iiinova — where your digital self keeps evolving

Reposted by Guilherme Duarte

Julian Gerez @juliangerez.com · 24/10/2025

Fun exercise for students in a causal inference class: draw a DAG for this debate.

Guilherme Duarte @guilhermeduarte.bsky.social · 26/10/2025

Interesting that out of the MHE's bag of tricks approach, the only design that definitely identifies a basic interventional estimand such as the ATE is SOB. Imbens & Angrist's IV identifies the CACE, RDDs identify cutoff estimands, and DID identifies ATT, but none of them identifies the ATE.

110

Guilherme Duarte @guilhermeduarte.bsky.social · 26/10/2025

If you have a massive disagreement on which DAGs were valid, but you don't use or get DAGs, then the work to solve those disagreement is just herculean

000

Guilherme Duarte @guilhermeduarte.bsky.social · 11/02/2025

Interesting response by @yiqingxu.bsky.social and others to @urisohn.bsky.social: arxiv.org/pdf/2502.05717 Causal inference is serious job. In Pearl's parlance, "define first, identify second, estimate last". If the 2 first parts are correct, one can use adaptive models in semiparametric fashion.

arxiv.org

260

Reposted by Guilherme Duarte

Giuseppe Cavaliere @cavalieregiu.bsky.social · 29/01/2025

A very cool Econometrics Journal editorial by @jaapabbring.bsky.social, @victorchernozhukov.bsky.social & Fernandez-Val on Wright's 1928 contribution to causal inference and IV. Very interesting stuff! Link: arxiv.org/abs/2501.16395

2245

Reposted by Guilherme Duarte

Pedro Sant’Anna @pedrosantanna.bsky.social · 30/12/2024

Here are the first five sets of slides: 01 Introduction: psantanna.com/DiD/01_Intro... 02 Classical 2x2 setup: psantanna.com/DiD/02_two_b... 03 Clustering issues: psantanna.com/DiD/03_Clust... 04 Functional form: psantanna.com/DiD/04_Funct... 05 Covariates: psantanna.com/DiD/05_Covar...

psantanna.com

50731181

Guilherme Duarte @guilhermeduarte.bsky.social · 18/12/2024

Obrigado mesmo, Eduardo. Aprendi muito com os seus papers.

100

Guilherme Duarte @guilhermeduarte.bsky.social · 17/12/2024

Obrigado mesmo Jamil. Fiquei ausente por um tempo por causa do market. Mas logo logo vamos tomar um café

010

Guilherme Duarte @guilhermeduarte.bsky.social · 17/12/2024

Thank you so much Melody. And thanks a lot for your help during this process

010

Guilherme Duarte @guilhermeduarte.bsky.social · 17/12/2024

Thank you so much Mike. It was really great to meet you at Polmeth

010

Guilherme Duarte @guilhermeduarte.bsky.social · 17/12/2024

Congrats Anton! This is excellent for you and for Madison!

020

Guilherme Duarte @guilhermeduarte.bsky.social · 17/12/2024

I am thrilled to announce that I will be joining Department of Government at Harvard University, first as a postdoctoral fellow (2025) and then as an assistant professor (2026). I am grateful and really excited for this new opportunity.

8722

Guilherme Duarte @guilhermeduarte.bsky.social · 04/12/2024

Just did, Claudia. Good to see you here.

000

Reposted by Guilherme Duarte

Edward H. Kennedy @edwardhkennedy.bsky.social · 22/11/2024

New paper! arxiv.org/pdf/2411.14285 Led by amazing postdoc Alex Levis: www.awlevis.com/about/ We show causal effects of new "soft" interventions are less sensitive to unmeasured confounding & study which effects are *least* sensitive to confounding -> makes new connections to optimal transport

35914

Reposted by Guilherme Duarte

Noah Greifer @noahgreifer.bsky.social · 16/11/2024

#Rstats `MatchIt` v4.6.0 is out! `MatchIt` implements propensity score matching and other matching methods for causal effect estimation. This isn't a major release, but here are the main updates: 🧵 #causalsky #econsky #episky #statsky

520533

Guilherme Duarte @guilhermeduarte.bsky.social · 15/11/2024

A clear example is the Russian Roulette case proposed by Anders Huitfeldt, and then studied by Pearl and Cinelli (2021). Unfortunately, Anders, Carlos Cinelli, and Pearl don't seem to be here on bsky

020

Guilherme Duarte @guilhermeduarte.bsky.social · 15/11/2024

In fact, no regression method can ensure external validity by itself. You need structural and sometimes functional assumptions. I feel like people in causal inference usually use the CATE invariance assumption. But in many cases this is not 100% guaranteed.

140

Reposted by Guilherme Duarte

Richard Everitt @bayesianstats.bsky.social · 15/11/2024

I made a starter-pack for Statistics and Statistics-related groups, departments or organisations. Please share, and suggest accounts that I have missed. go.bsky.app/q6MfWL

76423

Reposted by Guilherme Duarte

Aaron Roth @aaroth.bsky.social · 15/11/2024

Econ has some ML gems. One of my favorite's is Sandroni's sweeping result that no empirical test can distinguish an informed from an uninformed forecaster. I teach it in my ML class: www.youtube.com/watch?v=7OAI... But it is presented as negative, when it is in fact a sweeping positive result.

youtube.com

CIS 6200: Learning with Conditional Guarantees, Lecture 21.

