Re: ?Value of information without utilities?

Russell Almond (ralmond@ets.org)
Mon, 24 May 1999 13:29:12 -0400 (EDT)

>The other obvious answer is entropy, and its relative, the cross-entropy
>between two random variables. The introductory probability text by Ross
>(Prentice-Hall 1998) has a nice discussion of entropy and Shannon
>information. My favorite discussion of entropy and cross-entropy is in
>Bernardo and Smith, *Bayesian Theory* (Wiley 1994). However, it turns out
>these are value-related.

I find it hard to envision a notion of value of information that is not
somehow value related. I'd be interested in anyone who thinks he/she can.

Kathy Laskey

I.J. Good used the notion of a Quasi-Utility to describe some of these
information theoretic measures. David Madigan put a review of some of
that work into:

\refrence {Madigan, D. and R.G. Almond [1996]} ``Test
Selection Strategies for Belief Networks'' In D. Fisher and H-J Lenz
(eds.) {\it Learning from Data: AI and Statistics IV\/} Springer
Verlag, New York, 89--98.

(I think that it might also still be available as a Tech Report from
http://www.stat.washington.edu/
)
--Russell