Laurent Bergé
LightNews LightNews
banner
lrberge.bsky.social
Laurent Bergé
@lrberge.bsky.social
500 followers 540 following 120 posts
Ass prof in econ at the University of Bordeaux

Too much code.

https://sites.google.com/site/laurentrberge/
Posts Replies Media Videos
Pinned
lrberge.bsky.social
Laurent Bergé @lrberge.bsky.social · Sep 10
#rstats #econsky

fixest v0.13.0 is finally out!

It's still about making OLS and GLM estimations easy.

Some major changes:
- *default* VCOV becomes iid always!
- singletons are removed by default!

See all the changes here:
github.com/lrberge/fixe...
fixest: Fast Fixed-Effects Estimations
Fast and user-friendly estimation of econometric models with multiple fixed-effects. Includes ordinary least squares (OLS), generalized linear models (GLM) and the negative binomial. The core of the p...
cran.r-project.org
Reposted by Laurent Bergé
libbyheeren.bsky.social
Tell me something you do when you code that other people would tell you that you shouldn't do.

Tell me the rules you break!

I'll go first: I work in untitled files in the wrong project directories all the time. Like, all the time. Yes, I do tend to lose things 😂 #databs #rstats #python
September 24, 2025 at 5:52 AM
Reposted by Laurent Bergé
gmcd.bsky.social
`tinyplot` 0.5.0 is now available from CRAN & R-universe. Alongside bug fixes & internal improvements, this release includes some cool new features like bubble chart + dual legend support, element dodging, & ephemeral themes.

Detailed changelog: grantmcdermott.com/tinyplot/NEW...

#rstats #dataviz
tinyplot bubble chart example with dual legend, using penguins data. Taken from https://grantmcdermott.com/tinyplot/vignettes/gallery.html
September 23, 2025 at 2:41 AM
Reposted by Laurent Bergé
srvanderplas.bsky.social
Sarcasm, but actually something brought up by one of the deans, that having statisticians around is *less* important in the age of AI because AI can just do experimental design and data analysis for researchers. No joke. #sendhelp
September 15, 2025 at 6:15 PM
Reposted by Laurent Bergé
emilyriederer.bsky.social
Really insightful post from Julie Tibshirani (spotted in LinkedIn, can't find on Bsky) reflecting on #rstats 's unique governance structure and what can be learned for other languages

jtibs.substack.com/p/if-all-the...
If all the world were a monorepo
The R ecosystem and the case for extreme empathy in software maintenance
jtibs.substack.com
September 14, 2025 at 11:29 PM
Reposted by Laurent Bergé
gmcd.bsky.social
Important note for Mac users: make sure your installation properly exploits `fixest` multi-threading capabilities:

1. Follow @coatless.bsky.social's simple shell script and OpenMP instructions here: github.com/coatless-she...

2. Install from source: `install.packages("fixest", type = "source")`
September 11, 2025 at 4:33 PM
lrberge.bsky.social
#rstats #econsky

fixest v0.13.0 is finally out!

It's still about making OLS and GLM estimations easy.

Some major changes:
- *default* VCOV becomes iid always!
- singletons are removed by default!

See all the changes here:
github.com/lrberge/fixe...
fixest: Fast Fixed-Effects Estimations
Fast and user-friendly estimation of econometric models with multiple fixed-effects. Includes ordinary least squares (OLS), generalized linear models (GLM) and the negative binomial. The core of the p...
cran.r-project.org
September 10, 2025 at 7:34 PM
lrberge.bsky.social
The EEA conference starts tomorrow in Bordeaux!
I'll be around, so if you want to meet, just reach out!

#rstats #econsky

www.eeassoc.org/events/eea-b...
LinkedIn
This link will take you to a page that’s not on LinkedIn
lnkd.in
August 24, 2025 at 8:17 PM
Reposted by Laurent Bergé
gmcd.bsky.social
#rstats #econsky PSA: The next release of `fixest` will include some important changes (plus cool new features).

