Janis Hardwick

University of Michigan

Connie Page

Michigan State University

Quentin F. Stout

University of Michigan

Information maximization considerations, and analysis of the asymptotic
mean square error of several estimators, leads to the following
adaptive procedure: use the maximum likelihood estimator to estimate *p*,
and if this estimator is below (above) the cut-point *a_r*, then observe an
individual (product) trial at the next stage.
An exact analysis of this adaptive procedure for fixed sample sizes shows that
it behaves roughly as the asymptotics predict, and that several other
estimators and procedures behave far worse than their asymptotics indicate.
Further, the adaptive procedure exhibits negative regret over a portion of
the parameter range.

**Keywords:** batch testing, risk assessment, infection rate,
grouped data, omniscient allocation, composite testing

Copyright © 1997, 1996. | Last modified: 5 Mar 1997 |