using data for Bayesian prior

From: Geert Verdoolaege (geert.verdoolaege_at_ugent.be)
Date: 11/09/04 

Date: Tue, 09 Nov 2004 10:36:44 +0100

I am not sure why exactly, in Bayesian analysis, it is a bad idea to use
information on the measured data for the construction of a prior
distribution of the parameters.

For example, I used my data to derive some information on the
distribution of the parameters, information that I trust highly (e.g.
multimodal distribution, location of the modes). Can I not use this
information then? It would seem a waste to me not to use this additional
knowledge.

Thank you,
Geert