Today, we received the news that our (very short) paper on the use of Causal DAGs for designing experiments was accepted to be presented in EASE 2022.

The paper argues that we should design our experiments with the use of DAGs. That improves communication between stakeholders, allows us to verify our domain knowledge with causal discovery checks and even calculate directs and indirect effects that are not possible with just the experiment result (which only gives the total effect).

The paper can be found at: https://arxiv.org/pdf/2204.08743.pdf

The abstract follows:

Randomized field experiments are the gold standard for evaluating the impact of software changes on customers. In the online domain, randomization has been the main tool to ensure exchangeability. However, due to the different deployment conditions and the high dependence on the surrounding environment, designing experiments for automotive software needs to consider a higher number ofrestricted variables to ensure conditional exchangeability. In this paper, we show how at Volvo Cars we utilize causal graphical models to design experiments and explicitly communicate the assumptions of experiments. These graphical models are used to further assess the experiment validity, compute direct and indirect causal effects, and reason on the transportability of the causal conclusions.