Linear Mixed Model For A Group Of Depression Efficacy Data

This paper fits unbalance longitudinal data from repeated measure data ofHamilton depression rating scale and establish the linear mixed effects.First, basedon exploratory data anlysis and test of normality assumption, this paper confirmsthe normality of the model assumption is reasonable and gives the basic structure ofmodel. Then we respectively investigate possible covariance structures of therandom effect and random error. Maximum likelihood estimation and restrictedmaximum likelihood estimation are used to take optimal choice on with-subject andbetween-subject covariance structures. Two methods get the same results thatwithin-subject covariance is AR(1) structure,between-subject covariance is diagonalstructure. Then we obtain that the best model is linear mixed effect model with AIC,BICcriterion and likelihood ratio test. By the plot of standardized residuals and itsQQ-plot, we can see that linear mixed effect model well fit Hamilton depressionrating scale data. We draw the conclusion by predicting: after study the relationabout Hamilton depression rating scale with week, sex, level of IMI, level of DMIand endogenous/nonendogenous, state the drag is more effective in patients withendogenous depression. We use R software to complete all calculations of this paper.