools.