In this dissertation address the problem of an ill-conditioned information matrix in logistic regression in the presence of errors in the response variable. The problems that ill-conditioning cause in the generalized linear model framework are reviewed. Some alternative estimators are suggested, evaluated and compared to commonly used estimators The logistic regression model and the misclassification logistic regression model are particular cases of the GLM. The misclassification model is affected by problems of an ill-conditioned information matrix as are all GLM's. A near singular information matrix can often make desirable properties of estimation, prediction and parameter testing unattainable. This dissertation discusses the problems associated with the problem of ill conditioned information matrices in the GLM framework. In this research we found that some kinds of ill-conditioning have an adverse effect on the bias of the maximum likelihood estimator (MLE). Some estimators usually assumed to be more biased than the MLE, may be less biased than the asymptotically unbiased MLE. Some of the alternative estimators suggested for correcting the problem of an ill-conditioned information matrix are reviewed, and some other estimators are proposed
