Combined With Multiple Imputation Method To Deal With Linear Mixed Model

Linear mixed model is a very important statistical model It has unique advantages when dealing with longitudinal data,interval data and spatially correlated data.Especially for longitudinal data,linear mixed model has been widely used as an analytical tool Linear mixed model has two parameters:fixed effects and variance components.The research on the methods of parameter estimation of linear mixed model has been one of the hot research problems in recent years.This thesis mainly studies the parameter estimation of linear mixed model under longitudinal data.Combined with multiple imputation,a new parameter estimation method is proposed.The theoretical part of this paper introduces the methods of parameter estimation of fixed effects and variance components in linear nixed models firstly,and then introduces the background of missing data,including the reason of missing data,the patterns of missing data and the mechanisms of missing data,highlights the main steps of multiple imutation and the common multiple imputation methods.This part provides the theoretical basis for the parameter estimation method proposed in this paper.In this paper,two linear mixed models-simple linear mixed model and random intercept model are established under the specific longitudinal data-dental data.Based on the consistency of the least square estimation of parameters and moment estimation in the univariate linear regression model,the initial values of the parameters are obtained by using the moment estimation and the least square estimation,and then the final parameter estimations are calculated by combining the multiple imputation method and the matlab software.The validity of the method is verified by using a large number of simulations.Finally,the model selection is made and the results show that the random intercept model is more suitable for the dental data.