Abstract:
|
The methods and tools of Uncertainty Quantification (UQ) are used in defense and national security applications for model-supported inference, producing statistically defensible posterior distributions on quantities of interest. A qualitative confidence in the analysis and/or conclusions is often appropriate and increasingly a requirement. The interpretation of a qualitative "high, medium, low" judgment can be unsatisfactory both to the domain expert assigning the qualitative information, concerned about how this will be interpreted, and to the downstream consumer of the annotation, who may not be able to evaluate the issue behind the annotation. This poster will support discussion on a proposal for how qualitative information can be understood by both parties, by considering how it would be integrated if it did represent quantitative probability. There are distinct categories of expressions that may be made about the analysis relating to expected bias or uncertainty. We also propose that qualitative results can be best communicated in a systematic framework of graphical inference.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.