Default and Inference Options: Use in Truncated and General Risk Analyses. James D. Wilson, Resources for the Future, 1616 P Street N.W., Washington, DC 20036
How "default options" should be used in health risk analysis is exceptionally contentious within the risk analysis profession. Some argue that these should be "hard": set by policy, generally biased to be "health protective" and requiring a substantial body of evidence to replace by decision-specific alternatives. Others argue that they should be science-based, identified by agreement within the professional community, replaced by whatever decision-specific information may be available to the analyst. This paper identifies two different roles played by "inference" or "default options" and suggests that these two terms be adopted as descriptive of their different characteristics and uses. Inference options constitute an integral part of human health risk analysis, providing practitioners with consensus theories, models, or parameters that can be used to bridge knowledge gaps in specific analyses. Because human health risk analysis is both considered "scientific" and employs scientific reasoning, inference options must be treated as priors in an empirical-Bayesian inference process. Decision- or case-specific information modifies each prior according to the reliability of this information, with conflicts resolved by a scientific, weight-of-evidence process. Inference options are science-based "best estimates" and evolve through consensus within the professional community. Default options can be used as integral components of decision systems used by organizations to simplify and control frequent but nontrivial decisions that turn on judgments made by experts. We call these "recurrent" decisions. Authority to make these recurrent decisions is often delegated (sometimes tacitly) to staff members who have the requisite expertise. Yet delegation of decision authority is always conditioned by various forms of controls, including limits on the authority. We suggest that conditional delegation occurs because delegation always carries risks for organizations: those to whom authority is delegated may not make the decisions thought appropriate by senior officials, thus potentially harming them or the organization. We postulate that policies which define what models, theories, etc., and the input data to be used serve as means to control delegation risk (as well as simplifying decisions). These default options need to be tailored to the policy ends served, which generally means that some or all will be biased. They must be "hard," with "departure from" them requiring a high standard of evidence and also assurance that choosing an alternative will still satisfy the policy ends of the decision process in which these are a part. Default options need to be developed in the same way as any other policy-implementation practices, including deliberations that engage those who will be affected by the decisions.