pith. sign in

arxiv: 2601.21749 · v2 · submitted 2026-01-29 · 💰 econ.EM

fixest: A fast and feature-rich framework for econometric estimations in R

Laurent R. Berg\'e , Kyle Butts , Grant McDermott This is my paper

Pith reviewed 2026-05-16 10:07 UTC · model grok-4.3

classification 💰 econ.EM
keywords R packageeconometricsfixed effectsregressionfixed-point algorithmC++ implementationdifference-in-differencesinstrumental variables
0
0 comments X

The pith

fixest is an R package for fast econometric estimations using a novel C++ fixed-point acceleration algorithm for fixed effects.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces fixest as a unified R package supporting ordinary least squares, instrumental variables, generalized linear models, maximum likelihood, and difference-in-differences. It focuses on efficient fixed-effects estimation through a new acceleration algorithm written in C++ that speeds convergence in varied data settings. An expressive formula interface lets users run multiple estimations, stepwise regressions, and variable interpolations in one call. Inference options such as robust standard errors can be adjusted after estimation, and the package outputs publication-ready tables and plots directly. Benchmarks position it ahead of alternatives across R, Python, and Julia for speed on typical tasks.

Core claim

fixest supplies a single framework for econometric estimation in R that particularly excels at fixed-effects models via a novel fixed-point acceleration algorithm coded in C++. This algorithm enables rapid convergence even for models with varying slopes. The package supplies an expressive formula syntax that supports multiple estimations and stepwise procedures in a single command, built-in choices for robust standard errors that can be changed without re-estimation, and direct methods for producing regression tables and coefficient plots.

What carries the argument

The novel fixed-point acceleration algorithm implemented in C++ that solves fixed-effects problems with rapid convergence across data contexts.

If this is right

  • Researchers can estimate models with many fixed effects or varying slopes in far less time than with prior R tools.
  • A single formula call can generate multiple related regressions including stepwise variants and interpolated variables.
  • Robust standard error choices can be swapped after the main estimation step without recomputing coefficients.
  • Publication-ready tables and plots are produced directly from fitted objects without extra formatting code.
  • The package matches or beats specialized alternatives from Python and Julia on speed for standard econometric tasks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Wider adoption of fixed-effects specifications becomes practical in applied work once computational barriers drop.
  • The R-plus-C++ pattern shown here could encourage similar performance tuning in other statistical packages.
  • Extensions to additional model families or integration with large-scale data tools would follow naturally from the current design.
  • Fields relying on difference-in-differences or instrumental-variables methods gain a quicker prototyping option.

Load-bearing premise

The novel fixed-point acceleration algorithm delivers genuine speed and convergence gains over existing methods on real econometric data without numerical instability or other hidden costs.

What would settle it

A head-to-head runtime and convergence benchmark on a large panel dataset with many fixed effects where fixest either fails to finish substantially faster than packages such as lfe or produces numerically different coefficient values.

Figures

Figures reproduced from arXiv: 2601.21749 by Grant McDermott, Kyle Butts, Laurent R. Berg\'e.

Figure 2
Figure 2. Figure 2: Default iplot graph for DiD estimations. Effect on y Year Estimate and 95% Conf. Int. −5 0 5 10 −10 −5 0 5 TWFE Sun and Abraham 39 [PITH_FULL_IMAGE:figures/full_fig_p039_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Example of graphs created with coefplot (above) and iplot (below), with default (left) and customized (right) layout. Effect on Ozone Estimate and 95% Conf. Int. −30 −20 −10 0 10 Temp Wind 6 7 8 9 Month Effect on Ozone Estimate and 95% Conf. Int. −30 −20 −10 0 10 9 8 7 6 Wind Temp Month Effect on Ozone Month Estimate and 95% Conf. Int. −30 −20 −10 0 10 5 6 7 8 9 Effect on y period Estimate and 95% Conf. In… view at source ↗
read the original abstract

fixest is an R package for fast and flexible econometric estimation. It provides a unified framework for applied research, with comprehensive support for a diverse class of models: ordinary least squares, instrumental variables, generalized linear models, maximum likelihood, and difference-in-differences. The package particularly excels at fixed-effects estimation, supported by a novel fixed-point acceleration algorithm implemented in C++. This algorithm achieves rapid convergence across a variety of data contexts and enables efficient estimation of complex models, including those with varying slopes. An expressive formula interface facilitates multiple estimations, stepwise regressions, and variable interpolation in a single call. Users can adjust inference strategies on the fly, choosing from an array of built-in robust standard errors. The package also provides methods for publication-ready regression tables and coefficient plots. Benchmarks demonstrate that fixest offers best-in-class performance against leading alternatives in R, PYTHON, and JULIA.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The manuscript introduces the fixest R package as a unified, high-performance framework for econometric estimations including OLS, IV, GLM, ML, and difference-in-differences models. It highlights a novel fixed-point acceleration algorithm implemented in C++ for rapid fixed-effects estimation (including varying slopes), an expressive formula interface supporting multiple estimations and stepwise regressions, on-the-fly robust standard error options, and methods for publication-ready tables and plots. Benchmarks against leading alternatives in R, Python, and Julia are presented to support claims of best-in-class performance.

