© 1990 by Biometrika Trust
Articles |
Data-translated likelihood and Jeifreys's rules
Department of Statistics, Carnegie-Mellon University Pittsburgh, Pennsylvania 15213, U.S.A.
Received for publication 1 October 1988.
Revision received 1 June 1989.
| Abstract |
|---|
According to the definition used by Box & Tiao (1973) for a likelihood to be data translated it must have location form in terms of a sufficient statistic. In contrast to Jeffreys's arguments for a uniform prior, theirs does not cover cases such as the Cauchy location family, and is in this sense stronger than the group-theoretic criterion of invariance. Their definition is easily modified to cover such cases through the introduction of an ancillary statistic, and their argument in favour of a uniform prior then becomes group-theoretic. Their concept of approximate data-translated likelihood may also be modified to produce a sharper local approximation.
Key Words: Ancillary statistic Information metric Invariant prior Reference prior