Partial Identification of State Dependence

经济学中的许多实证问题都需要回答以下问题的某种形式：过去的结果在多大程度上决定了当前和未来的结果？例如，如果雇主做出雇佣决定的部分依据是申请人过去的工作成绩，那么如果雇主认为这是申请人作为一名工人质量的负面信号，那么以前的失业可能会导致申请人保持失业状态。使用就业结果的数据来衡量这种现象发生的程度是一个众所周知的难题。本提案开发了新的统计模型来解决这个问题，因为它既适用于就业动态，也适用于经济学中的其他类似问题。该项目通过开发新的数据分析方法推进了该领域的发展。对就业结果的应用也促进了国家的繁荣，为我们提供了关于失业如何影响长期就业前景的更好信息。衡量状态依赖性的基本概念上的困难在于，过去结果的影响将与代理之间暂时持续的不可观察的异质性相混淆。例如，在一个小组中观察到，以前失业的代理人不太可能被雇用，这可能是由于过去失业的负面因果效应，但如果失业的代理人不太可能被雇用，这也可能是由于其他不可观察的因素，如偏好或生产力。迄今为止，绝大多数用于测量状态依赖性的统计方法都使用了高度参数化的动态二元选择模型的变体。这些模型依赖于许多强有力的假设，包括对异质性形状的任意功能形式限制。在许多应用程序中，它们很可能被严重错误地指定。本提案的目标是开发和应用透明的非参数方法来从未观察到的异质性中识别状态依赖性。由于状态依赖性点识别的困难，本文提出的方法采用部分识别技术。特别是，PI开发了一个新的动态潜在结果模型，研究了其性质，并将其应用于就业结果中的国家依赖问题。

Many empirical questions in economics require the answer to some form of the following question: To what extent do past outcomes determine current and future outcomes? For example, if employers make hiring decisions based in part on the past employment outcomes of an applicant, then previous unemployment may cause an applicant to remain unemployed, if employers view this as a negative signal of the applicant's quality as a worker. Using data on employment outcomes to measure the extent to which this phenomenon occurs is a notoriously difficult problem. This proposal develops new statistical models for addressing this problem, both as it applies to employment dynamics, and to other similar problems in economics. The project advances the field by developing new methods for data analysis. The application to employment outcomes also advances the national prosperity by giving us better information about how unemployment affects long run job prospects. The fundamental conceptual difficulty with measuring state dependence is that the effect of past outcomes will confound with temporally persistent unobservable heterogeneity across agents. For example, observing in a panel that previously unemployed agents are less likely to be employed could be due to a negative causal effect of past unemployment, but it could also result if unemployed agents are less likely to be employed due to other unobservable factors such as preferences or productivity. To date, the vast majority of statistical methods designed to measure state dependence use variants of highly-parameterized dynamic binary choice models. These models depend on many strong assumptions, including arbitrary functional form restrictions on the shape of heterogeneity. They are quite likely to be severely misspecified in many applications. The goal of this proposal is to develop and apply transparent nonparametric approaches for identifying state dependence from unobserved heterogeneity. Owing to the difficulty of point identifying state dependence, the proposed methods use partial identification techniques. In particular, the PI develops a new dynamic potential outcomes model, studies its properties, and applies it to questions of state dependence in employment outcomes.