Significance. If the reported benchmarks hold under scrutiny, fixest would represent a substantial practical advance for applied econometricians working with high-dimensional fixed-effects models. The C++ implementation of the fixed-point accelerator and the comprehensive feature set (formula interface, inference flexibility) merit explicit credit as strengths that could reduce computation time and coding burden in large-scale empirical work.

major comments (1)
  1. [Benchmarks section] Benchmarks section (likely §5 or equivalent): the central claim of best-in-class performance and 'rapid convergence across a variety of data contexts' for the novel fixed-point algorithm is load-bearing, yet the manuscript provides insufficient detail on dataset characteristics, convergence failure rates, numerical precision comparisons, or stability checks; without these, the superiority over existing methods cannot be fully evaluated.
minor comments (2)
  1. [Abstract] The abstract asserts 'best-in-class performance' without qualification; the benchmarks section should explicitly state the scope of tested models and any limitations to avoid overstatement.
  2. [Tables] Ensure all benchmark tables report both timing and any accuracy or convergence metrics side-by-side for direct comparison.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive assessment and the recommendation for minor revision. We address the single major comment below.

read point-by-point responses

  1. Referee: [Benchmarks section] Benchmarks section (likely §5 or equivalent): the central claim of best-in-class performance and 'rapid convergence across a variety of data contexts' for the novel fixed-point algorithm is load-bearing, yet the manuscript provides insufficient detail on dataset characteristics, convergence failure rates, numerical precision comparisons, or stability checks; without these, the superiority over existing methods cannot be fully evaluated.

    Authors: We agree that greater transparency on these dimensions will strengthen the paper. In the revised manuscript we will expand the benchmarks section (and add a short appendix) with: (i) a table reporting the exact characteristics of each dataset (observations, number and type of fixed effects, presence of varying slopes, source and any preprocessing); (ii) convergence failure rates for the fixed-point accelerator across all specifications and tolerance settings; (iii) side-by-side numerical comparisons of coefficient vectors and standard errors against reference implementations (lfe, reghdfe, etc.) to document precision; and (iv) additional stability diagnostics (sensitivity to starting values and tolerance). These additions will be presented without altering the existing performance conclusions. revision: yes

Circularity Check

0 steps flagged

No significant circularity: software implementation paper with empirical benchmarks only

full rationale

The paper presents the fixest R package and its novel fixed-point acceleration algorithm for fixed-effects models. No mathematical derivation chain exists; claims rest on code implementation details and runtime benchmarks versus R/Python/Julia alternatives. No equations reduce to self-inputs, no fitted parameters are relabeled as predictions, and no self-citation forms a load-bearing uniqueness theorem. The reader's assessment of score 1.0 aligns with the absence of any derivation that could be circular. This is a standard non-circular software description.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a software package paper rather than a theoretical derivation, so no free parameters, axioms, or invented entities are introduced.

pith-pipeline@v0.9.0 · 5450 in / 905 out tokens · 29118 ms · 2026-05-16T10:07:28.507538+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

14 extracted references · 14 canonical work pages · 1 internal anchor

  1. [1]

    Vincent Arel-Bundock, Noah Greifer, and Andrew Heiss

    doi: 10.18637/jss.v103.i01. Vincent Arel-Bundock, Noah Greifer, and Andrew Heiss. How to interpret statistical models using marginaleffects forRandPYTHON.Journal of Statistical Software, 111(9):1–32,

    work page doi:10.18637/jss.v103.i01

  2. [2]

    Bezanson , author A

    URL https://doi.org/10.1137/141000671. Johannes Boehm. GLFixedEffectModels.jl,

    work page doi:10.1137/141000671

  3. [3]

    doi: 10.1016/j.spl.2020.108945

    ISSN 0167-7152. doi: 10.1016/j.spl.2020.108945. URL https://www.sciencedir ect.com/science/article/pii/S0167715220302480. John C Driscoll and Aart C Kraay. Consistent covariance matrix estimation with spatially dependent panel data.Review of economics and statistics, 80(4):549–560,

    work page doi:10.1016/j.spl.2020.108945 2020

  4. [4]

