AGCI Session II: Characterizing and Communicating Scientific Uncertainty
Session Chairs: Dr. Richard H. Moss and Dr. Stephen H. Schneider
July 31 to August 8, 1996
Subjective Probability Rankings and the IPCC Process
Richard Moss
IPCC Technical Support Unit
U. S. Global Change Research Program
Washington, DC
The issues of representation of uncertainty, how to characterize outlier events in assessments, and how to combine expert judgments in very diverse fields, have been around for a long time. Moss' interest is focused on how we can do a better job next time than was done in the latest IPCC process. A lot of progress has already been made, including in the area of characterizing uncertainty, but much more remains to be accomplished. How to best rank confidence levels is still a difficult issue and so far the effort to deal with it has not been very systematic.
Moss presented a framework he and colleagues developed as one possible way of standardizing subjective probability rankings in the IPCC process. The draft framework consists of placing conclusions into one of four categories:
Well-established This category denotes wide agreement, based on multiple findings through multiple lines of investigation. A finding could be removed from this category not by a single hypothesis, observation or contention, but only by a plausible alternative hypothesis, based on empirical evidence or explicit theory, and accepted by a substantial group.
Well-posed controversy A well-established finding becomes a well-posed controversy when there are serious competing hypotheses, each with good evidence and a number of adherents.
Probable This category indicates that there is a consensus, but not one that has survived serious counter-attack by other views or serious efforts to "confirm" by independent evidence.
Speculative Speculative indicates not so much "controversy" as the accumulation of conceptually plausible ideas that haven't received serious attention or attracted either serious support or serious opposition.
As shown in Figure 2.12, there are two dimensions to this framework with regard to confidence: amount of evidence, and level of agreement. This framework was developed too late in the process to be used in the 1995 IPCC assessment just completed, so for this version, they fell back on using high, medium and low confidence levels, designated by one, two or three asterisks, which is a less precise framework.
How to best rank
confidence levels is still a difficult issue and so far the effort to
deal with it has not been very systematic.
A more systematic process and tools for ranking confidence levels are needed. It is both particularly important and difficult to do this in an environment that is becoming increasingly political. We must also pay attention to how uncertainties are characterized in the media and in the political process.
We must also pay
attention to how uncertainties are characterized in the media and in
the political process.