Causal Inference Using Antidotal Variables

This paper shows that incorporating what we call antidotal variables (AV) into a causal treatment effects analysis can with one cross-sectional regression identify the causal effect, the spillover effect, as well as possible biases from selectivity. We apply the AV technique to analyze leave taking arising from the California Paid Family Leave (CPFL) program. Our analysis yields between a 55% and 70% larger treatment effect than the traditional DID methods, which we attribute to confounding effects and spillovers, neither of which are found in traditional studies.

本研究表明，将我们定义的纠偏变量（antidotal variables，AV）纳入因果处理效应分析框架后，仅通过一次横截面回归即可同时识别因果效应、溢出效应，以及选择性偏差带来的潜在偏误。我们将AV分析方法应用于加州带薪家庭休假（California Paid Family Leave，CPFL）计划引发的休假行为研究。本研究测算得到的处理效应较传统双重差分法（Difference-in-Differences，DID）高出55%至70%，我们将该结果差异归因于混杂效应与溢出效应——而这两类效应在传统研究中均未被发现。