A statistical test for a parameter describing the association in correlated binary data.

摘要

In this thesis we develop a test of statistical hypothesis for the association parameters in the non-linear dynamic conditional probability model ( NLDCP) suggested by Farrell and Sutradhar (2006) for longitudinal correlated binary data. In the development of such a test we consider the simplest case for the above model where the probability associated with the response at a particular time is only conditionally related to the previous response. We then derive the Rao-Fisher scoring test statistic as well as both its asymptotic and non-asymptotic probability distributions, and derive a power function for a decision rule based on the above test.