Greg Faletto

cran.r-project.orgfetwfe: Fused Extended Two-Way Fixed EffectsCalculates the fused extended two-way fixed effects (FETWFE) estimator for unbiased and efficient estimation of difference-in-differences in panel data with staggered treatment adoption. This estimato...

Version 1.10 of {fetwfe}, my package for difference-in-differences under staggered adoptions, is now available on CRAN: cran.r-project.org/web/packages... Blog post: gregoryfaletto.com/2026/05/23/n... Brief explainer on the package, and description of the updates, in the thread below.

Greg Faletto@gregoryfaletto.com

Interested in doing some causal inference over the long weekend? Version 1.5.0 of the {fetwfe} package for difference-in-differences under staggered adoptions is now on CRAN! Details on what's new below: cran.r-project.org/web/packages...

Hello! I made a skill for R package development. It's a multiagent system running an end-to-end workflow for converting a GitHub issue into a PR solving the issue, with subagents for planning, implementation, and review. github.com/gregfaletto/... 🧵

github.comgithub.com

gregoryfaletto.comNew Multi-Agent Coding Skill for Developing Statistical R Packages – Gregory Faletto

There are instructions for installation and use on GitHub. I've been using the skill for a while to develop my {fetwfe} and {cssr} packages. If you check it out, I'd appreciate your feedback! Here's a blog post that explains a bit more about the skill: gregoryfaletto.com/2026/09/30/n...

Greg Faletto@gregoryfaletto.com

Finally, the package also supports {broom} functions like `tidy`, `glance`, and `augment` for easier interactions with other tidyverse packages.

github.comIssues · gregfaletto/fetwfePackageImplementation of fused extended two-way fixed effects (Faletto 2024) - Issues · gregfaletto/fetwfePackage

Please let me know if you have any questions or feedback, and feel free to create an issue at github.com/gregfaletto/... if you notice any bugs or want a new feature.

Greg Faletto@gregoryfaletto.com

There are also expanded features for statistical inference. P values and confidence intervals are available for all estimated treatment effects, and there's a "selected" binary column indicating whether the cohort ATTs were estimated as exactly 0--which serves as an asymptotic 100% CI.

Cluster-robust standard errors are also now implemented. It is also worth noting that this update fixes errors in the previous package versions in the standard error calculations. This affects both point estimates and p-values.

Greg Faletto@gregoryfaletto.com

FETWFE is a middle ground. It regularizes the coefficients in the ETWFE regression towards each other, so that if e.g. within a cohort the treatment effect is the same in time periods t and t + 1, we only estimate one TE for those two time periods, preserving unbiasedness with better efficiency.

What's new in version 1.10.0 compared to the previous CRAN release: first of all, the package now provides convenience functions for event study estimates and plots.

link.springer.comTwo-way fixed effects, the two-way mundlak regression, and difference-in-differences estimators - Empirical EconomicsI derive a result on the equivalence between the two-way fixed effects (TWFE) estimator and an estimator obtained from a pooled ordinary least squares regression that includes unit-specific time avera...

Greg Faletto@gregoryfaletto.com

One solution for this, extended two-way fixed effects (Wooldridge 2025), solves this problem with an extended regression that estimates a separate regression coefficient for the treatment effect for each cohort at each time since starting treatment. link.springer.com/article/10.1...

This works, and is unbiased. But maybe all of these separate TEs are overkill, and some of them might actually be equal (like the ATT for the same cohort in adjacent time periods). If so, ETWFE is inefficient (if you speak econometrics) or has a poor bias/variance tradeoff (if you speak ML).

Greg Faletto@gregoryfaletto.com

Briefly, under staggered adoptions, the two-way fixed effects estimator is biased for the average treatment effect on the treated units. It turns out that estimating a single coefficient for the treatment effect on all cohorts at all times doesn't aggregate into a sensible estimate of the ATT.

arxiv.orgMIRAGE: The Illusion of Visual UnderstandingMultimodal AI systems have achieved remarkable performance across a broad range of real-world tasks, yet the mechanisms underlying visual-language reasoning remain surprisingly poorly understood. We r...

26 Apr

Gonna throw a fit if I don't get 48x points

Greg Faletto reposted

Grace

@gracekind.net

30 Mar

As many standardized-test-takers know, there are two ways to score highly on a test: learn the material, or guess what the question-writer is thinking. LLMs have gotten really good at the second part!

Grace@gracekind.net

There’s some cool recent research on this phenomenon! It turns out vision language models excel at image benchmarks *even when the actual images aren’t provided,* because the answers are implicit in the questions! arxiv.org/abs/2603.21687

arxiv.orgTabICLv2: A better, faster, scalable, and open tabular foundation modelTabular foundation models, such as TabPFNv2 and TabICL, have recently dethroned gradient-boosted trees at the top of predictive benchmarks, demonstrating the value of in-context learning for tabular d...

23 Mar

Oh wow, this is how I'm finding out the transformers finally came for IID data

Gaël Varoquaux@gaelvaroquaux.bsky.social

The state of the art for learning or tabular data is: arxiv.org/abs/2602.11139 It comes with a high-quality software implementation: tabicl.readthedocs.io/en/latest/

Greg Faletto reposted

brennan

@brennan.computer

14 Mar

man uses GPT to guide him through sequencing his dying dog's DNA, identifying the mutated genomes, and developing a bespoke cancer mRNA vaccine www.theaustralian.com.au/business/tec...

