The Statistics of Causal Inference

We start with experiments and move on to look at ways to approach the experimental ideal with observational data. We'll cover three pathways to causal inference: 

• instrumental variables
• difference-in-differences
• the regression discontinuity design. 

For each method, we will start with intuition and indicative examples, move on to the identification assumptions, and proceed to estimation.

We will go back to applications, this time thinking more about what type of robustness checks we can implement to assess the extent to which we can plausibly expect the identification assumptions to hold. For each design we will also cover recent extensions. 

Each technique will be accompanied by a lab session, where you will work with fellow participants to replicate an existing study. We will provide code in R and in Stata for two examples per method.

How the course will work online 

We will discuss the pre-recorded lectures during the live sessions. Every session will assume you have seen the recording and will be based on questions and discussions on the topics covered. 

Rather than the two-hour lecture that would be happening if this were a face-to-face course, we will provide short videos, each on a specific topic. 

All slides are given one week in advance. The TA will provide support with theory and code. We'll contact you at the end of the second day to discuss how the format is working, and whether you think changes or ad hoc adjustments would be helpful.

We will set up a Slack # for the group, to discuss readings, slides, and code. We may replace this with a Google Colab, but we’ll make this decision after all participants are registered. 

During ‘refreshment sessions’ on Wednesday, Thursday and Friday, the Instructor and TA will meet with participants to discuss their own work and research.

You should have a solid understanding of linear regression and even mechanical knowledge of hypothesis testing at the level of any introductory econometrics textbook.