Refining Poisson Confidence Intervals
George Casella, Christian P. Robert · 1988
April1986 Revised July 1987 Second Revision, March 1988 Third Revision, November 1988 A computational method is presented which, when applied to an existing confidence procedure, produces a uniformly superior procedure. This method, called refinement, actually produces a family of procedures that constitute a complete class. Although details are only given for the Poisson distribution, the refinement process is applicable to any discrete distribution. Tables of refined Poisson confidence intervals are given for confidence levels of .95. An example of refining negative binomial confidence intervals is also included.