#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...

#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.

Vis method for decomposition now merged to main, feedback welcome!

Love to see this, of course =)

Currently revamping the default output view for the Gelbach Decomposition. Goal is to help both academics but also data scientists in industry to communicate results easily. Are the column and panel headers immediately clear? Any better way to summarize decomp results?

Really awesome new formulaic feature! We'll update pyfixest to make it happen (and might get some perf improvements thanks to polars along the way).

𝐏𝐲𝐅𝐢𝐱𝐞𝐬𝐭 0.30.1 is on 𝐏𝐲𝐏𝐢 and I am extremely happy with this release, mostly because of two 𝐧𝐞𝐰 𝐟𝐞𝐚𝐭𝐮𝐫𝐞𝐬: a Rust back end for all performance critical parts and quantile regression.

nerdsniped by @s3alfisc.bsky.social's pyfixest quantile regression PR into implementing a bunch of QR solution methods. Anyone have strong views on these things [was taught the LP one and good alt solvers exist, FN - which is implemented in quantreg - is very fast] gist.github.com/apoorvalal/3...

Presenting an invitation to the open source python econometrics ecosystem with @s3alfisc.bsky.social on Thursday at a NABE webinar. Please come [and/or send us PRs!] They also have talks from @jondr44.bsky.social and @lihualei.bsky.social lined up for later in the summer nabe.com/NABE/Events/...

Great slides! ... and we have a Python implementation for you in pyfixest: py-econometrics.github.io/pyfixest/ref...

Hi, I am looking for someone with a Stata license to help me verify an implementation for cluster robust errors in quantile regression as implemented in the qreg2 routine. Any help would be much appreciated! Gh issue with details here: github.com/py-econometr...

Reading the Koenker book on quantile regression and having a blast. Top quotes: "It is a somewhat unhappy fact of life that the asymptotic precision of quantile estimates in general (...) depend on the reciprocal of a density function ..."

I've started to work on a quantile regression implementation for pyfixest. There's a basic PR that implements the Frisch-Newton solver in numpy + a quantreg interface. Still plenty of things to do (standard errors, performance improvements), so please let me know if you want to contribute!

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...

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.

I am currently revamping small sample defaults of pyfixest (the goal is to match fixest SEs out of the box). Now some unit tests for Poisson comparing Python vs R fail for derived statistics (tstats, pvalues, CIs) because (I think) of restrictive tolerances.

New paper, of the "your toys are probably fine, just make sure they don't have any serrated edges" variety

TIL I learned that of course you can set breakpoints in pytest scripts, you just have to move up the callstack by typing "u"... Increasingly coming to the conclusion that whenever there is a Python dev tool that I think everyone would want to have, it's already been shipped with the main libraries.

Thanks to @apoorvalal.com, we now have some benchmarks for PyFixest on the GPU: 2-3x improvements for large problems via JAX relative to numba! We also released pyfixest 0.28.0, mostly with small bug fixes, but it also contains JAX support and GLMs (no fixed effects yet) as experimental features.

More cool things coming to pyfixest!

NEW R PACKAGE: I created the {fetwfe} package to implement fused extended two-way fixed effects (FETWFE), my estimator for difference-in-difference under staggered adoptions. You can install and use it by following the instructions on GitHub here: github.com/gregfaletto/.... Short thread:

I’d like to set up some benchmarks for pyfixest using a cloud service and would like access to multiple threads and a GPU. Any recommendations? Colab pro only seems to offer two threads. Sane pricing and ease of use are the main criteria. Maybe @apoorvalal.com @gmcd.bsky.social @alemartinello.com ?

Something cool for the next pyfixest release 😊

I just released Coding for Economists v1.0.2. This update fixes a lot of minor issues and switches to Astral's uv as the recommended package manager. We have some new contributors too. And, big news, CfE also hit 750 stars! ⭐⭐⭐ aeturrell.github.io/coding-for-e...

`etwfe` v0.5.0 is now available on CRAN (and R-universe). grantmcdermott.com/etwfe/ As a reminder, `etwfe` makes it easy to estimate Extended TwoWay Fixed Effects a la @jmwooldridge.bsky.social. Super flexible and good for nonlinear models, recovering heterogeneous TEs, etc. #econsky #rstats 1/3

Wonderful! Cheers Dear readers, fyi we are doing a #Python update for #GaborsDataAnalysis github.com/gabors-data-...

