Heterogeneous and Uncertain Knowledge: Beyond Probabilities

Abstract

Individual and group judgments about heterogeneous causal evidence are essential to modeling public health outcomes. Those judgments are often made using results from exposure-response models or other damage functions. Preferably, those functions should be mechanistic. In public health, for example, they represent fundamental pathophysiological and other processes such as carcinogenesis, the spread of infections, and mass casualties that may occur after the catastrophic incident has occurred. Alternatively, the models are statistical and consist of causal associations, rather than attempt to represent mechanisms. In all cases, heterogeneous and conflicting scientific evidence generates alternative expert opinions that can affect regulatory law decisions. Scientific agreements and disagreements about causation made on behalf of the public and for or against a public choice may have to meet legal challenges. Scientifically, those opinions must be objective, formal, and replicable. As a first step toward meeting these three qualifiers, we discuss and exemplify fusing heterogeneous and uncertain information using probabilistic Monte Carlo simulations (i.e., human exposure to waterborne inorganic arsenic and cancer). We also discuss fuzzy integrals and Dempster-Shafer combination rules. These represent uncertainty using fuzzy or possibilistic measures, rather than probabilistically, and often more closely reflect the meaning of policy and legal phrases and terms-of-art. We exemplify fusing uncertain knowledge and information using: (i) system-biology probabilistic model, and (ii) a fuzzy causal network. Aggregated scientific evidence from multiple experts about causal models used in support of public decisions about setting precautionary or other risk averse regulatory standards is critical. The reason is twofold. First, different opinions are essential to scientific debates and, second, public choices are fully informed. Conflicts can be resolved through panels convened by learned societies or by scientific advisory groups to national and international agencies to justify the choice of a causal model for regulatory law. Rather than relying on consensus on the whole of the evidence, we parse the scientific evidence into assumption or premises, rules or models, and results. We then discuss some essential methods and criteria that can result in majority or other agreements that avoid paradoxical results. We extend these discussions in Chap. 8, Aggregating Individual Judgments in Precautionary Decision-Making.