Synthetically Controlled Bandits
This addresses the cost of experimentation in settings with interference or coarse units, such as online platforms, but is incremental as it builds on existing designs like fixed and switchback methods.
The paper tackles the problem of minimizing regret in coarse experimental units, such as region-split experiments on online platforms, by introducing Synthetically Controlled Thompson Sampling (SCTS), which achieves near-optimal regret with minimal loss to inferential ability, as demonstrated through theoretical guarantees and experiments on synthetic and real-world data.
This paper presents a new dynamic approach to experiment design in settings where, due to interference or other concerns, experimental units are coarse. `Region-split' experiments on online platforms are one example of such a setting. The cost, or regret, of experimentation is a natural concern here. Our new design, dubbed Synthetically Controlled Thompson Sampling (SCTS), minimizes the regret associated with experimentation at no practically meaningful loss to inferential ability. We provide theoretical guarantees characterizing the near-optimal regret of our approach, and the error rates achieved by the corresponding treatment effect estimator. Experiments on synthetic and real world data highlight the merits of our approach relative to both fixed and `switchback' designs common to such experimental settings.