The Application Of Cumulative Multilevel Model:bases On Bayesian Approach

[objective]:In this article,the cumulative multilevel model which bases on theBayesian approach is used to analyze the hierarchical structured data with intraclasscorrelation,in order to acquire the methodology and verify advantage of themodel,additionally,provide a practicality analysis method to hierarchical structureddata.[method]:The data of the cross-sectional study always be hierarchical,thecharacteristic of which is obvious,they are intraclass correlation,cumulative dependenceand variation in distinct level.Traditional logistic regression model can not satisfy thesecharacteristic,while multilevel cumulative logistic regression is able to solute thisproblem.Besides,the bias of the result from the traditional estimate method iscertified.Thus,the multilevel cumulative logistic regression model basing on bayesianapproach should be more applicable.First,the basis principles of cumulative multilevel model and bayesianapproach,MCMC method will be introduced in this essay.Then,cumulative multilevelmodel bases on bayesian approach will be applicated to analyze the data from thesmoking control investigation in GuangZhou for the result and the process of theanalysis.At last,the differences between the traditional analysis methods and thecumulative multilevel model is represented,according to DIC andparameters.Meanwhile,the simulate trace graph and Raftery–Lewis information areused to evaluate convergence of the parameters.[result]:The analytic result of the smoking control investigation in GuangZhoushows that the data of which is correlated,and the sex,race,workingcondition,Environmental tobacco smoke,the score of the smoking control knowledgeand smoking knowledge are the influenced factors to legislative attitude of the publicsmoking control.In addition,random slope model displays that the more positiveapproval to the smoking legislation from the community,the more positive approval tothe smoking legislation from the resident,retired or waiting for employment,in the samecommunity.Contrastly,cumulative logistic regression ignores the higher levelinformation.In addition,DIC and standard error of the parameters indicates that the fitness to the data by cumulative multilevel model is better and the estimate value of thecumulative random slope model is preciser than cumulative logistic model.The residualanalysis also explore the community with the most positive or negative attitude.The simulate trace graph indicates that the markov chain converges to theobjective distribution,and Raftery–Lewis information presents the practical iteration isadequate to acquire the accuratep2.5、p97.5,which means that the result can be used asstatistical inference.[conclusion]:The result from the investigating hints that what population shouldbe noticed by further research.Cumulative multilevel model with better fitness to thedate considers the intraclass correlation and reduces the bias of the parametersestimate by resolving the residual,besides,acquires more significant information fromthe data.Hence,the cumulative multilevel model is more outstanding than cumulativelogistic.The bayesian approach takes the uncertainty of the variance component intoaccount,so the estimation precision of higher level variation is enhanced.In conclusion,cumulative multilevel model with bayesian approach is a reasonableanalytic method to the hierarchical structured data.