Alex

Oh interesting, this does not look good. For me everything has been working well, I just upgraded to the newest fixest release via conda-forge a few weeks ago without any problems (as it happens, on windows!).

No global R or Python installation needed! And if you have these, no interference with your global installations.

Big fan. pixi.sh/latest/

Got a new PC and getting back to pyfixest dev is a breeze with pixi (via conda-forge, which seems to get too little love?). This just works: clone pf, install pixi, type pixi r tests in the shell, and then pixi installs Python, R, all Python and R deps, all R deps, and then starts running tests.

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...

#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...

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.

#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.

Next step is to rewrite the vignette (more to the point & I need to make a better case for the methods usefulness in business contexts).

Also works with "negative" effects:

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

Love to see this, of course =)

In contrast to uv, it is cross-language and also allows to pin down non-python dependencies in the same env (R, Julia, etc). And it also has nice "pixi run command" options (you can even define tasks as in justfiles, which I personally use a lot). pixi.sh/latest/

For pyfixest's development, we've been using pixi for a while, which I really enjoy. More or less it is the conda-forge (free & open source) equivalent to uv (it actually runs uv for PyPi dependency resolution). Really fast both for PyPi & conda-forge dependency resolution.

This I will still have to explore (as matplotlib does not always auto-scale). Generally there are two options to handle this as of now - users can keep / drop covariates via a function arg, or simply combine many of them into a single "combined" covariate in the estimation process.

Yes, negative contributions should be handled and would be colored red and should add to the "initial difference" bar

My attempt - how to tell you I work in industry without telling you ;) what do you think?

For the ladder I'll make sure that all defaults can be easily changed.

I agree - optimally I am looking for default labels that make it immediately clear to the econometricians what is being shown (this is the status quo) but that also allows non-technical stake holder to easily understand the substantive message.

Thanks - I actually control for x21 in the short regression ... will have to think about a good naming convention for this case.

Yes this makes a lot of sense. Had also wondered if anyone had seen any great visualizations. Am sold that we should add a plot method. Thanks!

cc @gelbach.bsky.social , and also @nickchk.com, whose blog post (nickchk.substack.com/p/decomposit...) convinced me of the value of decomp methods in industry - maybe you have some tips on effective communication?

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).

Re fast routines for the quantile regression process - we follow Chernozhukov et al (2019) and their algo's are quite an improvement over "naive" for loops over quantiles:

This release has had 8 new contributors (none of which I think are around here) - thanks all for your help!

You can find the full changelog here: py-econometrics.github.io/pyfixest/cha...

Second, we have added support for 𝐪𝐮𝐚𝐧𝐭𝐢𝐥𝐞 𝐫𝐞𝐠𝐫𝐞𝐬𝐬𝐢𝐨𝐧 based on a Frisch-Newton Solver, including iid, heteroskedastic, and cluster robust standard errors, as well as fast routines for fitting the entire quantile regression process.

Plus, moving to Rust gets rid of the 5-15s you'd have to wait for the numba backend to compile at the first attempt. For now, the Rust back end is optional, but we plan to make it the default with the next PyFixest release.

For large problems, the Rust back end can lead to 2-5x performance improvements. In the release note, we show how fitting a regression with N = 10Mio, 3 fixed effects and 10K clusters can be fit in 3 instead of 15s via Rust.

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.

Yes I also still struggle with it. On the other hand, if we check "what the markets says" - the Machado & Santos Silva paper stands at 1.7K citations

With preprocessing applied, I run a qr on N = 1 mio and k = 3 in 1.5 seconds, vs 18 with the "pure" FN. Nice. PR here: github.com/py-econometr...