Confidence Coefficients for Usual Binomial Interval Estimates under Hypergeometric Models.

摘要

Usually when one discusses confidence intervals for the population success proportion,theta,constructed under binomial sampling it is often presumed that the coverage probability under a hypergeometric model would be at least as great. There is good intuitive support for this contention. The binomial distribution has heavier tails (i.e.,larger variance) than the hypergeometric. Moreover when hypergeometric intervals are constructed often the finite sampling correction is used thereby shortening the interval to achieve the same nominal level. In the paper the authors establish conditions under which the above presumption is valid and demonstrate that it is not true without reservation. (Author)