On the Difficulties Inherent in Fisher's Fiducial Argument
A. P. Dempster · Journal of the American Statistical Association · 1964
The terms predictive and postdictive are introduced for two basic and implicitly well-known modes of interpreting probabilities, and the roles of these two modes in statistical inference are discussed with emphasis on the confidence argument of Neyman and the fiducial argument of Fisher. The conflicts between the confidence approach and the fiducial approach are analyzed in a way which generally supports Fisher's position. Finally, difficulties internal to the fiducial approach are discussed. It is concluded that the general form of the fiducial argument is appealing but that many of the restrictions imposed by Fisher are awkward or ambiguous and ought, perhaps, to be replaced.