الفهرس | Only 14 pages are availabe for public view |
Abstract Multicollinearity is one of the major problems in regression estimation methods. The presence of this problem in logistic regression causes an inflation in the variance of the Maximum Likelihood (ML) estimator. It can also inaccurate unstable estimates which affects confidence intervals and hypothesis tests. To overcome this serious problem, some biased estimators such as: ridge estimator, Liu estimator and Liu-type estimator, were proposed before as a way of having smaller Mean Squared Error (MSE) than ML estimator. These different biased estimators in the logistic regression are discussed and, the shrinkage parameters are presented in this thesis. A Monte Carlo simulation is conducted to assess the performance of ridge and Liu-type estimators, applying some estimators from the literature, in terms of MSE and Bias criteria. Different levels of correlation between the explanatory variables, and different sample sizes are considered. It was concluded that the new applied estimators outperform ML estimator in the ridge and Liu-type estimation. Three empirical applications of real data sets with different sample sizes were conducted to prove the findings of the simulation study. It is concluded that the results of the applications agree with the results of the simulation study |