Abstract: China's higher education has achieved new breakthroughs in research and education, innovation capabilities and social services, while at the same time we also notice that the traditional planning and allocation of resources under the "211" and "985" engineering models has resulted in asymmetric and unbalanced input and output of higher education. The overall level of Chinese higher education is still low and the capacity for independent innovation and society service function is still weak. Under the background of building world-class universities and first-class disciplines, the resource allocation model based on cooperation and competition requires that China increase the input-output efficiency of higher education.In this paper, the Window-Malmquist index method and spatial clustering analysis method are used to measure and analyze the input-output efficiency of higher education in 31 provinces (cities) in China from 2004 to 2015 and its evolution rules and spatial differences. The results show that:(1) the input-output efficiency of China's higher education shows an effective state of DEA, and technological progress is the main contributor to its improvement. Meanwhile, the existence of "regional chasm" in higher education has led to catch-up effect which lowers the efficiency value; (2) From the view of time evolving, the efficiency of input-output of higher education is generally relatively stable, but the upward trend is not obvious, and there is a possibility of polarization or multi-polarization; (3) in terms of spatial disparity, among the 31 provinces and municipalities, the input-output efficiency in Jiangsu Province is the highest. The input-output efficiency of higher education in eastern coastal areas is characterized by "high" with a prominent higher ratio comparing to other regions (4) from the perspective of spatial agglomeration characteristics, there is a positive spatial relationship among input-output efficiencies in different regions under the influence of technological progress, while catch-up effect reduces the spatial linkage. The main inspirations of this article are as follows:firstly, from the prospective of increasing input-output efficiency of higher education, it is necessary to narrow the regional gap among regions. Secondly, from the prospective of keep stability of higher education input-output efficiency, it is necessary to ensure the stability of higher education input and reform policy. Thirdly, from the prospective of strengthening spatial agglomeration, it is necessary to enhance exchange and cooperation of higher education among provinces and regions.

Key words: higher education, input-output efficiency, spatial difference, Window-Malmquist index method