Fiducial Nonparametrics and linear regression

Abstract: 

R.A.Fisher's theory of fiducial probability -- a way of finding non-Bayesian posteriors for parameters -- has little or no acceptance among current statisticians, but separately there is a fiducial representation of an unknown cdf F, with a simple proof. Nonparametric models often contain an unknown error distribution, so there are numerous further applications, called Fiducial Nonparametrics. Among such methods, some have interesting theory but are not competitive with traditional methods. Others, however, are already in useful form, such as the fiducial distributions for parameters in linear models, with no assumptions being made about the form of the common error distribution. While easy to understand in principle, they can have a complex combinatorial nature. This suggests the use of resampling schemes to implement the methods and to organise the results.

Bruce Brown was an undergraduate at the University of Melbourne and obtained a Ph D at Purdue University, Indiana, USA in 1968.   Since then he worked until retirement at a number of universities in Australia and overseas.  His primary loves within Statistics have been consulting work and Nonparametrics.

The seminar will be followed by drinks and finger food with the speaker in the staff room (RC-3082). All attendees are welcome!