E.g. Fixed-effects regs will now default to 'iid' SEs rather than clustered. github.com/lrberge/fixe...

You can install and test drive the dev version from R-universe; see the README.
From the fixest NEWS file: the new default VCOV is iid for all estimations. To change the default to the way it was, place setFixest_vcov(all = "cluster", no_FE = "iid") in your .Rprofile.
July 18, 2025 at 3:36 PM
lrberge.bsky.social
Announcing sircon: a new R console! (for Windows)

It's in beta and actively developed (new patched version just released).

I've been using it for a while now and I think it's rather bug free. But if you encounter some: let me know!

github.com/lrberge/sircon
July 2, 2025 at 3:23 PM
lrberge.bsky.social
Just gave a presentation a the REGIS summer school we organised in Bordeaux.

Algorithms for empirical research

Slides and material here: github.com/lrberge/2025...

The conclusion is that, even in the age of AI, using your brain is still relevant!

#RStats #EconSky
June 20, 2025 at 2:09 PM
Reposted by Laurent Bergé
kylefbutts.bsky.social
github.com/kylebutts/vs...
If you are using vscode and stata, you should try out my extension. It uses interactive window which let's you write in a `.do` file but get a notebook type experience.
June 8, 2025 at 3:07 PM
lrberge.bsky.social
Good post that generates discussions on student's use of LLMs.

