Application Research On Dichotomous Data Of Multilevel Statistical Mode

Objective: hiberarchy data are existed in practical research in large amount, with the characteristic of clustering in geographical, administrative or any specific regions. Traditional regressive mode is inappropriate to analyse these data. Single hypothesis of traditional regressive mode is not accordance to it for it may cause bias and lead to a wrong result. This study is to explore the application of multilevel statistical mode which is to analyse the hiberarchy data. Data of the Fourth National Health Service Survey (western Chongqing area) were adopted to study with double level logistic regression. This study is to investigate the methodology of multilevel mode in practical application, so that to provide reference for further study.Methods: two-week morbidity was recruited as response variable. Sex, age, marital status, educational level, medical insurance and economic status of the interviewees (Chongqing city residents, 15 years old or above) were the explain variables. Based on the results of descriptive study and univariate study, MLwiN was adopted to fit the double level logistic regressive mode; DIC was adopted to respond the goodness of fitting. To compare the results of double level mode with the dichotomous single level logistic regression mode. Results: 1. There was hiberarchy in two-week morbidity of Chongqing residents by double level mode. They had significance among different streets. 2. Double level logistic regressive mode and dichotomous single level logistic regressive mode were almost the same in selecting variables and hypothesis testing. However, to the indices such as economic status which were obviously clustering, double level logistic regressive mode may be more sensitive than that dichotomous single level logistic regressive mode did. 3. Mode fitting goodness test showed that double level logistic regressive mode was better than dichotomous single level logistic regressive mode, with DIC value being 4515.06 and 4583.17, respectively.Conclusions: multilevel mode was adopted in categorical variable of hiberarchy data in this study. Being a new access to analyse this kind of data, this method is more applicable. However, in practical use, for the limitation of statistical software development, there was imperfection in multilevel mode test, diagnosis and residual analysis of category variable. It will be a new research direction to explore the category variable multilevel mode.