Hollow village is the epitome of China's economic and social changes in the village scale, characterized by the decline of industries, the outmigration and the large existences of derelict lands and homestead, which poses a real threat to land resources, food security, village environment, social stability and rural governance. The research on the formation mechanism of homesteads hollowing is not only helpful for understanding the hollow village phenomenon in China, but also an important prerequisite for the promotion of hollow village governance and the implementation of rural revitalization strategy. Based on 1238 households’ questionnaires, the paper establishes a mathematical model and uses quantile regression method to explore theoretically and empirically the differential causes of homestead hollowing. The quantile regression results show that: (1) the average estimated hollow rate of homesteads is 29.14%, which is much higher in mountainous and hilly areas than in plains areas, that is, 30.93%, 35.58%, and 20.30%, respectively, and higher in central and western regions than in eastern regions, 31.21%, 34.29%, and 22.41%, respectively. (2) Homesteads hollowing is the result of more factors than one, of which, socio-economic development, terrain, village location and social security system are the dominant factors for homesteads hollowing, while the influences of cultivated land resource endowment, family size, the peasants’ notion of "Jiaye", outmigration proportion and institutional factors vary across quantiles. (3) The positive impacts of household income and the distance from the township increases with the rising quantiles, while the impacts of land titling and the policy implementation of “one homestead for one "hu" only rise in the lower and median quantiles, but hardly statistically in upper quantile, which was mostly caused by integral outmigration. Additionally, the farmer's participation of new rural social pension insurance can significantly decrease the homesteads hollowing rate, while the negative effects of the terrain present a U-shaped distribution. Finally, the paper proposes some further research directions that can be developed, such as exploring the role of social network in the outmigration with the help of large-scale micro population database, and evaluating the land titling policy with the difference-in-difference method. The findings of this study can not only contribute to understanding the formation mechanism of homesteads hollowing, but also shed light on future studies on the targeted governance.