8 Difference in differences

8.1 Introduction

Nov 5. Slides. No reading is required, but for reference see Card & Krueger 1994

Today we study the effect of a policy change in New Jersey, drawing on evidence from the neighboring state of Pennsylvania.

Difference in difference is an identification strategy to be used when one or more units become treated at some time point while others do not. If we believe an assumption of parallel trends, then we can use the change over time for the never-treated units to estimate the change over time that would have been experienced by the units who become treated, in a counterfactual world where they had not become treated.

8.2 Lab

Nov 6 Slides. Download the R Markdown file here. You can find the knitted version here.

In lab, you will implement a difference in difference estimator in a specific setting. The example comes from Malesky, Nguyen, & Tran 2014 and will most closely follow the re-analysis of these data by Egami & Yamauchi 2023.

8.3 Extensions of DID

Nov 7. Slides. No reading is required, but for reference see Egami & Yamauchi 2023 Slides with annotations from lecture: Annotated Slides.

How do we know when the parallel trends assumption holds? And what do we do if it does not hold? This class discusses recent extensions of the DID framework.