Statistical Programs |
College of Agriculture | University of Idaho |
Seminar Announcement |
"Applied Statistics in Agriculture" |
Uninformative Priors for Bayes' Theorem
Presented By |
Dr. Harold J. Price |
Sandia National Laboratories (Retired) |
Tuesday, Nov. 14 3:30 P. M. Room 62 College of Agriculture |
Perhaps the major objection to the use of Bayes' Theorem is the choice of prior
distribution. An important case arises when there is no prior information available;
this situation calls for an uninformative prior. LaPlace, who made excellent use of Bayes'
Theorem to obtain significant results, postulated that the uninformative prior should be constant.
He later came to question the use of this prior. One of the
objections
to the constant prior is that it is not independent of variable transformations. Later, Jeffreys, using heuristic arguments, postulated that a better uninformative prior was the 1/sigma function. It has the desirable feature of being independent of integer-power variable transformations. It also has the unfortunate property of extreme weighting for low values of sigma. In this talk, I postulate an uninformative prior that is independent of virtually any variable transformation, and yet, does not give preferential weight to any value of the variable. A simple example will illustrate the difference between the new and 1/sigma priors. Paper (pdf file) Overheads (pdf file) Figure 1 (jpg file) |
Return to Seminar Listing. |