标题： 几何混合下的开关实验 显示英文标题

标题： Switchback Experiments under Geometric Mixing

摘要： 回溯开关是一种实验设计，通过反复打开和关闭整个系统的干预措施来测量治疗效果。 回溯开关实验是一种克服单元间溢出效应的强大方法；然而，它们容易受到时间延续偏倚的影响。 本文考虑了在以几何速率混合的马尔可夫系统中的回溯开关实验的性质。 我们发现，在这种情况下，标准的回溯开关设计因延续偏倚而受到严重影响：其估计误差随着实验范围$T$的增加以$T^{-1/3}$的速度衰减，而在没有延续偏倚的情况下，可以达到更快的$T^{-1/2}$衰减速率。 我们还表明，明智地使用预热期可以大大改善这种情况，并使误差以$\log(T)^{1/2}T^{-1/2}$的速度衰减。 我们的正式结果在经验评估中得到了反映。 显示英文摘要

摘要： The switchback is an experimental design that measures treatment effects by repeatedly turning an intervention on and off for a whole system. Switchback experiments are a robust way to overcome cross-unit spillover effects; however, they are vulnerable to bias from temporal carryovers. In this paper, we consider properties of switchback experiments in Markovian systems that mix at a geometric rate. We find that, in this setting, standard switchback designs suffer considerably from carryover bias: Their estimation error decays as $T^{-1/3}$ in terms of the experiment horizon $T$, whereas in the absence of carryovers a faster rate of $T^{-1/2}$ would have been possible. We also show, however, that judicious use of burn-in periods can considerably improve the situation, and enables errors that decay as $\log(T)^{1/2}T^{-1/2}$. Our formal results are mirrored in an empirical evaluation.