𝗣𝘆𝗙𝗶𝘅𝗲𝘀𝘁 𝟬.𝟮𝟳: (𝗣𝗿𝗲-𝗖𝗵𝗿𝗶𝘀𝘁𝗺𝗮𝘀 🎅) 𝗥𝗲𝗹𝗲𝗮𝘀𝗲 Hi all, a new version of PyFixest (0.27) has found it's way to PyPi! Highlights: we now support @gelbach.bsky.social’s regression decomposition & Westfall and Young's multiple testing correction.

We have a new notebook that motivates multiple testing corrections and shows how to use them via pyfixest. At the end, it produces this pretty figure: py-econometrics.github.io/pyfixest/mul...

Call for Help: As AE for JSS, I am looking for reviewers for a neat paper (and #Rstats package) with low/mid/high precision float support, something of interest for (very) large models and machine learning. If you,dear reader, could review this please get in touch via email. Thanks in advance.

What's the standard approach for synthetic control with many treated units? One at the time? If the goal is to match the average outcome in the treatment group over time, why not estimate the propensity score using daily outcomes as covariates, and then build a SC using propensity weights?

As it’s still a little under the radar - @vincentab.bsky.social & team have also developed a Python version of marginaleffects: github.com/vincentarelb... Works with statsmodels and pyfixest!

Turns out this was a bug after all, I missed a few intercepts …

Happy to share that my and @cconlon.bsky.social's micro BLP paper was just accepted at the Journal of Econometrics! jeffgortmaker.com/files/Incorp... It's our 2nd tied to our PyBLP software, so here's a thread on surprising (to me) benefits of combining methods research with open source work. 1/6

We’re about to finish an implementation of @gelbach.bsky.social‘s regression decomposition for the next pyfixest release. But there‘s a problem! We only match Stata‘s b1x2 to the 3-4th digits after the comma. Is this a bug, or just numerical error?

Is there a package in #RStats R that calculates Westfall & Young familywise error rate adjusted p values?

Okay I couldn't find anything like this, so I had ChatGPT help me make one by picking the most important tools from the index of Cameron & Trivedi Stata book & tell me the Python equivalents (plus a few I added). I've tried to check it, but comments very welcome! docs.google.com/document/d/e...

Hey #EconSky, I'm on the job market with an #EconJMP about open source software. OSS is a global public good, widely used and provided by the private sector, but the target of recent industrial policy. Paper: jeffgortmaker.com/files/Open_S... 1/

Hey #RStats people. How do you feel about startup messages? Would this nudge be too obnoxious if I added it to my package?

The great-tables library now (experimentally) supports Tex! This is pretty cool and we will likely refactor pyfixest’s etable() to produce latex output based on this. GT is a truly great library all around, big endorsement from my side! posit-dev.github.io/great-tables...

(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...

Questions to the Bayesians here - in practice, do you ever use the Bernstein-von Mises Theorem to approximate the posterior for large samples? encyclopediaofmath.org/wiki/Bernste... Context: trying to reduce the annoyingly long run time of a Bayesian model with large N.

Looking forward to present on pyfixest tomorrow!

Day 1 of #CDSM24 wrapped up with an inspiring keynote from Professor Susan Athey! Looking forward to Day 2, packed with more causal data science insights, an industry roundtable, and a live podcast session. Don't miss it! Program: www.causalscience.org#Programme

We’ve released a patch to pyfixest (0.26.2) with the following new features: - feols and fepois have new split and fsplit arguments (just as in fixest!) - you can check for separation in Poisson models via the „iterative rectifier“ - we’ve fixed a 🐞 in the predict method for WLS with newdata

From Wooldridge‘s paper, we know that for panel data, the TWFE estimator and the 2-way Mundlak are identical. So for a model as Y = D + unit + time + e and panel data we can instead estimate Y = D + ave(D, unit) + ave(D, time) + u Does this generalize to „higher dimensional panels“?

We’ve started a user forum for PyFixest over at discourse: pyfixest.discourse.group We hope it will be useful for discussing the implementation of estimators and the econometrics of linear models in general!

Is there someone with a Stata license around here who’d like to help us integrate @gelbach.bsky.social‘ regression decomposition into pyfixest? I’d like to run some tests against the b1x2 module but don’t have access to Stata. Would be much appreciated! GitHub issue here: github.com/py-econometr...

need to work out some adaptation of bag-of-little-bootstraps / subsampling for this setting [or work out some closed-form solution, which seems far-fetched] but otherwise i'm pretty happy with this. constraints that impose groupby+means get you pretty far

You'll get a kick out of this: arxiv.org/abs/2410.09952 (h/t @apoorvalal.com @s3alfisc.bsky.social )

Pleased to announce the launch of a website that tallies the number of times a software has been used in replication files + gives you badges. softwarecite.com e.g., softwarecite.com/package/pand...