Expanding the Data Science Toolkit: Using a Mixed Methods Approach to Improve Causal Inference

Mixed method approaches, which carefully integrate qualitative and quantitative data, are widely used in the social sciences and public health. Yet their potential to strengthen causal inference remains underexplored. This course introduces mixed methods as an innovative extension of the causal inference toolkit. We begin with the challenges of estimating causal effects in non-randomized settings. Many statistical designs and analysis approaches exist, including propensity scores, instrumental variables, newer proximal causal inference methods, etc. However, each involves untestable assumptions that are often hard to understand and think through in applied studies. Traditional tools, such as sensitivity analyses or bounding techniques, only partially address these underlying assumptions. Mixed methods offer a complementary and emerging framework to incorporate "on the ground" content knowledge into study design and analysis. Qualitative methods systematically capture key information from individuals in a structured way to learn more about, for example, how treatments were assigned to different individuals. Participants will learn how qualitative insights can help identify relevant causal questions, clarify mechanisms of action, and assess assumptions such as the plausibility of no unobserved confounding. Through lecture, discussion, and applied exercises including case studies, we discuss how mixed methods can help produce more credible and interpretable causal conclusions.