27 Feb

Can we all agree that naming your ice cream flavor a multisyllable made-up word that includes an intentional misspelling of "mocha" is insufficient notice that it contains a nontrivial amount of caffeine

23 Feb

"We are projecting onto an astral plane of heretofore unknown quality, with perfect genius that God himself could ne'er have foreseen." --an LLM after a couple rounds of iteration to get code that actually runs and does what it's supposed to do

8 Feb

It sort of feels like if you pay $8 million for a 30 second Super Bowl ad, they should throw in the rights to refer to it as the Super Bowl instead of the "big game"

@eugenevinitsky.bsky.social

23 Jan

So it seems like one of the highest-leverage things you can do is learn one of VS Code/Positron/Antigravity/Cursor/Windsurf. Because then you pretty much know the rest of them too

Greg Faletto reposted

Eugene Vinitsky 🍒

8 Jan

As a person who does believe capable AI risks are real, it’s frustrating to see the community not exactly behaving responsibly. If you can’t find it in yourself to say or do something about a literal, present, AI harm (Grok being used for abuse) what is the point of you

Greg Faletto reposted

Paul Goldsmith-Pinkham

@paulgp.com

2 Dec 2025

Please enjoy this blog post based on my colleague Kyle Jensen's talk surveying different approaches to AI coding abilities paulgp.com/ai-coding/20...

Greg Faletto reposted

Julia M. Rohrer

@dingdingpeng.the100.ci

19 Nov 2025

Confidence interval discussion time! The perfect opportunity to repost this blog post answering the question you haven’t dared to ask: www.the100.ci/2024/12/05/w...

the100.ciWhy you are not allowed to say that your 95% confidence interval contains the true parameter with a probability of 95%A shibboleth is a custom, such as a choice of phrasing, that distinguishes one group of people from another. The term goes back to the Hebrew Bible, in which the inhabitants of Gilead identify members...

Darren Dahly@statsepi.bsky.social

This is an education problem, not a tool problem; and we don't want people simply moving from thinking p-values are magic to thinking confidence intervals are.

Greg Faletto reposted

rainbowR

@rainbowr.org

6 Nov 2025

🌈 The website for the inaugural rainbowR conference is now live! 🎉 🗓️ Save the date: Feb 25th-26th, 2026 📣 Call for submissions is open 👯 We're bringing together LGBTQ+ R users to promote our work and foster connections among community members conference.rainbowr.org

conference.rainbowr.org– rainbowR conference

Greg Faletto reposted

rainbowR

@rainbowr.org

22 Sept 2025

📣We're hosting a rainbowR conference, to connect and promote LGBTQ+ people in the #RStats community and to showcase examples of working with LGBTQ+ data 🌈🎉 To help us plan, if you'd be interested in attending or speaking, please fill in this very short form (1-2 mins) docs.google.com/forms/d/1Tx0...

docs.google.comInterested in a rainbowR Conference?rainbowR (https://rainbowr.org) is hosting a virtual conference (tentative date: February 26, 2026). The aim of the conference is to connect and promote LGBTQ+ people in the R community, and to showca...

trends.google.comGoogle TrendsExplore search interest for help with mortgage by time, location and popularity on Google Trends

16 Sept 2025

Huh, the spike is actually March 2009, well after the worst of the financial crisis (Lehman Brothers declared bankruptcy in September 2008) trends.google.com/trends/explo...

This post is unavailable.

21 Aug 2025

Starting to think a big divide in whether or not people like AI is: "do you do the kind of work where it's much easier to check and fix an 80% reliable solution than it is to just do it yourself from scratch?"

chatgpt.comChatGPT - Alt Text WriterA conversational AI system that listens, learns, and challenges

18 Aug 2025

I made a GPT for generating alt text. I made it because I couldn't find one that works the way this one does: you just upload an image and press send and it generates alt text. It follows the W3C Web Accessibility Initiative recommendations. chatgpt.com/g/g-6820994e...

18 Aug 2025

No biggie, but feel free to use it, and if you like it leave a review!

7 Aug 2025

...so when do I get to try GPT-5? 😥

9 July 2025

The way the Bluesky notifications count works, sometimes I log on and think I must have been canceled but it turns out I just got 4 likes and a response on my post

If you're comparing different packages for difference-in-differences under staggered adoptions, I've created two functions to help: * attgtToFetwfeDf() takes a data set that is formatted for did::att_gt() and converts it to the format needed for the methods in {fetwfe}.

5 July 2025

* etwfeToFetwfeDf() does the same for data sets formatted for etwfe::etwfe(). Please let me know if you have any questions or feedback!

Second, if you want to compare these estimators, there's now a set of functions that make it easy to conduct simulation studies. Check out the new vignette for details. cran.r-project.org/web/packages...

gregoryfaletto.comNow on CRAN: {fetwfe} version 1.5.0, with new features – Gregory Faletto

5 July 2025

Finally, there's now an S3 class for the output of fetwfe(), which works with functions like print() and summary(). You'll notice the output looks nicer now. See, for example, the getting started vignette cran.r-project.org/web/packages...

I wrote a blog post with more details on what's new, but I'll talk about the basics here: * New estimators * Workflow for simulation studies * More convenience functions, nicer output gregoryfaletto.com/2025/07/03/n...

5 July 2025

First of all, I added an implementation of extended two-way fixed effects (@jmwooldridge.bsky.social 2021), etwfe(), with inputs and outputs aligned with fetwfe(). There's also now betwfe(), which implements a bridge-penalized (includes lasso and ridge regression) version of etwfe().

28 June 2025

this rules. Haven't taken the plunge to pay for Gemini access, been making the most of my free daily 2.5 Pro prompts for math. The CLI preview access is a super helpful workaround. I just put my prompt in a text file along with the pdf I had uploaded in a question I had asked in the web interface.