Title: Bounding program benefits when participation is misreported
Authors: Lina Zhang - University of Amsterdam (Netherlands) [presenting]
Denni Tommasi - Monash University (Australia)
Abstract: Instrumental variables are commonly used to estimate treatment effects in case of non-compliance. However, the endogenous program participation is often misreported in survey data, and standard techniques are not sufficient to point identify and consistently estimate the effects of interest. We first establish a link between the true and mismeasured effect which is mediated by a parameter of the misclassification probabilities. Second, we provide an instrumental variable method to partially identify the heterogeneous treatment effects when both non-compliance and misreporting of treatment status are present. Third, we formalize a strategy to combine external information about misclassification probabilities of treatment status to tighten the bounds or to obtain a point estimate. Finally, we develop ivbounds, a new Stata package that we use to reassess the benefits of participating in the 401(k) pension plan on savings. Our method has several applications. First, it can be used as the leading strategy in any setting where the practitioner knows that the endogenous binary treatment is not well measured. Second, it can be used as the leading robustness check in case misreporting is only suspected. Third, it can be used to assess the sensitivity of program benefits under different assumptions of the misclassification probabilities.