YouTube video by Aaron Roth

29215

Reposted by Guilherme Duarte

apoorva lal @handle.invalid · 12/11/2024

(brand new, WIP) python package for synthetic control estimators with a fast weight solver (pyensmallen). Currently implements jacknife CIs since inference in single-treated setting is basically made up anyway. hope this passes muster, @paulgp.com ? github.com/apoorvalal/s...

github.com

GitHub - apoorvalal/synthlearners: fast synthetic control estimators for panel data problems

fast synthetic control estimators for panel data problems - apoorvalal/synthlearners

54914

Guilherme Duarte @guilhermeduarte.bsky.social · 05/11/2024

I guess I am a DAG person. Also an ADMG person.

120

Reposted by Guilherme Duarte

Adam L @adam-lg.bsky.social · 22/10/2024

"Recent Developments in Partial Identification" by Kline and Tamer (2023). #stats 📉📈 For those curious about Manski bounds and partial identification more generally, a nice review! Open access: www.annualreviews.org/content/jour...

Identification strategies concern what can be learned about the value of a
parameter based on the data and the model assumptions. The literature on
partial identification is motivated by the fact that it is not possible to learn the
exact value of the parameter for many empirically relevant cases. A typical
result in the literature on partial identification is a statement about char-
acterizing the identified set, which summarizes what can be learned about
the parameter of interest given the data and model assumptions. For in-
stance, this may mean that the value of the parameter can be learned to be
necessarily within some set of values. First, the review surveys the general
frameworks that have been developed for conducting a partial identifica-
tion analysis. Second, the review surveys some of the more recent results on
partial identification.

0147

Guilherme Duarte @guilhermeduarte.bsky.social · 21/10/2024

Sure. I have to update it. Let me know your recs.

210

Reposted by Guilherme Duarte

Anton Strezhnev @astrezh.bsky.social · 18/09/2024

The only part of Twitter worth preserving are those Wooldridge threads on Poisson QMLE.

2271

Guilherme Duarte @guilhermeduarte.bsky.social · 01/09/2024

I wonder who is going to be the first person on this website to claim python is better than R.

210

Guilherme Duarte @guilhermeduarte.bsky.social · 01/09/2024

Hahahaha. You're already in the starter pack bsky.app/profile/guil...

110

Reposted by Guilherme Duarte

Marcelo Soares @msoares.bsky.social · 01/09/2024

🚨GRAVE: usuário do Bluesky cria lista que permite bloquear de uma vez uma penca de perfis caça-clique que espalham fofoca, exagero, inutilidade e desinformação. bsky.app/profile/did:...

1602214585

Reposted by Guilherme Duarte

Amanda Weiss @amandaweiss.bsky.social · 31/08/2024

I, of course, would never pander to the delightful new community of folks joining Bluesky by pointing out that Brazilian mayoral elections are a major use case for difference-in-differences methods.

1142

Guilherme Duarte @guilhermeduarte.bsky.social · 01/09/2024

Only be arriving in the US on tuesday, so I got only bluesky to use.

000

Guilherme Duarte @guilhermeduarte.bsky.social · 01/09/2024

A starter pack with causal inference I know. Unsure if this is the correct way of doing this, but I'm learning go.bsky.app/FdemGAZ

36427

Guilherme Duarte @guilhermeduarte.bsky.social · 01/09/2024

That is indeed a pretty good book. I bought a physical copy.

010

Guilherme Duarte @guilhermeduarte.bsky.social · 01/09/2024

Yeah, that's an amazing book. I was just worried that it might not be free. Thanks for that.

110

Guilherme Duarte @guilhermeduarte.bsky.social · 01/09/2024

I wonder if I forgot any book. I feel like free books on causal discovery / algorithmic ID are sort of non-existent.

210

Guilherme Duarte @guilhermeduarte.bsky.social · 01/09/2024

Judea Pearl's causality is not available online, but as the book is based on his papers, you can find them online bayes.cs.ucla.edu/BOOK-2K/. For instance, chapter 8 is extracted from Balke and Pearl (94, 97) and Chickering and Pearl (96).

bayes.cs.ucla.edu

CAUSALITY, 2nd Edition, 2009

140

Guilherme Duarte @guilhermeduarte.bsky.social · 01/09/2024

Now from an identification point-of-view, we have Hernan-Jamie Robins' textbook 'What if'. This is an amazing bible that covers a lot of graphical models, including SWIGs. www.hsph.harvard.edu/miguel-herna...

hsph.harvard.edu

151

Guilherme Duarte @guilhermeduarte.bsky.social · 01/09/2024

Finally, another resource with a strong focus on incorporating ML models are these causal inference notes by Stefan Wager. web.stanford.edu/~swager/stat...

web.stanford.edu

120

Guilherme Duarte @guilhermeduarte.bsky.social · 01/09/2024

The second free resource is the recent book by Victor Chernozhukov, Hansen, Syrgkanis, and others. www.causalml-book.org Amazing book if you are interested in Double ML, or incorporating ML models in the estimation of causal effects with root-n consistency.

causalml-book.org

CausalML

Applied Causal Inference Powered by ML and AI. Victor Chernozhukov, Christian Hansen, Nathan Kallus, Martin Spindler, Vasilis Syrgkanis.

250