This paper discusses the problem of estimation and inference on the effects of time-varying treatment. We propose a method for inference on the effects treatment histories, introducing a dynamic covariate balancing method combined with penalized regression. Our approach allows for (i) treatments to be assigned based on arbitrary past information, with the propensity score being unknown; (ii) outcomes and time-varying covariates to depend on treatment trajectories; (iii) high-dimensional covariates; (iv) heterogeneity of treatment effects. We study the asymptotic properties of the estimator, and we derive the parametric convergence rate of the proposed procedure. Simulations and an empirical application illustrate the advantage of the method over state-of-the-art competitors.
翻译:本文讨论了对时间变化治疗效果的估计和推论问题。我们提出了对效果变化治疗史的推论方法,采用了动态的共变平衡法,结合惩罚性回归。我们的方法允许(一)根据任意的过去信息指定治疗方法,其倾向性分数不详;(二)结果和时间变化共变法取决于治疗轨迹;(三)高维共变体;(四)治疗效果的异质性。我们研究了测算器的无药可治特性,并得出了拟议程序的准参数趋同率。模拟和实验应用说明了该方法相对于最先进的竞争者所具有的优势。