Pedro H. C. Sant'Anna Profile
Pedro H. C. Sant'Anna

@pedrohcgs

Followers
25K
Following
18K
Media
385
Statuses
4K

Associate Professor @EmoryEconomics | Econometrics | Causal Inference | Difference-in-Differences. Dad 5x

Atlanta, GA
Joined June 2012
Don't wanna be here? Send us removal request.
@pedrohcgs
Pedro H. C. Sant'Anna
1 month
We have some updates on our DDD paper. In our latest version, we have introduced three new applications and an open-source R package to facilitate the usage of all our DDD tools!. 🚀R package: 🔎Updated paper:
Tweet card summary image
arxiv.org
Triple Differences (DDD) designs are widely used in empirical work to relax parallel trends assumptions in Difference-in-Differences (DiD) settings. This paper highlights that common DDD...
@pedrohcgs
Pedro H. C. Sant'Anna
3 months
🚨New paper alert🚨. My paper with @marcelortizv on "Better Understanding Triple Differences Estimators" is finally out. 📰 Main msg: Nuances in DDD designs challenge the practice of generally viewing DDD as the diff btw two DiDs. We can do better!. 1/
Tweet media one
6
65
308
@pedrohcgs
Pedro H. C. Sant'Anna
1 day
If you want some DiD-related stickers, just ask me when you see me! . A new batch has just arrived!. #rstats #Econometrics #causalinference
Tweet media one
6
5
145
@pedrohcgs
Pedro H. C. Sant'Anna
3 days
I was chatting yesterday and that convo reminded me of this paper by Carone, Luedtke, and van de Laan: I really like the writing style, too!!!. Sharing in case you enjoy efficient inference procedures like I do!
Tweet media one
1
0
16
@pedrohcgs
Pedro H. C. Sant'Anna
1 month
PolMeth is now among my favorite conferences!!!.
0
0
15
@pedrohcgs
Pedro H. C. Sant'Anna
1 month
Here we go for another week full of GenAI, a DiD seminar, and a DiD workshop!. For those attending PolMeth at @EmoryUniversity: looking forward to connecting!.
1
0
17
@pedrohcgs
Pedro H. C. Sant'Anna
1 month
But also that sometimes we can further robustify important findings by dropping comparison groups that could be harder to motivate (like never-treated units)
Tweet media one
2
0
5
@pedrohcgs
Pedro H. C. Sant'Anna
1 month
It also highlights that sometimes our methods lead to different conclusions than using too-rigid regression specifications
Tweet media one
1
0
2
@pedrohcgs
Pedro H. C. Sant'Anna
1 month
Aligned wiht our results, our applications (and simulations) highlight that we can provide much more informative inference procedures (🥰 smaller confidence intervals 🥰 ---Figure 5)
Tweet media one
1
2
5
@pedrohcgs
Pedro H. C. Sant'Anna
2 months
RT @ecmaEditors: Econometric analysis typically ignores researcher incentives. This article frames experimental data analysis as a mechanis….
0
49
0
@pedrohcgs
Pedro H. C. Sant'Anna
2 months
What an fun and productive week!. Co-teaching with @jmwooldridge and having the help of @SofiaTsarsi and @ianLal with the labs was just awesome!.
@SofiaTsarsi
Sofia Tsarsitalidou
2 months
The econometricians, Athens edition. @pedrohcgs @jmwooldridge
Tweet media one
4
2
101
@pedrohcgs
Pedro H. C. Sant'Anna
2 months
Thank you, @BeatrizGietner, for covering our paper!!!.
@BeatrizGietner
Bia
2 months
We have a new post 🙃.Three recent DiD papers, one shared goal: better estimates in messy settings.☑️DiD that hits efficiency bounds.☑️Synthetic DiD that mimics an oracle.☑️TWFE vs. the new wave.👉
Tweet media one
1
5
42
@pedrohcgs
Pedro H. C. Sant'Anna
2 months
RT @jmwooldridge: It’s a real pleasure team teaching a course with ⁦@pedrohcgs⁩. Many are referring to us as “Fire and Ice.” .
0
12
0
@pedrohcgs
Pedro H. C. Sant'Anna
2 months
We hope this paper will be helpful for you!. We will work on packaging everything so you can adopt all this seamlessly. The paper is here: Thanks!!!. 17/17.
Tweet card summary image
arxiv.org
This paper investigates efficient Difference-in-Differences (DiD) and Event Study (ES) estimation using short panel data sets within the heterogeneous treatment effect framework, free from...
1
2
11
@pedrohcgs
Pedro H. C. Sant'Anna
2 months
All in all, we have attempted to gain a deeper understanding of the role of time and comparison groups in DiD setups. This enabled us to provide more informative DiD and ES estimators. In practice, this matters if you care about getting short confidence intervals. 16/.
1
0
3
@pedrohcgs
Pedro H. C. Sant'Anna
2 months
Finally, we also discuss how to assess the stability of our estimators using a type of "specification curve" that plots all possible (non-collinear) event-study constructions. In our application, the ES coefficients are very stable, which is great news!. 15/
Tweet media one
1
0
3
@pedrohcgs
Pedro H. C. Sant'Anna
2 months
We expect that people will want to understand how we achieve such gains in precision, and the key is to effectively combine different pre-treatment and comparison groups in a non-uniform way. You can visualize these using our proposed visualization tools. 14/
Tweet media one
1
0
2
@pedrohcgs
Pedro H. C. Sant'Anna
2 months
We also have an empirical illustration showing that these gains in precision arise in real applications, and not in simulations. In our application, we highlight that other methods can require 50% more data to achieve the same precision as our efficient DiD procedure. 14/
Tweet media one
1
0
2
@pedrohcgs
Pedro H. C. Sant'Anna
2 months
We also conducted calibrated simulations using staggered setups with Compustat data and demonstrated that gains in efficiency are also important in this context. The gains in efficiency vary according to the degree of correlation between outcomes over time. 13/
Tweet media one
1
0
3