Variable Selection In Balanced Longitudinal Model

In daily life, we often encounter with such a kind of data: The same individual or the unit that will be tested is observed at different times, then we get the data that are observed from the individual at different times. This kind of data are called the longitudinal data. When we deal with the kind of data, we can construct the model of the longitudinal data, and use the method of regression analysis to analyze this problem. So we can find the key factors that affect the response variable.When we use the longitudinal modle to deal with the practical problems, one of the most important steps is the selection of the variables. In general, when we analyze the longitudinal model, we often place the variables that are related to the response variables into the model according to the professional knowledge or experience, and the result of doing this is to place the variables that affect the response variables a little or the variables that have nothing to do with the response variables into the model. Then the calculation is very large, and the accuracy of estimation and prediction will be reduced. Furthermore, in some cases, the costs of the data that we observed from some variables are very expensive. If we place all the variables that affect the response variable a little or the variables that have nothing to do with the response variables into the model, the result adds the costs of collection the data that observed and the costs of the model that is applied. So, when we analyze the model of longitudinal data, it is necessary to choose the variables very carefully.The main content of this paper is to apply the method of Lasso(Least absolute shrinkage and selection operator) that is proposed by Tibshirani(1996) to shrinkage the coefficients of the model of longitudinal data and set some coefficients to zero, and use AIC or BIC to cut the coefficients that equal to zero, so we can determine the order of the model, and achieve the objectives of model selection.