If you're a teacher, have a look!
andrew.heiss.phd Andrew Heiss @andrew.heiss.phd · Jun 5
Updated my "AI, LLMs, and bullshit" mini-lecture post in my dataviz class with a new section on vibe coding datavizs25.classes.andrewheiss.com/resource/ai-...
Vibe coding and learning Remember that LLMs are (philosophical) bullshit machines. They don’t care if the answer if gives you is wrong or right. They don’t care if the code runs or not. All they care about is if the answer looks plausible. That said, in part due to the massive amount of publicly available code that’s inside their training data, LLMs can work okay with code in limited circumstances. Think of them as fancy documentation—if a programming language or package has good documentation with lots of good examples online, the stuff that LLMs generate about it will generally have good coverage and won’t be completely made up. They work well enough that a whole new approach to programming has emerged: vibe coding. When vibe coding, you essentially tell an LLM what you want it to make and then let it actually build the whole thing. Some code editors actually allow LLMs to create and edit files directly on your computer. Like, you can say Make a website that has a login page where users can log in and then create and edit blog posts. Create a theme for it that is mostly blue, but with some nice contrasting colors, and use a warm, friendly sans serif font. Add some photos from Flickr too. … and it will attempt to do that, creating all the necessary files and code. It feels neat and magical to code without coding. If you know what you’re doing, and you mistrust everything the LLM spits out, and hand-check everything it says, vibe coding can save you some time. HOWEVER it is (1) potentially dangerous, and (2) bad for learning. Danger! Because LLMs are designed to make plausible looking code, they can reference external packages that don’t exist. For instance, you might want to make a lollipop chart, and it’ll say something like Certainly! First, run remotes::install_github("someone/gglollipop") and then run library(gglollipop) to load it. BUT GUESS WHAT. There is no package named {gglollipop}. It’s not a thing. Malicious actors also vibe code and look for the names of packages that LLMs think should exist, then they go and create those packages themselves, but include malware in those fake packages. If you’re using an editor that allows LLMs to make changes to your computer, it can run all this code without you knowing what happens and bad things can happen. Additionally, vibe-coded code can run well but be full of security vulnerabilities and issues and inefficiencies and bugs. The code it generates is essentially an average of all other code online—it will have bugs and security issues. It’s also incredibly easy to accidentally expose private information to LLMs, like feeding them passwords, usernames, private data, and so on, especially if you’re mindlessly generating and running code. Always distrust and be suspicious about all code that LLMs generate. Bad for learning! Some people claim that vibe coding helps them learn how to code. This is most likely not the case—pure vibe coding likely hurts learning. Think of learning to code like learning a language. You could try to learn French by watching hours and hours of French movies, and you’d certainly pick stuff up, but you won’t learn nearly as well as you would by taking a dedicated French class. You need the foundation first. If you take a few years of French classes and then watch hours and hours of French movies, you’ll learn a lot. If you jump right into a pure media-based approach to learning, it’ll be detrimental to learning the language. Recent research explores this idea with LLMs: Students who substitute some of their learning activities with LLMs (e.g., by generating solutions to exercises) increase the volume of topics they can learn about but decrease their understanding of each topic. Students who complement their learning activities with LLMs (e.g., by asking for explanations) do not increase topic volume but do increase their understanding. We also observe that LLMs widen the gap between students with low and high prior knowledge. (Lehmann, Cornelius, and Sting 2025) Prior knowledge is essential if you want to do any code-related stuff with LLMs, and even then, LLMs are helpful when asking for explanations (e.g. “This code works, but I don’t know why! Can you explain what’s happening with geom_point() here?”), but far less helpful for learning than just generating plausible-looking answers. I can typically tell when you submit code generated by an LLM; it has a certain vibe/style to it and often gives you extraneous code that you don’t actually need to use. Like this—this comes right from ChatGPT: # Load the required libraries library(dplyr) # Read a file named cars.csv into the R session data <- read.csv("data/cars.csv") # Calculate the average value of cty by class average_cty_by_class <- data %>% group_by(class) %>% summarize(average_cty = mean(cty, na.rm = TRUE), .groups = "drop") # Show the results print(average_cty_by_class) The “tells” Here’s how I can tell that that ↑ code comes from an LLM: Comments at every stage read.csv() instead of read_csv(), The older %>% pipe instead of the newer |> pipe na.rm = TRUE inside mean() .groups = "drop" inside summarize() Using print() to show the created object In general, don’t ever just copy/paste from ChatGPT and hope it runs. If you do use these tools, use them to ask for explanations. Make sure you know what every line of code is doing. For example, if it spits out summarize(average_cty = mean(cty, na.rm = TRUE), .groups = "drop"), and you don’t know what those na.rm = TRUE or .groups = "drop" things mean, look at the help page for mean() and see what na.rm actually does so you can decide if you really need to use it or not, or ask the LLM why it included it. I will call out code that looks vibe-coded in your assignments. Please don’t do it.
June 6, 2025 at 11:49 AM
