James Dickey

Traditionally, statistical theory has been concerned with the problem of choosing a function of experimental data to use as an automatic decision rule or formal inference procedure. Such a function would usually be evaluated by its performance in an imagined long sequence of independent repetitions of an experiment. Probability typically enters the analysis as a description of the imagined long-run behavior. I am interested in probability, more broadly, as a language for expressing expert opinion--for example, articulating uncertainty about the safety of a nuclear power plant design; about the future of economic series such as commodity prices or a consumer price index; or about the longevity of road surfaces. Expert opinion probability models range in structure from jointly consistent or "coherent" probabilities for a finite number of events to a multivariate predictive density represented as an average of a multiple regression sample density. The problems I am working on include: (i) using computers to elicit expert probabilities, with continual feedback of model implications for coherence of elicited values or model rejection; (ii) foundational issues regarding formal inference methods; (iii) uses of expert probability distributions in statistical data analyses; (iv) predictive use of models constructed from expert opinions; (v) questions of computational feasibility, algorithms, computing capacity and efficiency, and parallel processing and supercomputing. My research in these and other areas has been supported in the U.S. by the National Institutes of Health and by the National Science Foundation.