|
Abstract:
|
Multinomial Logistic Regression has wide applications in many fields. Most of the literature on this topic depends on classical methods which assumes that the design variable X is non-stochastic in nature. In recent years, however, it has been recognized that non-stochasticity of the design variable is too restrictive and nonrealistic: see, Oral (2006). By the use of the modified maximum likelihood methodology, the model has been opened up to stochastic design variable. The aim of this study is to consider the estimation and hypothesis-testing problems in multinomial logistic regression models with a stochastic variable X. MML method is used in all of the theoretical derivations. MML estimators are known to be highly efficient and robust. Mathematical properties of these estimators and test statistics will be analyzed through simulations and real life examples will be carried out.
|
