multinomial response models, for modeling and determining important factors in different contraceptive methods in women

difference aspects of multinomial statistical modelings and its classifications has been studied so far. in these type of problems y is the qualitative random variable with t possible states which are considered as classifications. the goal is prediction of y based on a random vector x ? ir^m. many methods for analyzing these problems were considered. one of the modern and general method of classification is classification and regression trees (cart). another method is recursive partitioning techniques which has a strange relationship with nonparametric regression. classical discriminant analysis is a standard method for analyzing these type of data. flexible discriminant analysis method which is a combination of nonparametric regression and discriminant analysis and classification using spline that includes least square regression and additive cubic splines. neural network is an advanced statistical method for analyzing these types of data. in this paper properties of multinomial logistics regression were investigated and this method was used for modeling effective factors in selecting contraceptive methods in ghom province for married women age 15-49. the response variable has a tetranomial distibution. the levels of this variable are: nothing, pills, traditional and a collection of other contraceptive methods. a collection of significant independent variables were: place, age of women, education, history of pregnancy and family size. menstruation age and age at marriage were not statistically significant.