Dimension Reduction Methods And Its Applications In Panel Data Regression Model

Panel data section can be considered as a combination of time series and cross-section data; it can produce a large number of variables with a more complex structure. However, a large number of the variables will result in a dilemma in data analysis. This paper introduces LASSO, SCAD, adaptive LASSO and Group LASSO methods under three criteria to the panel regression model, and compares the simplification and the accuracy of interpretation in different models.Upon the analysis of average lifetime of twelve countries panel data of ten years, we found that the results of SCAD and adaptive LASSO method is very effective, they have more accurately results in the dynamic model with lag,and can eliminate more variables than LASSO, as well as in dealing with three determination criteria. Group LASSO considers the each group as a unit in the model; it still remains the collinearity problem and can’t effectively determine the lag periods.In three determination criterions, the GCV criteria and the AIC criteria have similar performance in variable selection, but the BIC criterion is better than other twos. Incorporated herein by dimensionality reduction methods, we reduce the variables which have no significant influence in the dynamic model of variable coefficients, and it still can remain the interpretation. And on this basis, we can analyze the differences of main factors influencing the average lifetime in different time and cross-section dimension.