Conclusion

This study found that the combination of stepwise regression analysis and data envelopment analysis can effectively improve the validity of data envelopment analysis.The stepwise regression method removed data of little relevance from the input data system and optimized the regression model.

According to the results of stepwise regression，the variables that explained GDP better are fixed assets，junior middle school education human capital，senior middle school education human capital，college and above degree education human capital.The primary school education human capital is erased from the production function model，which means that the contribution of primary school education human capital in economic growth is weak.

DEA analysis shows that the regional human capital is obviously different，heterogeneity of human capital investment and output efficiency is generally not high.Only a few areas achieve the heterogeneous human capital input and output DEA effective.The DEA evaluation of the heterogeneous human capital shows that the input redundancy mainly because of the junior high school education human capital.In the regression model，the coefficient of human capital of the junior high school education is negative，and the negative effect of the level of human capital on the output is also explained.The directions of heterogeneous human capital structure optimization mainly are human capital flow to other levels or flow to other regions.

And the main way is promote a large number of junior middle school education human capital improve to a higher degree human capital，that is to improve the level of human capital of the junior middle school education through higher education or vocational education，so as to meet the actual needs of the industrial structure transformation in China at present.