Model Selection For Linear Mixed Models

The regression model selection is an important research direction of modern application statistic. It mainly researches the rationality of regression model and the independent variables selection. Linear mixed model is a kind of linear models, which includes both fixed effects and random effects. It is frequently used in biology, medicine, economics, determining, sampling designs and quality control procedures, and so on. People usually put all relational independent variables into the regression model. It may increase the amount of the regression model calculation, and may decrease the precision of the estimation and the prediction. Therefore it is costly to acquire some independent variables data and put them into the model to solve some practical problems,This thesis will focus on the model selection problem in linear mixed model, and put forward a new criterion. This thesis uses the Mean Square Error Method, and it constructs the Revised MSB criterion. By reviewing the Revised MSB, we can choose the independent variables of the regression model and decrease the calculation. And this criterion doesn't depend on distribution of the model.In this thesis, the criterion is used to solve the salary problem, which is one of the problems of Human Resource Management. The application of the Revised MSB receives right results, and it has the same effect with the other criterion. It can be used as a new way to solve the policy of compensation and benefit system problem in Human Resource Management.