III: Medium: Design and analysis of experiments on networked populations

现代技术使人们能够大规模地收集和处理有关线上和线下社交过程的详细数字痕迹。例如，在社交媒体平台上运行的应用程序提供了与数亿用户互动的机会，在一个环境中，任何人的参数都可以根据其他人的行为进行指定/更改。从技术上讲，在这类应用中，我们通常可以访问所谓的社交网络信息，而现有的策略通常不会利用这些信息来设计和分析实验。在这种联网系统中获取和处理知识的工具是关键。特别是，设计和分析实验的算法和统计策略将为解决许多重要的公开问题和政策问题提供必要的垫脚石。该研究开发了一个集成的研究和教育计划，解决了三个关键问题：(1)如何消除用于获取网络人口数据的流行链接跟踪算法的偏差；(2)如何设计和评估优化给定目标的新网络抽样算法；以及(3)如何在网络人口上设计随机实验，使其能够估计因果影响，而不是关联影响。它将开发两个案例研究来在实践中演示这些工具：(1)在美国和欧洲对教育和网络资本对向上流动的影响进行大规模研究；(2)对改进分布式软件系统延迟的根本原因分析的因果策略进行实证分析。有关更多信息，请参阅项目网站：http://www.people.fas.harvard.edu/~airoldi/iis-design_analysis_experiments_networks.html

Modern technology enables the collection and manipulation of detailed digital traces about online and offline social processes, at scale. Apps that run on social media platforms, for instance, provide the opportunity to engage with hundreds of millions of users, within an environment whose parameters for any individual can be specified/changed in response to the behavior of others. Technically, in such applications we often have access to so-called social network information, which is typically not leveraged by existing strategies to design and analyze experiments. Tools for knowledge acquisition and manipulation in such networked systems are key. In particular, algorithmic and statistical strategies to design and analyze experiments that leverage information about connectivity among the components of a system of interest individuals, and can operate at scale, will provide necessary stepping stones for tackling many important open problems and policy questions.This research develops an integrated research and educational program that addresses three key problems: (1) how to remove bias from popular link-tracing algorithms used to acquire data on networked populations; (2) how to design and evaluate new network sampling algorithms that optimize a given objective; and (3) how to design randomized experiments on networked populations that enable the estimation of causal effects, rather than associative effects. It will develop two case studies to demonstrate these tools in practice: (1) A large-scale study of the effects of education and network capital on upward mobility, in the United States and Europe; and (2) An empirical analysis of causal strategies to improve of root-cause analysis of latency in distributed software systems.For further information see the project web site located at: http://www.people.fas.harvard.edu/~airoldi/iis-design_analysis_experiments_networks.html