Analysis Of Binary Outcomes With Missing Smoking Data

Longitudinal data combines the Cross-section data and time data characteristics,and thus not only be able to analyze the trends over time better,but also be able to more accurately reflect the differences between between-subjects and within-subjects.However,if the study time is too long,some individuals will drop-out,so we can't get the measurements of these individuals at some moments,thus result in missing data.If we can't be reasonable to deal with such missing data,then the inferences will be discredible.How to deal with the missing data,is a very practical matter.In this paper,we consider a data set which comes from Donald Hedeker'article,evaluated the effectiveness of different smoking cessation treatments,for the original author make up three methods:Missing = smoking,last observation Carried forward(LOCF) and a little multiple imputation.First,I establish some reasonable graphical models according to the actual trial design,then analyze the graphical models which are identifiable,impute the missing data using EM algorithm,return to the experimental background,based on the missing data's value of imputation,assume whether the group effect is significant or not and the impact of baseline, we also compare the results with the results of the original article.