Last updated: 20 August 2021
This repository tracks the recent developments in the Difference-in-Difference (DiD) literature. Currently, it is just a dump of my bookmarks from different websites including Twitter, GitHub, YouTube etc. This will be sorted out over time as the literature converges to some consensus. But this might still take a while.
This is a working document, if you want to contribute, just e-mail, open an issue or start a discussion on GitHub. Since paths and links are also subject to change, please report these changes so this repository is kept as up to date as possible.
TO BE ADDED...
Treatments across different units can occur in various configurations. The figures below some of the most common treatment types. Here the x-axis represents the time scale, the y-axis the different groups (panel IDs) and the orange represents the time periods in which the units are treated. The figures below show types of differential treatment timings:
Some package paths have been split across lines but adding spaces to keep table formatting intact. Just make sure they are in one line when copying them in the Stata window or dofile.
For individual packages, check their helpfiles for example code.
For using and plotting multiple DiD packages in Stata, the event_plot
command (ssc install event_plot, replace
) by Kirill Borusyak is highly recommended. It estimates and combines results from five different estimators. Example of how to do event study plots using different packages is given in the five_estimators_example.do dofile on GitHub.
The event_plot
usage example has been extended twice:
-
David Burgherr
has a dofile on Dropbox.
-
Pietro Santoleri
has a dofile on GitHub that plots seven different estimators.
Scott Cunningham has sample dofiles as part of the CodeChella DiD event.
Packages installation paths have been split across lines to preverse table formatting in Markdown.
This is still being updated. See links below for details. There is also a Julia DiffinDiff group on GitHub which contains various packages.
Name | Installation | Package by | Reference paper |
---|---|---|---|
InteractionWeightedDIDs.jl | GitHub | ||
SynthControl.jl | GitHub |
Papers are in alphabetical order by last name. Papers without journals are pre-prints. Please click on paper links for details. SORTABLE TABLE TO BE ADDED.
Dmitry Arkhangelsky , Guido Imbens, Lihua Lei
, Xiaoman Luo (2021). Double-Robust Two-Way-Fixed-Effects Regression For Panel Data.
Susan Athey, Guido Imbens (2006). Identification and inference in nonlinear difference-indifferences models. Econometrica.
Kirill Borusyak , Xavier Jaravel
, Jann Spiess
(2021). Revisiting Event Study Designs: Robust and Efficient Estimation.
Brantly Callaway, Andrew Goodman-Bacon, Pedro H.C. Sant'Anna (2021). Difference-in-Differences with a Continuous Treatment.
Brantly Callaway, Pedro H.C. Sant'Anna (2020). Difference-in-Differences with multiple time periods, Journal of Econometrics.
Clément de Chaisemartin, Xavier D'Haultfoeuille (2018). Fuzzy differences-in-differences. The Review of Economic Studies.
Clément de Chaisemartin, Xavier D'Haultfoeuille (2020). Two-Way Fixed Effects Estimators with Heterogeneous Treatment Effects. American Economic Review.
Clément de Chaisemartin, Xavier D'Haultfoeuille (2021). Two-way fixed effects regressions with several treatments.
Clément de Chaisemartin, Xavier D'Haultfoeuille (2021). Difference-in-Differences Estimators of Inter-temporal Treatment Effects.
Xavier D’Haultfoeuille, Stefan Hoderlein, Yuya Sasaki (2013). Nonlinear difference-indifferences in repeated cross sections with continuous treatments.
Xavier D’Haultfoeuille, Stefan Hoderlein, Yuya Sasaki (2021). Nonparametric Difference-in-Differences in Repeated Cross-Sections with Continuous Treatments.
Bruno Ferman , Cristine Pinto (2021). Synthetic Controls with Imperfect Pre-Treatment Fit. Quantitative Economics.
Simon Freyaldenhoven, Christian Hansen, Jesse M. Shapiro (2019). Pre-event Trends in the Panel Event-Study Design. American Economic Review.
Hans Fricke (2017). Identification based on difference-in-differences approaches with multiple treatments. Oxford Bulletin of Economics and Statistics.
John Gardner (2021). Two-stage differences in differences.
Andrew Goodman-Bacon (2021). Difference-in-differences with variation in treatment timing. Journal of Econometrics.
Jonathan Roth (2021). Pre-test with Caution: Event-study Estimates After Testing for Parallel Trends.
Jonathan Roth , Pedro H.C. Sant'Anna
(2021). Efficient Estimation for Staggered Rollout Designs.
Pedro H.C. Sant'Anna , Jun Zhao (2020). Doubly robust difference-in-differences estimators, Journal of Econometrics.
Liyang Sun, Sarah Abraham (2020). Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. Journal of Econometrics.
Scott Cunningham (2020). Causal Inference: The Mix Tape.
Nick Huntington-Klein (2021). The Effect.
Here are people who are actively involved in curating information on the latest DiD developments. This includes blogs, lecture series, tweets.
Pedro H.C. Sant'Anna . Seminar on DiD on 24-27 August, 2021 organized by Statistical Horizons
is now open for registeration.
Scott Cunningham : CodeChella the ultimate DiD event Workshop 1: Friday July 16th, 2021 and Workshop 2: Friday July 23, 2021 which will be live on Twitch. The videos from the workshop are now up on YouTube.
Chloe East organizes an online DiD reading group.
Taylor J. Wright organizes an online DiD reading group. The lecture recordings can also be viewed on YouTube.
Scott Cunningham : Scott's Substack is the goto place for an easy-to-digest explanation of the latest metric-heavy DiD papers.
