Abstract: With the continuous development of information technology, the communication between individuals becomes easier. Meanwhile, the emergence of various types of e-commerce platforms makes the decisionsof individuals become more and more interdependent. The researches of innovation diffusion (starting from the classical Bass Model)focus on the diffusion process from the perspective of market level. However, the existing studies have some limitations, such as the completely connected network which divorced from practice,and the overlook of the heterogeneity between individuals. Thus, more researchers focus the innovation diffusion from the perspective from individual level, and connect individuals' adoption behavior with markets' diffusion results. And, this connection should not lack the help of social networks. Luckily, benefited from the improvement of accessibility of social network data, the studies are able to realize.This study focuses on the influence of peers on the individual behavior of new product adoption, and connects the individual behavior and market diffusion process via social networks. A unique dataset contains more than 1.2 million users' information is taken, which come from the biggest online social network in China. It has several information including a complete social network, individual attributes and adoption behavior. Firstly, a shuffle test is used to identify the peer influence. But, the endogenous from homophilycan not be ignored. Then, the PSM model with both static and dynamic model is used to control the endogenous effectively. And the effect of peer influence is found to exist in adoption process. Furthermore, it accounts for 72% of the total factors. In addition, it is found that the network structure of individuals impacts the peer influence, and the relationship between the tie strength and peer influence has a positive moderating effect. A method and empirical evidence is provided for the future studies.

Key words: innovation diffusion, peer influence, social network, data-driven