Representations of risk and ambiguity

Projects

Implications of imprecision in models for environmental decision analysis
Funding source: Swiss National Science Foundation

Expert elicitation of the deep uncertainty surrounding the market and non-market damages of climate change
Duration and funding source: 1 year (October 2011 – September 2012.); U.S. Environmental Protection Agency


Implications of imprecision in models for environmental decision analysis

Funding source: Swiss National Science Foundation

Personnel: M. Borsuk (Dartmouth), P. Reichert, H.R. Kuensch, S.L. Rinderknecht (ETH-EAWAG)

Synopsis: Bayesian statistical inference offers a mathematical framework to describe a learning process by combining prior knowledge with new data. In this framework, prior knowledge is typically formulated with a precise probability distribution to describe the subjective belief of either an individual expert or the joint belief of several experts about the value of a specified variable or model parameter. In practice such belief statements are often ambiguous. This is particularly the case if intersubjective belief is being expressed that is intended to represent the current state of knowledge of the scientific community.

One way to take this ambiguity into account is to replace a single prior probability distribution by a set of distributions that spans the range of appropriate distributions. Many specifications of such sets of probability distributions over continuous variables, so-called classes of distributions or imprecise probabilities, have been proposed. Despite this theoretical development, the concept of imprecise probabilities, which leads to a robustification of probability statements, is still very rarely applied. A reason for this may be that it is felt to be too difficult to implement. Difficulties could occur during elicitation, when updating priors with data, or when propagating imprecise distributions through models. To overcome the first of these potential difficulties, we developed an elicitation technique for the Density Ratio Class, which we believe to be the most satisfying class of probability distributions from a conceptual point of view (Rinderknecht et al. 2011a). This technique was then applied to several case studies, to demonstrate that a wide range of ambiguity can occur in practical applications (Rinderknecht et al. 2011b).  Finally, in (Rinderknecht et al. 2012), we address the remaining potential obstacles by showing how Bayesian inference, marginalization, and model propagation with Density Ratio Class priors works and how it can be easily implemented numerically.


Expert elicitation of the deep uncertainty surrounding the market and non-market damages of climate change

Duration and funding source: 1 year (October 2011 – September 2012.); U.S. Environmental Protection Agency

Personnel: M. Gerst (PI), M. Borsuk, and R. Howarth (Dartmouth)

Synopsis and progress:  Society’s reaction to climate change is ultimately an exercise in assessing exposure to poorly-defined, long-term risk.  Therefore, consideration of uncertainty should be central, and not secondary, in cost-benefit analysis.  Making uncertainty central to cost-benefit analysis, however, is hindered by current methods of representing climate damages; namely, representation of uncertainty, if considered at all, is focused on weak (first-order) instead of strong (second-order, Knightian, or pure) uncertainty.  Building on advances in the climate change damages and expert elicitation literature, we have designed protocol to assess expert knowledge on damages of climate change in a manner that is consistent with representing strong uncertainty.  The protocol explicitly assesses expert knowledge on adaptation, damages due to the rate and level of climate change, and willingness-to-pay to prevent non-market damages, as well as differences in the aforementioned parameters between the developed and developing world.  The expert elicitation process is currently in progress.