Primary Submission Category: Randomized Studies
Experimenting under Stochastic Congestion
Authors: Shuangning Li, Ramesh Johari, Stefan Wager, Kuang Xu,
Presenting Author: Shuangning Li*
Stochastic congestion, a phenomenon in which a system becomes overwhelmed by fluctuations in demand, occurs frequently in many industries. In this paper, we study how randomized experiments can be conducted under stochastic congestion to help gain better insights into the behavior of stochastic systems. In particular, we study two switchback experiments, where treatments are simultaneously randomized sequentially for every unit, and a local perturbation experiment, where treatments are randomized independently for each unit. We focus on a simple stochastic system that has a single queue with an outside option. Aiming to estimate the effect of a system parameter on the average arrival rate to the system, we establish central limit theorems for the proposed estimators. We establish that the estimator from the local perturbation experiment is asymptotically more accurate than the estimators from the switchback experiments.