Network Formation and Peer Effects in the USAFA

pi建议研究社会网络的形成。该项目包含两个组成部分，一个是经验性的，一个是方法论的。在实证部分，pi计划从美国空军学院（USAFA）的学员那里收集有关友谊联系的新数据。这些数据将允许pi调查同伴效应的潜在机制。卡雷尔和韦斯特早期的研究通过将学员随机分配到中队，证明了同伴效应的存在。他们的研究表明，在他们的中队中，其他学习成绩稍好的学员会使学员受益。在这里，pi将调查这些影响背后的机制，特别是这些影响是否通过友谊和社会网络的形成而起作用。在方法学方面，pi计划为有限人口中的网络形成开发一种新的结构模型，并开发贝叶斯方法来对这些模型进行推理。该模型是一个网络形成的战略模型，它假设，从一个空的网络开始，一系列的机会出现，让成对的个体形成友谊联系。如果有机会，两个潜在的朋友都会评估他们从链接中获得的效用，考虑到他们和朋友的特点，考虑到网络的状态，并决定是否建立友谊。经过一系列的机会，网络最终确定。这些模型将允许同质性，在相似的个体之间更有可能形成联系。学员随机分配到中队，将允许pi评估网络形成的因果效应，例如，改变小组的组成是否会影响中队中友谊的数量和社区的程度。这些模型的一个新颖之处在于，pi允许友谊链接的效用取决于已经建立的友谊的数量和性质。更广泛的影响：社会网络的研究已经在许多学科中得到广泛应用。在这个与经济学家和一位社会学家的跨学科提案中，pi计划开发和说明将用于研究社交网络的各种领域的新方法。

The PIs propose to study the formation of social networks. The project contains two components, one empirical, and one methodological. In the empirical component the PIs plan to collect new data on friendship links from cadets at the United States Airforce Academy (USAFA). These data will permit the PIs to investigate the mechanisms underlying peer effects. Earlier work by Carrell and West demonstrated, through randomized assignment of cadets to squadrons, the presence of peer effects. Their work suggested that cadets benefit from the presence in their squadron of other cadets who have slightly higher academic performance. Here the PIs will investigate the mechanisms behind these effects, and in particular whether those effects work through the formation of friendships and social networks.In the methodological component, the PIs plan to develop a new structural model for network formation in a finite population and develop Bayesian methods for conducting inference for such models. The model is a strategic model of network formation, which assumes that, starting with an empty network, a sequence of opportunities arises for pairs of individuals to form friendship links. Given the opportunity, both potential friends evaluate the utility they get from the link, given their and their friends' characteristics, and given the state of the network, and decide to form a friendship or not. After a sequence of opportunities the network is finalized. These models will allow for homophily where links are more likely to be form between similar individuals. The random assignment of cadets to the squadrons will allow the PIs to assess causal effects on network formation, e.g., whether changing the composition of the group affects the number of friendships and the degree of community in the squadron. A novel feature of these models is that the PIs allow the utility of a friendship link to depend on the number and nature of friendships already established.Broader Impacts:The study of social networks has become widespread in many disciplines. In this cross-disciplinary proposal with economists and a sociologist, the PIs plan to develop and illustrate new methods that will be of use for a variety of areas where social networks are studied.