On Unconditional Exact Tests For Comparing Two Proportions with
Incomplete Data
ABSTRACT:
Various asymptotic test procedures have been developed for
testing the equality of two bionomial proportions with partially
incomplete paired date under random mechanism. Test procedures
that discard incomplete obervations have been shown to have
inferior power in comparison to those utilizing all observations. On
the other hand, asymptotic test procedures that utilize all
observations could still be questionable in small smaple problems
or sparse data structures. In this talk, unconditional exact test
procedures are proposed for testing the equality of two binomial
proportions with partially incomplete paired data under random
mechanism. A real example from a neurological study is used to
illustrate the applicability of the proposed unconditional exact test
procedures. Empirical study is conducted to investigate the
performance of various test procedures with respect to size and
power.