    Rcpp : Seamless R and C++ Integration

    doi: 10.18637/jss.v040.i08. URL https://doi.org/10.18637/jss.v040.i08. Dirk Eddelbuettel, Romain Francois, JJ Allaire, Kevin Ushey, Qiang Kou, Nathan Russell, Inaki Ucar, Douglas Bates, and John Chambers. Rcpp: SeamlessRandC++Integration,

    work page doi:10.18637/jss.v040.i08

  5. [5]

    doi: 10.1137/10079687X

    ISSN 1064-8275, 1095-7197. doi: 10.1137/10079687X. URL http://epubs.siam.org/doi/10.1137/10079687X. Ragnar Frisch and Frederick V . Waugh. Partial Time Regressions as Compared with Individual Trends. Econometrica, 1(4):387–401,

    work page doi:10.1137/10079687x

  6. [6]

    doi: 10.2307/1907330

    ISSN 0012-9682. doi: 10.2307/1907330. URL https://www.jstor.or g/stable/1907330. Simen Gaure.lfe: Linear group fixed effects.The R Journal, 5(2):104–117, 2013a. Simen Gaure. OLS with multiple high dimensional category variables.Computational Statistics & Data Analysis, 66:8–18, 2013b. Matthieu Gomez. FixedEffectModels.Jl: Fast Estimation of Linear Models ...

    work page doi:10.2307/1907330

  7. [7]

    9 Ventsislav Chonev, Joel Ouaknine, and James Worrell

    Association for Computing Machinery. ISBN 978-1-4503-2029-0. doi: 10.1145/2488608.2488724. URL https://doi.org/10.1145/2488608.2488724. Michael C. Lovell. Seasonal Adjustment of Economic Time Series and Multiple Regression Analysis. Journal of the American Statistical Association, 58(304):993–1010,

    work page doi:10.1145/2488608.2488724 2029

  8. [8]

    doi: 10.230 7/2283327

    ISSN 0162-1459. doi: 10.230 7/2283327. URL https://www.jstor.org/stable/2283327. Daniel Lüdecke, Mattan S. Ben-Shachar, Indrajeet Patil, and Dominique Makowski. Extracting, com- puting and exploring the parameters of statistical models using R.Journal of Open Source Software, 5(53):2445,

    work page arXiv

  9. [9]

    doi: 10.21105/joss.02445. James G. MacKinnon and Halbert White. Some heteroskedasticity consistent covariance matrix esti- mators with improved finite sample properties.Journal of Econometrics, 29(3):305–325,

    work page doi:10.21105/joss.02445

  10. [10]

    Grant McDermott

    doi: 10.1016/0304-4076(85)90158-7. Grant McDermott. ggfixest: Dedicated ggplot2 Methods for fixest Objects,

    work page doi:10.1016/0304-4076(85)90158-7

  11. [11]

    URL https://doi.org/10.32614/RJ-2021-053

    doi: 10.32614/RJ-2021-053. URL https://doi.org/10.32614/RJ-2021-053. Amrei Stammann. Fast and Feasible Estimation of Generalized Linear Models with High-Dimensional k-way Fixed Effects.ArXiv e-prints,

    work page doi:10.32614/rj-2021-053 2021

  12. [12]

    Fast and Feasible Estimation of Generalized Linear Models with High-Dimensional k-way Fixed Effects

    URL https://arxiv.org/pdf/1707.01815.pdf. Liyang Sun and Sarah Abraham. Estimating dynamic treatment effects in event studies with heteroge- neous treatment effects.Journal of econometrics, 225(2):175–199,

    work page internal anchor Pith review Pith/arXiv arXiv

  13. [13]

    Achim Zeileis, Susanne Köll, and Nathaniel Graham

    doi: 10.18637/jss.v016.i09. Achim Zeileis, Susanne Köll, and Nathaniel Graham. Various Versatile Variances: An Object-Oriented Implementation of Clustered Covariances inR.Journal of Statistical Software, 95:1–36, October

    work page doi:10.18637/jss.v016.i09

  14. [14]

    Various Versatile Variances: An Object-Oriented Implementation of Clustered Covariances in

    ISSN 1548-7660. doi: 10.18637/jss.v095.i01. URL https://doi.org/10.18637/jss.v095.i01. 56

    work page doi:10.18637/jss.v095.i01