Reposted by Laurent Bergé
andrew.heiss.phd
Updated my "AI, LLMs, and bullshit" mini-lecture post in my dataviz class with a new section on vibe coding datavizs25.classes.andrewheiss.com/resource/ai-...
Vibe coding and learning Remember that LLMs are (philosophical) bullshit machines. They don’t care if the answer if gives you is wrong or right. They don’t care if the code runs or not. All they care about is if the answer looks plausible. That said, in part due to the massive amount of publicly available code that’s inside their training data, LLMs can work okay with code in limited circumstances. Think of them as fancy documentation—if a programming language or package has good documentation with lots of good examples online, the stuff that LLMs generate about it will generally have good coverage and won’t be completely made up. They work well enough that a whole new approach to programming has emerged: vibe coding. When vibe coding, you essentially tell an LLM what you want it to make and then let it actually build the whole thing. Some code editors actually allow LLMs to create and edit files directly on your computer. Like, you can say Make a website that has a login page where users can log in and then create and edit blog posts. Create a theme for it that is mostly blue, but with some nice contrasting colors, and use a warm, friendly sans serif font. Add some photos from Flickr too. … and it will attempt to do that, creating all the necessary files and code. It feels neat and magical to code without coding. If you know what you’re doing, and you mistrust everything the LLM spits out, and hand-check everything it says, vibe coding can save you some time. HOWEVER it is (1) potentially dangerous, and (2) bad for learning. Danger! Because LLMs are designed to make plausible looking code, they can reference external packages that don’t exist. For instance, you might want to make a lollipop chart, and it’ll say something like Certainly! First, run remotes::install_github("someone/gglollipop") and then run library(gglollipop) to load it. BUT GUESS WHAT. There is no package named {gglollipop}. It’s not a thing. Malicious actors also vibe code and look for the names of packages that LLMs think should exist, then they go and create those packages themselves, but include malware in those fake packages. If you’re using an editor that allows LLMs to make changes to your computer, it can run all this code without you knowing what happens and bad things can happen. Additionally, vibe-coded code can run well but be full of security vulnerabilities and issues and inefficiencies and bugs. The code it generates is essentially an average of all other code online—it will have bugs and security issues. It’s also incredibly easy to accidentally expose private information to LLMs, like feeding them passwords, usernames, private data, and so on, especially if you’re mindlessly generating and running code. Always distrust and be suspicious about all code that LLMs generate. Bad for learning! Some people claim that vibe coding helps them learn how to code. This is most likely not the case—pure vibe coding likely hurts learning. Think of learning to code like learning a language. You could try to learn French by watching hours and hours of French movies, and you’d certainly pick stuff up, but you won’t learn nearly as well as you would by taking a dedicated French class. You need the foundation first. If you take a few years of French classes and then watch hours and hours of French movies, you’ll learn a lot. If you jump right into a pure media-based approach to learning, it’ll be detrimental to learning the language. Recent research explores this idea with LLMs: Students who substitute some of their learning activities with LLMs (e.g., by generating solutions to exercises) increase the volume of topics they can learn about but decrease their understanding of each topic. Students who complement their learning activities with LLMs (e.g., by asking for explanations) do not increase topic volume but do increase their understanding. We also observe that LLMs widen the gap between students with low and high prior knowledge. (Lehmann, Cornelius, and Sting 2025) Prior knowledge is essential if you want to do any code-related stuff with LLMs, and even then, LLMs are helpful when asking for explanations (e.g. “This code works, but I don’t know why! Can you explain what’s happening with geom_point() here?”), but far less helpful for learning than just generating plausible-looking answers. I can typically tell when you submit code generated by an LLM; it has a certain vibe/style to it and often gives you extraneous code that you don’t actually need to use. Like this—this comes right from ChatGPT: # Load the required libraries library(dplyr) # Read a file named cars.csv into the R session data <- read.csv("data/cars.csv") # Calculate the average value of cty by class average_cty_by_class <- data %>% group_by(class) %>% summarize(average_cty = mean(cty, na.rm = TRUE), .groups = "drop") # Show the results print(average_cty_by_class) The “tells” Here’s how I can tell that that ↑ code comes from an LLM: Comments at every stage read.csv() instead of read_csv(), The older %>% pipe instead of the newer |> pipe na.rm = TRUE inside mean() .groups = "drop" inside summarize() Using print() to show the created object In general, don’t ever just copy/paste from ChatGPT and hope it runs. If you do use these tools, use them to ask for explanations. Make sure you know what every line of code is doing. For example, if it spits out summarize(average_cty = mean(cty, na.rm = TRUE), .groups = "drop"), and you don’t know what those na.rm = TRUE or .groups = "drop" things mean, look at the help page for mean() and see what na.rm actually does so you can decide if you really need to use it or not, or ask the LLM why it included it. I will call out code that looks vibe-coded in your assignments. Please don’t do it.
June 5, 2025 at 3:04 PM
Reposted by Laurent Bergé
hadley.nz
At Posit, we love @typst.app:

