Acknowledgements

Like many open-source software projects, PyFixest builds on work and ideas first developed in other packages. In this section, we want to acknowledge and express our appreciation for the authors of these packages and their creativity and hard work.

Software

Unless explicitly stated otherwise, all PyFixest code is written independently from scratch. The packages listed below have influenced PyFixest’s API design or algorithmic choices, or are used for testing PyFixest as reference implementations, but no source code has been copied except where explicitly stated (with license and permission details provided inline).

fixest (R)

If open source software is made by “standing on the shoulders of giants”, in case of PyFixest, there is mostly one very big giant - Laurent Bergé’s formidable fixest R package. fixest is so good we decided to stick to its API and conventions as closely as Python allows when starting to work on a fixed effects regression package in Python. Without fixest, PyFixest likely wouldn’t exist - or at the very least, it would look very different. Most importantly, fixest has shaped our understanding of how a user-friendly regression package should look like and what functionality it should offer.

More concretely, we have borrowed the following API conventions and ideas directly from fixest:

Feature	What PyFixest borrows
Formula syntax	feols(), fepois(), feglm() function and argument names; the i() interaction operator; multiple estimation syntax via sw(), sw0(), csw(), csw0(); fixed effects interactions via fe1^fe2
Multiple Estimation Optimizations	The core idea that most of the work of fixed effects regression can be pooled / cached when estimating multiple models with the same fixed effects structure
Demeaning / FWL	The alternating-projections algorithm in PyFixest is a standalone implementation in numba/rust, but uses the same convergence criteria and default parameters
Small-sample corrections	The ssc() function to control small sample adjustments, and all of its defaults - adj, fixef_K, cluster_adj, cluster_df - mirror fixest exactly (see fixest’s standard errors vignette)
Collinearity detection	The algorithm is a Rust/Numba re-implementation of Laurent Berge’s C++ routine in fixest, re-licensed under MIT with Laurent’s permission
Post-estimation	etable(), coefplot(), iplot(), coef() etc mirror fixest’s output and plotting functionality
On the fly variance covariance adjustments	As in fixest, you can adjust the vcov post estimation by calling a vcov() method on the results object (Feols in pyfixest and fixest in fixest)
Predict method for fixed effects	The predict() and fixef() methods in PyFixest mirrors fixest’s functionality for obtaining fitted values, fixed effects, and linear predictions

You can learn more about fixest on github, via its documentation, or by reading the associated paper.

PyFixest is tested against fixest via rpy2 to ensure numerical equivalence (usually rtol = 1e-08, atol = 1e-08) for coefficients, standard errors, t-statistics, p-values, confidence intervals, etc for OLS, IV, Poisson, and GLM models.

---

By functionality

Poisson regression

Package	Language	Role
ppmlhdfe	Stata	PyFixest’s Poisson estimator (fepois) implements the ppmlhdfe algorithm as described in Correia, Guimaraes & Zylkin (2020), including its separation detection and acceleration strategies. Test datasets for separation examples are taken from the ppmlhdfe repository (MIT license)

Instrumental variables

Package	Language	Role
ivDiag	R	The IV diagnostics implementations are validated against ivDiag by Lal et al.

Quantile regression

Package	Language	Role
quantreg	R	PyFixest’s quantreg() implementation is tested against R’s quantreg package (by Roger Koenker) for coefficient and NID standard error equivalence.
qreg2	Stata	PyFixest’s cluster-robust standard errors for quantile regression are tested against Stata’s qreg2 output, which is based on work by Parente & Santos Silva (2016).

Difference-in-Differences

Package	Language	Role
did2s	R	PyFixest’s DID2S estimator’s API is strongly inspired by Kyle Butts’ R package (MIT license) and we have relied on Kyle’s writeup of the method for our own implementation. Tests compare coefficients and standard errors against the R implementation. Additionally, PyFixest’s event_study() function is inspired by the event_study() function in did2s.
lpdid	R	PyFixest’s local-projections DID estimator is highly influenced by Alex Cardazzi’s R code (published under MIT) for the lpdid package. We also test against Alex’ package.
lpdid	Stata	We also test our implementation against Daniel Busch and Daniele Girardi’s Stata implementation of the local-projections estimator.

