Design & Analysis of Quasi-Experiments for Causal Inference
In many fields, randomized experiments are often considered the gold standard approach for learning about the causal effects of an intervention, program, or policy. However, randomized experiments are not always feasible or ethical. Furthermore, the increasing availability of large-scale observational datasets presents opportunities to investigate causal effects outside of the realm of designed experiments. This course surveys contemporary research design strategies for investigating questions about causal effects, focusing on the theory and application of quasi-experimental methods that can, under some conditions, provide strong warrants for drawing causal inferences. The focus of the course is on causal description of point-in-time interventions (“Is this intervention effective?”) rather than causal explanation (“Why is this intervention effective?”).
The course begins with an introduction to the potential outcomes framework for expressing causal quantities, followed by an examination of (idealized) simple and block randomized experiments as prototypes for learning about causal effects. The remainder of the course covers theory and data-analysis strategies for drawing causal inferences from four quasi-experimental designs: instrumental variables approaches, regression discontinuity designs, non-equivalent control group designs (using techniques such as matching and propensity score weighting), and comparative interrupted time series designs. For each design, we will consider (i) the core strategy for identifying a causal effect, (ii) corresponding statistical approaches for estimating the effect, and (iii) strategies and design elements for strengthening the design. Further, advanced topics will be covered based on student interest.