⚡ Make PDFs in milliseconds, not minutes
✨ The power of LaTeX with today's technologies
✍️ Modern typography (including emojis!)
🧠 Clear mental model

So I'm thrilled to announce that we're now supporting its development: posit.co/blog/posit-a...

#rstats
Posit
We’re happy to announce that we’re supporting Typst by funding one of their full-time engineers.
posit.co
June 3, 2025 at 1:47 PM
Reposted by Laurent Bergé
adrianhill.de
I had to endure some teasing from my collaborators for choosing to write our NeurIPS submission in @typst.app – looks like I have one more reason to believe I made the right call... 😅
My BlueSky feed; filled with three mentions of Overleaf being down due the upcoming NeurIPS deadline.
May 14, 2025 at 8:48 AM
Reposted by Laurent Bergé
alvespina.bsky.social
This is a great article in the latest issue of the JEP for anyone teaching panel methods (any level). Love this history of economic terms and ideas feature. #Econsky

pubs.aeaweb.org/doi/pdfplus/...
May 10, 2025 at 9:30 AM
Reposted by Laurent Bergé
pp0196.bsky.social
#rstats #SWE
2021 R Journal issue on software engineering practices in R dev community: journal.r-project.org/issues/2021-2/
Target paper calling for research : journal.r-project.org/articles/RJ-... Wonder if there are recent interventions on the state of SWE with R [beside great work by rOpenSci]
May 4, 2025 at 8:33 PM
lrberge.bsky.social
Blog post (fixest vignette) on "Collinearity": something everybody faces one day but for which there's not much guidance.

Trying to clarify how econometrics software handle collinearity and what one should do about it.

#RStats #EconSky

Comments welcome

lrberge.github.io/fixest/artic...
On collinearity
lrberge.github.io
April 21, 2025 at 4:11 PM
Reposted by Laurent Bergé
s3alfisc.bsky.social
PyFixest 0.29 is on PyPi!

- We have updated how small sample adjustments are computed. PyFixest inference matches fixest now 100%, out of the box and all of the time!
- We have added support for fully saturated event studies, which in turn enables @apoorvalal.com's DiD specification tests.
April 19, 2025 at 12:37 PM
Reposted by Laurent Bergé
kylefbutts.bsky.social
great little explanation on why to avoid passive voice refactoringenglish.com/chapters/pas...
Passive Voice Considered Harmful
Effective writing for software developers
refactoringenglish.com
March 25, 2025 at 12:34 PM
Reposted by Laurent Bergé
nanoporous.net
French researcher, going to a conference in Houston, was forbidden entry to US; his work and personal electronics were both confiscated.

Why? Because a “random search” of his cell phone revealed a negative personal opinion on Trump and the Trump administration. www.lemonde.fr/internationa...
Etats-Unis : un chercheur français refoulé pour avoir exprimé « une opinion personnelle sur la politique menée par l’administration Trump »
Le ministre de la recherche a dit sa « préoccupation », mercredi, après cette décision des autorités américaines. Le chercheur du CNRS aurait subi un contrôle aléatoire à son arrivée, avant que son or...
www.lemonde.fr
March 19, 2025 at 5:37 PM
Reposted by Laurent Bergé
datavisfriendly.bsky.social
Graphic #rstats reminder that vaccines work 📊
www.databrewer.co/R/gallery/gg...
Heatmaps for eight infections diseases in the US: poilio, mumps, rubbella. hepatitis A, ... Each one has the US states in rows and year in columns, with a thick black line showing the year that a vaccine against that disease was introduced. Cells are colored by the number of cases of that disease.
March 20, 2025 at 12:59 AM
Reposted by Laurent Bergé
libbyheeren.bsky.social
Help me write a playbook of best practices for giving data-specific technical workshops or code-heavy talks. This isn't about bashing bad talks, it's about sharing experiences from which we can all learn. We'll all need slightly different things from a learning environment, so let's share. #databs
libbyheeren.bsky.social Libby Heeren @libbyheeren.bsky.social · Mar 4
Have you ever attended a super great OR super terrible data + coding workshop (probably Python, R, shiny, js, etc)? What made it terrible or amazing?
March 4, 2025 at 7:23 AM
Reposted by Laurent Bergé
bsebordeaux.bsky.social
📣 Bordeaux School of Economics is pleased to invite you to participate in the 40th meeting of the European Economic Association @eeanews.bsky.social
📅 25-28 August, 2025
🔔 Submissions until Feb, 28
📍@bsebordeaux.bsky.social
@univbordeaux.bsky.social @cnrsaquitaine.bsky.social
⬇️ Info
eea2025.org
February 11, 2025 at 2:54 PM