Complex Sampling And Multiple Level Model-theestimations Of Per Capita Health Expenditures And Twoweeks Prevalence Of Shanxi Province

ObjectiveIn view of the different types of variables,we adopted the design-based method of statistical inference,selectd the data of fifth national health services survey in Shaanxi Province representative survey,and two types of variables have been estimated based on the design,which is per capita expenditure on health and two weeks prevalence.In addition,take the advantage of multi-stage sampling structure from the fifth national health services survey,comparing and then discussed how to analyze factors of fixed variables while data containing Hierarchical structure,difference between the traditional Logistic regression method and the the multi-level model method. For the multilevel model was used to analyze the influential factors are summarized in the paper.The estimation results for the health service supply and demand has considerable reference value and reference meaning on the assessment of Shaanxi province and lower per capita health expenditure and the two week prevalence rate of the common people.MethodThe fifth national health services survey using four stage stratified cluster sampling with unequal probability.For continuous variables and dichotomous variables, we chose per capita health expenditure and two week prevalence rate to estimate. With per capita health expenditure in the annual per capita income in the proportionand the two week prevalence rate of Shanxi province public health burden and health level. In addition, through the analysis offactors by logistic regression model and multi-level model is applied to the two variables influencing, to clear that there is obvious advantage for multi-levelmodel in comparison with the traditional logistic method while data has hierarchical structure. ResultsIn this study,we use the data of the representative survey data in Shaanxi Province,the fifth national health services survey,to estimate the weighted health expenditure per capita in Shaanxi province and two week prevalence rate.The mean of the per capita expenditure on health of residents in Shaanxi provinceis about 1048.31 yuan,median is 500.00 yuan,the estimation of mean of SRS is 1123.09(1024.48,1221.70)yuan,the DEFF is 1.68;Two week prevalencerate is about 17.48%, the estimation of mean of SRS is 17.27%(16.96,17.58), the estimation of mean of complex sampling is 12.29%(10.59,13.99). the DEFF is 30.17.In processing complex sample with obvious hierarchy, the general linear regression model and Logistic regression model were fitting traditional effect for continuous variables and dichotomous variablesis not ideal, the former adjustment R2 is only 0.0720,while the latter is only 0.3967.The result of comparison of multi-levelmodel likelihoodratio with the traditional regression model show that,in the face of such data,multi-level model has obvious advantages. ConclusionThe results of design-based statistical inference method vary from the individual values weighted or not and weight methods. The discovery of factors of per capita health expenditure and two week prevalence rate inlogistic regression model and multi-level model tells that,for a data with obvious hierarchical structure,the results of multi-level model is more reliable than the traditional logistic regression model.