Social and professional networks affect labor market dynamics, knowledge diffusion and new business creation. To understand the determinants of how these networks are formed in the first place, we analyze a unique dataset of business cards exchanges among a sample of over 240,000 users of the multi-platform contact management and professional social networking tool for individuals Eight. We develop a structural model of network formation with strategic interactions, and we estimate users' payoffs that depend on the composition of business relationships, as well as indirect business interactions. We allow heterogeneity of users in both observable and unobservable characteristics to affect how relationships form and are maintained. The model's stationary equilibrium delivers a likelihood that is a mixture of exponential random graph models that we can characterize in closed-form. We overcome several econometric and computational challenges in estimation, by exploiting a two-step estimation procedure, variational approximations and minorization-maximization methods. Our algorithm is scalable, highly parallelizable and makes efficient use of computer memory to allow estimation in massive networks. We show that users payoffs display homophily in several dimensions, e.g. location; furthermore, users unobservable characteristics also display homophily.
翻译:社会和专业网络影响劳动力市场的动态、知识传播和新的商业创造。为了了解这些网络是如何形成的决定因素,我们首先分析在多平台接触管理和专业社交网络工具的240,000多名个人用户抽样中进行商业卡交换的独特数据集。我们开发了具有战略互动的网络形成结构模型,并估算了取决于商业关系构成以及间接商业互动的用户报酬。我们允许在可观测和不可观察特点中的用户差异性影响关系的形式和保持。模型的固定平衡提供了一种可能性,这种可能性是我们可以以封闭形式描述的指数随机图形模型的混合。我们克服了在估算方面的若干计量和计算挑战,为此利用了两步估算程序、变化近似和微度-氧化方法。我们的算法是可缩放的,高度平行的,并且有效地利用计算机记忆来进行大规模网络的估算。我们显示,用户的付款在多个维度上表现出同质性,例如位置;此外,用户无法观测的显示性特征。