David Bickel, statistician

This paper considers a weighted version of the differential entropy of the posterior distribution of the probability of success conditional on the observed value of a binomial random variable. The uniform (0,1)prior distribution of the success probability is used to derive large-sample results.
The weighting function allows emphasizing some values of the parameter more than other values. For example, since the success probability value of 1/2 has special importance in many applications, that parameter value may be assigned a higher weight than the others. This differs from the more common Bayesian approach of assigning more prior probability to certain parameter values.
The author proves asymptotic properties not only of the weighted differential entropy but also of weighted differential versions of the Renyi, Tsallis, and Fisher definitions of entropy or information. The results are concrete in that they are specifically derived for the posterior distribution of the success probability given the uniform prior.

Reviewed by David R. Bickel