Applied Statistical Causal Inference in Software Engineering

Description

Software engineering (SE) research is currently subject to a paradigm shift in data analysis practices. Being content with identifying correlations from quantitative data limits our ability to infer valid conclusions that provide decision support to SE practitioners. To address these flaws, SE researchers began to adopt methods for statistical causal inference (SCI) from more mature scientific disciplines like evidence-based medical research, sociology, and econometrics. However, these methods exhibit a steep learning curve which hinders their adoption in SE research. This tutorial aims at gently introducing SE researchers to SCI methods in order to keep up with methodological advances of the field. The tutorial introduces the fundamentals of SCI with applications and examples particular to SE, enabling participants to keep up with the advances of data analysis practices.

Topics:

• Causal modeling: A graphical syntax to frame complex causal hypotheses that can be used to reason about SE phenomena.
• Causal relationships: Types of relationships between variables in hypotheses and their consequences for data analysis.
• d-separation: A systematic technique to mitigate confounders and other sources of bias when analyzing quantitative data.

Learning Outcomes:

By the end of the tutorial, participants will be able to:

• Understand the importance of causal over correlational inference for decision-making.
• Apply causal modeling techniques to frame complex causal hypotheses, and
• Analyze causal models to identify and remove potential confounders.

Ultimately, the tutorial will equip participants with the basic SCI techniques to draw causal conclusions even from observational data, i.e., data not gathered during a controlled experiment.