A forum sponsored by
Society for Risk Analysis,
U.S. Army Corps of Engineers’
Engineer Research and
Development Center, and
Electric Power Research Institute
Rosina Bierbaum, Director for the Environment, White House Office of Science and Technology Policy
Todd S. Bridges, U.S. Army Corps of Engineers Waterways Experiment Station
Ecological risk assessments are dominated by information collected at small spatial and temporal scales. The scales most commonly emphasized are those that correspond to levels of biological organization at or below the individual organism, even though these scales provide rather limited information about processes that control the dynamics of populations and communities. The attention we give to uncertainties regarding the exact form of the does-response relationship or the bioavailability of a specific contaminant reflects our focus on the processes operating at relatively small spatial and temporal scales. The reason commonly voiced for not giving adequate consideration to processes operating at larger, more ecologically relevant, spatial and temporal scales is that the uncertainties at such scales are large. This uncertainty comes from lack of information (e.g., about the factors controlling population dynamics for a given receptor) and natural variation evident at such scales (e.g., seasonal and natural disturbance induced fluctuations in abundance). By not giving adequate attention to the ecological processes operating at contaminated sites we are in danger of over-estimating the risk at sites where affected processes operating at small scales are overwhelmed by processes operating at the larger scales we currently ignore. However, risks may also be under-estimated at sites where subtle changes in small-scale processes are magnified at larger scales. Improving our ability to estimate and regulate ecological risks will require greater consideration of scale and uncertainty.
Vicki Bier, University of Wisconsin at Madison
Deborah Mayo, Virginia Polytechnic Institute and State University
The recognition that values (methodological, political, cultural, economic) may enter into every stage of risk management--even at the level of establishing evidence of risk--has often been the basis for denying the possibility of evaluating risk assessments objectively. Because of the public policy consequences that follow from risk assessments, the tasks of generating and interpreting risk data may be thought to introduce ethical and other value-laden considerations that may go beyond the accepted canons of "objective scientific reporting". Some have taken this to show that an ethical or responsible interpretation of evidence may warrant violating canons of scientific objectivity, and even that a scientist must choose between norms of morality and objectivity. The danger is that it may then become possible to declare "immoral" the objective reporting of scientific uncertainties in evidence. This conflicts with the generally accepted imperative for an ethical interpretation of scientific evidence. We need a much more careful understanding of the precise nature of the uncertainties in increasingly complex scientific inferences, especially those that form the basis for decisions about risky technologies.
Timothy Barry, U.S. Environmental Protection Agency
Kathryn Blackmond Laskey, George Mason University
A model is a representation of a system that can be used to answer questions about the system. More and more, public policy decisions are based on the predictions made by scientific models. Science commits to the search for common, agreed upon models enforced not by dogma and coercion, but by open public debate and tests of correspondence to empirical observation. Models serve the laudable purpose of encouraging society to base policy on sound science rather than on superstition or political power. But the Achilles heel of policy analysis in general and of formal policy modeling in particular is the "unknown unknown" -- when the world departs substantially from the assumptions of the model in ways that matter critically for policy making purposes. Model uncertainty is uncertainty about the degree to which a model is an adequate representation of the world for the problem at hand. This talk frames the issue of model uncertainty, contrasts model uncertainty with uncertainty in the world, describes some of the dangers of inadequate treatment of model uncertainty, and asks the question of how policy analysts can protect against these dangers.
Mark Colyvan, University of Tasmania
We discuss arguments that purport to prove that probability theory is the only sensible means of dealing with uncertainty. We show that these arguments can succeed only if some rather controversial assumptions about the nature of uncertainty are accepted. We discuss these assumptions and provide reasons for rejecting them. Finally, we discuss examples of what we take to be non-probabilistic uncertainty.
Teddy Seidenfeld, Carnegie Mellon University
Douglas Dixon, Electric Power Research Institute
Vicki Bier, University of Wisconsin at Madison
Adam Finkel, Director of Health Standards, Occupational Safety and Health Administration
Ronald Yager, Iona College
Richard Neapolitan, Northeastern Illinois University
I discuss the importance of not just propagating probabilities but propagating intervals in Bayesian networks. First I introduce the idea of a probability of a relative frequency and then give a simple example in the one variable world showing how this notion can be used in policy decision making. Next I describe qualitatively how intervals are propagated in Bayesian networks and show qualitatively how the intervals widen when variables are instantiated from below. Finally, I give an example showing the importance of considering intervals in risk analysis problems using Bayeisan networks.
Scott Ferson (organizer), Applied Biomathematics