Abstract: For  the  partially linear model with longitudinal data, we established the information criteria of variable selection based on empirical likelihood method, and proved the asymptotic properties of the new variable selection method. We simulated and compared the limited sample properties of parameter and information criteria based on empirical likelihood. The results show that the information criteria based on empirical likelihood method overcomes the difficulty that the parameter  likelihood functions are sometimes difficult to obtain, and the model selection effect is better.

Key words: empirical likelihood, longitudinal data, partially linear model, variable selection;