A Practical Proactive Proposal for Dealing with Attrition: Alternative Approaches and an Empirical Example
针对调查无回应和样本流失损害估计有效性的问题，提出主动为观测创建工具变量的方法，如事先随机分配不同追踪强度，并用“搬向机遇”项目数据验证，得到与完整行政数据相近的无偏估计且置信区间更窄。
Survey nonresponse and attrition undermine the validity of many and possibly most econometric estimates. We propose that survey administrators and evaluators proactively create an instrument for observation, for example, by ex ante randomizing participants to differing intensity of follow-up. We illustrate how to apply our proposed methodology using a carefully conducted randomized controlled trial, the Moving to Opportunity demonstration project, which de facto randomly assigned a subset of subjects to more intensive follow-up. The approach yields treatment effect estimates similar to the unbiased estimator based on complete administrative data and has narrower confidence intervals than alternative bounding approaches.