This article shows when a triple difference strategy using an imperfect control category improves on the double difference strategy for estimating an average treatment effect. For example, a product is treated in one place and not another leading to a double difference strategy. When does comparison with an untreated product in triple difference strategy improve accuracy?
