Applications of statistical causal inference in software engineering

Context: The aim of statistical causal inference (SCI) methods is to estimate causal effects from observational data (i.e., when randomized controlled trials are not possible). In this context, Pearl's framework based on causal graphical models is an approach that has recently gained popularity and allows for explicit reasoning about issues related to spurious correlations.Objective: Our primary goal is to understand to which extend and how Pearl's graphical framework is applied in software engineering (SE). Methods: We performed a systematic mapping study and analysed a total of 25 papers published between 2010 and 2022. Results: Our results show that the application of Pearl's SCI framework in SE is relatively recent and that the corresponding research community is fragmented. Most of the selected papers focus on software quality analysis. There is no clear and widespread community of practice (yet) on how to implement and evaluate SCI in SE.Conclusions: To the best of our knowledge this is the first time such a mapping study is done. We believe that SE practitioners might benefit from such a work, as it both provides an overview of the work and people involved in the application of causal inference methods, but also outlines the potential and limitations of such approaches.