Estimating Equilibrium Choice Models with Social Interactions and Network Effects: Theoretical Foundations and Emprical Applications

在许多经济和社会环境中，个人的行为或福利直接受到相关参照组中其他个人的存在、特征或行为的影响。仅举几个例子，孩子们可能会受到同学的影响，可能会受到邻居的房主的影响，或者会受到其他驾车者的影响。在许多这样的设置中，个人并不是随机分配到他们的参照组，而是主动决定加入哪个组。这种将个人随机分成小组的做法导致了对小组内其他小组成员的影响以及对小组最初选择的影响的不正确推断。例如，一个设施齐全的社区可能会吸引许多高收入居民，而且房价也很高。如果数据中看不到这些便利设施，一项天真的分析很可能会将高房价归因于高收入邻居的存在，并暗示高收入个人有强烈的同居偏好。该项目开发了一种方法，用于适当确定社会互动在将个人分类为参考群体方面的作用。这一策略利用了这样一个事实，即每个人对参照组的选择受到他们可选选项的性质的影响(在我们前面的例子中，可用邻居的集合)。这项研究首先开发了一种方法，使用这种变异源来识别分类模型中的纯拥堵和聚集交互作用。然后，这一方法将扩展到具有更复杂形式的社会互动的模型，在这些模型中，个人不仅受到同一参照组中那些人的人数的影响，而且还受到其属性的影响。目前确定的这种方法的应用范围从城市经济学、公共经济学和环境经济学扩展到产业组织中的非价格竞争模型。对于各种政策考虑，正确确定这些环境中的社交互动是极其重要的。我们通过一系列的实证应用来论证这项研究的意义，这些应用既是为了证明该方法，也是为了说明其实际重要性。第一个利用1990年美国人口普查数据对旧金山湾区进行社区分类研究。这一分析提供了家庭如何在社区重要特征(位置、学校、犯罪、社会人口、住房和价格)之间进行权衡的完整图景，以及这些权衡对于收入、种族、教育、就业和家庭结构等不同特征的家庭是如何变化的。在正确确定了支撑城市住房市场的一系列复杂偏好之后，该项目进行了一些模拟，旨在揭示种族隔离、犯罪模式、学校组成、通勤模式和房价地理分布等重要综合城市现象的根本原因。然后，这一应用程序将扩展到更具体地关注一系列教育政策问题。第二套申请涉及一个非常不同的政策问题：全球气候变化对发展中国家的影响。虽然已经开发了衡量全球变暖的市场化(农业)影响的技术，但由于缺乏描述住房和其他地理上非贸易商品价格在不同地点之间变化的关键数据，用传统技术衡量非市场化影响，例如对气候便利设施的影响，并不令人满意。为这些商品指定一个均衡模型，并根据观测到的定居模式和我们的识别战略对其进行估计，将产生对巴西全球变暖舒适成本的全面衡量，并讨论其对《京都议定书》后附件一国家与最不发达国家之间关于温室气体减排努力的谈判的影响。该项目还将资助收集数据，以便将这一分析推广到中美洲和南美洲的其他发展中国家。

In many economic and social settings, the behavior or welfare of an individual is directly affected by the presence, characteristics, or behavior of other individuals in an associated reference group. Children might be affected by their classmates, home-owners by their neighbors, or commuters by their fellow motorists, to give just a few examples. In many of these settings, individuals are not randomly assigned to their reference groups but rather actively decide which group to join. This non-random sorting of individuals into groups leads to improper inferences about the influence of other group members both within the group and in the initial choice of group. A community with great amenities, for example, is likely to attract many high-income residents and have high housing prices. If the amenities are not seen in the data, a naive analysis is likely to attribute the high housing prices to the presence of high-income neighbors and suggest that high-income individuals have strong preferences to live with one another. This project develops a methodology for properly identifying the role of social interactions in the sorting of individuals into reference groups. This strategy draws on fact that each individual's choice of reference group is affected by the nature of their alternative options (the set of available neighborhoods in our earlier example). The research begins by developing a methodology that uses this source of variation to identify pure congestion and agglomeration interactions in sorting models. The methodology will then be extended to models with more complex forms of social interactions in which individuals are affected not only by the number but also the attributes of those in the same reference group. The scope of the currently identified applications of this methodology extends from urban, public, and environmental economics to models of non-price competition in industrial organization. Properly identifying social interactions in these settings is extremely important for a wide variety of policy considerations. We demonstrate the significance of this research with a series of empirical applications designed to both demonstrate the methodology and illustrate its practical importance. The first uses 1990 US Census data for the SF Bay Area to study neighborhood sorting. This analysis provides a complete picture of how households trade-off between important features of neighborhoods (location, schools, crime, socio-demographics, housing, and price), as well as how these trade-offs vary for households with different characteristics including income, race, education, employment, and family structure. Having properly identified the complex set of preferences that underlies the urban housing market, this project conducts a number of simulations designed to uncover the underlying causes of important aggregate urban phenomena such as racial segregation, crime patterns, school compositions, commuting patterns, and the geographic distribution of housing prices. This application will then be extended to focus more specifically on a series of education policy questions. The second set of applications address a very different policy issue: the consequences of global climate change in developing countries. While techniques are well developed to measure the marketed (agricultural) impacts of global warming, non-marketed impacts, such as those on climate amenities, have not been satisfactorily measured with traditional techniques owing to a lack of key data describing inter-location variation in the prices of housing and other geographically non-traded commodities. Specifying an equilibrium model for these commodities and estimating it with observed settlement patterns and our identification strategy yields a full measure of the amenity cost of global warming for Brazil, and its implications for post-Kyoto negotiations over greenhouse gas abatement efforts between Annex I countries and LDC's are discussed. This project will also fund the collection of data for the extension of this analysis to other developing countries in Central and South America.