Group for Research in Decision Analysis


Minimax Estimation of a Constrained Binomial Proportion


The problem of estimating a binomial proportion constrained to lie in an interval of the form [a,b] "not equal to" [0,1] is considered. The minimax and linear minimax estimators are obtained for both quadratic and information-normalized loss functions. For parameter spaces of the form [a, 1 - a] and [0,b], the minimax estimators are used as benchmark to compare the performance of estimators belonging to other classes such as linear minimax estimators, the MLE, and Bayes estimators associated with translated Beta priors.

, 37 pages

This cahier was revised in October 2001