Andrew C. Baker has notes on Difference-in-Differences Methodology with supporting material on GitHub.
Paul Goldsmith-Pinkham has a brilliant set of lectures on empirical methods including DiD on GitHub. These are also supplemented by YouTube videos.
Jeffrey Wooldridge has made several notes on DiD which are shared on his Dropbox including Stata dofiles.
Fernando Rios-Avila has a great explainer for the Callaway and Sant'Anna (2020) CS-DID logic on his blog.
Christine Cai has a working document which lists recent papers using different methods including DiDs.
These (related) interactive R-Shiny dashboards showcase how TWFE models give wrong estimates.
Kyle Butts : https://kyle-butts.shinyapps.io/did_twfe/
Hans Henrik Sievertsen : https://hhsievertsen.shinyapps.io/kylebutts_did_eventstudy/
Some interesting Twitter threads in no particular sequence. In order to render these properly, you need to view them on the Jekyll website.
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>The diff-in-diff literature is plagued with the notion that more controls are always better (as long as the treatment effect has 2 stars). Someone please write a version of this @yudapearl paper aimed at the assumptions & examples of DiD. #EconTwitter https://t.co/JfxQHGlxgw
— Arthur Lewbel (@lewbel) July 29, 2021
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>🚨Hello #EconTwitter! I am very happy that my paper with Brantly Callaway, "Difference-in-Differences with multiple time periods", is now forthcoming at the Journal of Econometrics. https://t.co/zoNxNY9ugq
— Pedro H. C. Sant'Anna (@pedrohcgs) December 18, 2020
What are the main take aways? I will ask my daughter to help me out.
1/n pic.twitter.com/DNj3Cpxxlu
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>I finally got my TWFE/Mundlak/DID paper in good enough shape to make it an official working paper. I'll put it in other places but it's currently here:https://t.co/D4iQETUSlX
— Jeffrey Wooldridge (@jmwooldridge) August 17, 2021
Also, the Stata stuff is still with the Dropbox link:https://t.co/q1AnkhEF97#metricstotheface
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>In preparing a DID course, I discovered some other neat stuff. When you impose PT and allow flexible interactions with covariates the way I do, the POLS estimator is numerically identical to the comparable imputation estimator in Borusyak, Jaravel, Spiess (2021). @borusyak
— Jeffrey Wooldridge (@jmwooldridge) August 17, 2021
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>More DID stuff, re: parallel trends. If you have two pre-treatment periods there are various equiv ways to test for pre-trends. Here are three (no controls):
— Jeffrey Wooldridge (@jmwooldridge) August 18, 2021
1. Regress cy2 = y2 - y1 on 1, d using t = 2; obtain t test on d.
This is a standard placebo test.#metricstotheface
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>In my career I have spent a lot of time assessing parallel trends in dif-in-dif models and visualizing those; this one is on top of my current list... and developing this was a team effort!
— Arindrajit Dube (@arindube) August 20, 2021
Show placebo estimates for a whole set of placebo dates, each w dynamic treatment effects https://t.co/lxSfekefkY
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>I spent this past week catching up with the DiD literature. Here is my list🧵:
— Jesús Villero (@jotavillero) May 16, 2021
1. Read (if you haven't) Andrew Goodman-Bacon's "Difference-in-Differences with Variation in Treatment Timing." The paper that started all for me (there are others before).Link: https://t.co/GBFjBnHDcj
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>Navigating the DiD revolution from one applied researcher's perspective.
— Matthew A. Kraft (@MatthewAKraft) June 24, 2021
A LONG 🧵 on what I've learned & what I'm still trying to figure out. Advice/insights welcome!
My @michaelpollan 🥦🍅🥕🫑 inspired TL;DR take:
"Apply DiD in context, not every 2x2, mostly event studies" pic.twitter.com/CWmwyo1Btp
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>I've been catching up on staggered diff-in-diff/two-way fixed effects recently. Simulating helped me see how bad TWFE performs with dynamic treatment effects (see those pre-trends). I also tried implementing Sun & Abraham (2020)'s interaction-weighted estimator in Stata pic.twitter.com/FQBCQi0m7d
— Shan Huang (@ShanHuang_ec) June 16, 2020
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>🚨Hello #EconTwitter! I am very happy that my paper with Brantly Callaway, "Difference-in-Differences with multiple time periods", is now forthcoming at the Journal of Econometrics. https://t.co/zoNxNY9ugq
— Pedro H. C. Sant'Anna (@pedrohcgs) December 18, 2020
What are the main take aways? I will ask my daughter to help me out.
1/n pic.twitter.com/DNj3Cpxxlu
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>The diff-in-diff literature is plagued with the notion that more controls are always better (as long as the treatment effect has 2 stars). Someone please write a version of this @yudapearl paper aimed at the assumptions & examples of DiD. #EconTwitter https://t.co/JfxQHGlxgw
— Arthur Lewbel (@lewbel) July 29, 2021
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>This is a really interesting discussion of using potential outcomes to define causal effects. Not sure I agree with the critiques, but they're thought-provoking
— Peter Hull (@instrumenthull) July 26, 2021
key quote: "I don’t like the idea of types defined by something that happens in the future.."https://t.co/AWRK9Qgj4W pic.twitter.com/a1J8pDxfXK
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>As I was re-reading this paper recently, I wanted to draw attention to the fact that this paper by Gardner (2021) derives (in appendix B)( the event-specific weights for each event in a "stacked regression" used in Cengiz et al (2019).
— Arindrajit Dube (@arindube) July 21, 2021
1/https://t.co/ucvjcNtq3Q pic.twitter.com/drVnoE1xHb