300590 BGHS Method Class "Causal Inference from Oberservational Data" (S) (SoSe 2020)

Most research questions in the social sciences are causal questions. Experiments are widely considered as the gold standard for drawing causal inference because of the manipulation of the treatment and the random assignment to the treatment groups. Yet, experiments are not feasible for many research questions. Therefore, researchers are often faced with a causal question and non-experimental data at hand.

This course introduces the approach of modern causal analysis as an attempt to causal inference from observational data. The course addresses three key topics:
First, it introduces the idea of causality based on the potential outcome framework by Donald Rubin. Here, we will learn to differentiate between crucial definitions of causal effects and to pose properly formulated causal research questions.
Second, it introduces directed acyclic graphs (DAGs) as a simple and straightforward tool to guide causal model building. Here, we will learn the difference between (self-)selection processes into the X-variable of interest and understand issues of endogenous selection bias, common-cause confounding and over-control bias.
Third, it discusses methodological approaches to estimate causal effects from observational data based on the considerations of topic 1 and 2. This part mostly draws on conventional multiple-linear regression models with a focus on control variables. Here, we will understand which control variables to include and which control variables to leave out in our models to disentangle a causal effect.