Panel data visualization

Package	Language	Role
panelView	R	PyFixest’s panelview() function for visualizing treatment patterns and outcomes in panel data is inspired by the panelView R package by Mou, Liu and Xu.

Randomization inference

Package	Language	Role
ritest	R	PyFixest’s ritest() method’s API heavily borrows from Grant McDermott’s R package and is tested against it.
ritest	Stata	Grant’s ritest is itself inspired by Simon Heß ritest Stata package.

Wild cluster bootstrap

Package	Language	Role
wildboottest	Python	PyFixest loads classes from wildboottest to run wild bootstrap inference. wildboottest is a Python port of fwildclusterboot.
fwildclusterboot	R	An R implementation of the “fast and wild” algorithm by Roodman et al.
boottest	Stata	Roodman et al. (2019). The fast wild cluster bootstrap methodology traces back to Roodman’s Stata boottest package

Multiple hypothesis testing (Romano-Wolf)

Package	Language	Role
wildrwolf	R	PyFixest’s rwolf() Romano-Wolf correction is tested against the wildrwolf R package for both HC and CRV inference.
wildwyoung	R	An R implementation of the Westfall-Young correction using the wild bootstrap.
rwolf	Stata	A Stata implementation of the Romano-Wolf stepdown procedure that inspired development of rwolf.
wyoung	Stata	A Stata implementation of the Westfall-Young stepdown procedure by Jones, Molitor & Reif.

Causal cluster variance

Package	Language	Role
TSCB-CCV	Stata	Pailanir & Clarke. PyFixest’s CCV implementation (Abadie et al., QJE 2023) is tested against Daniel Pailanir and Damian Clarke’s Stata implementation.

Gelbach decomposition

Package / Author	Language	Role
b1x2	Stata	PyFixest’s decompose() method is tested against hardcoded results from Gelbach’s b1x2 Stata package.
Apoorva’s Linear Mediation Gist	Python	The initial implementation of Gelbach’s decomposition in PyFixest was based on Apoorva’s gist

Demeaning and fixed effects recovery

Package	Language	Role
lfe	R	We based our first implementation of the MAP algorithm on the description in the “how lfe works” vignette.
pyhdfe	Python	PyFixest’s demeaning results are tested against Jeff Gortmaker’s pyhdfe. pyfixest’s first MVP was built using pyhdfe it ran its demeaning algorithm via pyhdfe MAP algo.

---

Test infrastructure

The following packages are used in PyFixest’s test suite to bridge between Python and R:

Package	Language	Role
rpy2	Python	The bridge between Python and R that powers all cross-language test comparisons.

Other Software

Here we list other foundational software without which a project like PyFixest would not be possible:

• formulaic
• numpy
• numba
• pandas
• scipy
• matplotlib
• great-tables / maketables
• pyo3

Papers and Algorithms

• Bergé, L. R., Butts, K., & McDermott, G. (2026). “Fast and user-friendly econometrics estimations: The R package fixest.” arXiv:2601.21749.
• Correia, S., Guimarães, P., & Zylkin, T. (2020). “ppmlhdfe: Fast Poisson estimation with high-dimensional fixed effects.” The Stata Journal, 20(1). arXiv:1903.01690.
• Gaure, S. (2013). “OLS with multiple high dimensional category variables.” Computational Statistics & Data Analysis, 66, 8-18. Vignette: How lfe works.
• Guimarães, P. & Portugal, P. (2010). “A simple feasible procedure to fit models with high-dimensional fixed effects.” The Stata Journal, 10(4), 628-649. DOI:10.1177/1536867X1101000406.
• Koenker, R. & Ng, P. (2005). “A Frisch-Newton Algorithm for Sparse Quantile Regression.” Acta Mathematicae Applicatae Sinica, 21(2), 225-236. DOI:10.1007/s10255-005-0231-1.