Combining Diverse Expert Opinions in Risk Analysis Using Relative Causal Knowledge

Disagreements among domain experts are common in risk analysis, especially when models differ in data sources, causal assumptions, or levels of abstraction. Conventional methods-such as averaging or pooling quantitative probabilities or forcing consensus-risk distorting important differences in causal understanding. We propose a knowledge-based framework for expert resolution: Relative Causal Knowledge for Expert Resolution (RCKER). Treating each expert's causal model as a potentially valid but partial perspective, RCKER uses tools from causal inference and category theory to translate, compare, diagnose, and, if possible, reconcile differences in expert beliefs. Rather than assuming a priori that harmonization is possible, RCKER identifies structural and interventional consistency across models and evaluates whether they can be reconciled at some level of abstraction. We illustrate the approach with two case studies. In the first, two experts disagree about the health effects of PM2.5, but their models are shown to be conditionally compatible after causal abstraction and confounder alignment. In the second, competing models of formaldehyde and leukemia are found to be structurally and interventionally irreconcilable. Together, these examples demonstrate how RCKER provides a principled framework for understanding and managing expert disagreement without collapsing substantively distinct views into misleading consensus. We discuss implications for risk communication, regulatory deliberation, and the design of expert panels and AI-assisted review tools.