Skip to content

Releases: py-econometrics/pyfixest

v0.27.0

15 Dec 12:31
a4272b0
Compare
Choose a tag to compare

PyFixest 0.27.0

Highlights

New Features

  • Adds support for Gelbach's (JoLe 2016) Regression Decomposition method by introducing a decompose() method for Feols.
  • Adds support for the multiple hypothesis correction by Westfall & Young via the pf.wyoung() function.
  • Input data frames to pf.feols() and pf.fepois() are now converted to pandas via narwhals.
    As a result, users can now provide duckdb or ibis tables as inputs, as well as pandas and polars data frames. polars and pyarrow
    are dropped as a dependencies.

Documentation

  • Adds a vignette on multiple testing corrections: link
  • Adds a vignette on Gelbach's regression decomposition: link

Bug Fixes

  • Fixes a bug in the wildboottest method, which incorrectly used to run a regression on the demeaned dependend variable in case it was
    applied after a fixed effects regression. My apologies for that!
  • Fixes a bug in the ritest method, which would use randomization inference coefficients instead of t-statistics whenever the "randomization-t" option would be selected, leading to incorrect results. This had consequences for the rwolf() function, which, in case of running ri-inference, would default to run the "randomization-t". My apolgies!

Full Changelog

v0.26.2

01 Nov 11:53
5027952
Compare
Choose a tag to compare

No changes relative to 0.25.4, I (Alex) simply messed up the PyPi release.

v0.25.4

01 Nov 11:40
bbf5f96
Compare
Choose a tag to compare

Highlights

  • Bugfix in fixef() method for WLS @s3alfisc (#682) that would lead to incorrect predictions with WLS when the newdata argument was employed. Thanks to @marcandre259 for the 🐛 report!
  • feols() and fepois() obtain "split" and "fsplit" arguments, just as you might know them from fixest::feols().
  • @leostimpfle has added the "iterated rectifier" check for detecting separation in Poisson models.
  • We have added timing benchmarks against linearmodels (@marcandre259 ).

Changes

v0.25.2

07 Oct 17:13
0724002
Compare
Choose a tag to compare

Highlights

  • Environments are now managed via pixi. This should greatly facilitate new contributions!
  • We have added a pf.dtable() function for producing summary statistics.
  • We now provide outcome plots via the panelview() function.
  • All difference-in-differences estimators can now be loaded via pf.lpdid(), pf.did2s(), etc
  • We implement in-memory regression compression a la duckreg, via polars.

What's Changed

Full Changelog: v0.24.2...v0.25.2

v0.24.2

31 Aug 15:42
05f44bd
Compare
Choose a tag to compare

Changes

v0.24.1

29 Aug 07:15
70e2b0c
Compare
Choose a tag to compare

Changes

v0.24.0

22 Aug 19:18
a961ccd
Compare
Choose a tag to compare

Changes

  • Layout improvements and new function arguments for etable, in particular for LaTex output: check out the etable notebook for examples and details @dsliwka (#583)
  • Small extension to etable that allows to relabel variables @dsliwka (#577)
  • Add labels argument to coefplot() and iplot() @s3alfisc (#580)
  • jit compile multicollinearity detection #466 @s3alfisc (#579)

PyFixest 0.23.0

06 Aug 06:37
c3c1823
Compare
Choose a tag to compare

Hightlights

  • PyFixest now includes Wald and F-Tests through the wald_test() method.
  • The new Wald Test feature enables us to provide basic IV diagnostics on the first stage, thanks to a significant overhaul of the Feiv class internals.
  • The pf.etable() function for regression output has been significantly enhanced for better aesthetics.
  • We've greatly improved the speed of the predict() method.
  • A new panelview function has been added to inspect staggered treatment rollouts in Difference-in-Differences studies.
  • We've introduced new arguments to feols() and fepois(), including a "lean" argument to clear memory-heavy objects from Feols instances.
  • RI support has been added for the rwolf function, for experimenters who want to control the family-wise error rate of their tests.
  • An update() method has been implemented, which allows to update regression coefficients based on incoming data (useful for online learning setups).
  • Additionally, new vignettes on regression tables and using PyFixest with the marginaleffects package have been added.

Take a look at the quickstart notebook for a brief intro to all new features.

Breaking Changes

  • We're adjusting the default small sample corrections to exactly match the fixest defaults, and a new vignette demonstrates equivalence between fitting models via R-fixest and py-fixest =) This change will lead to minor differences in computed standard errors and vcov's between versions 0.22.0 and 0.23.0.

New Contributors

What's Changed

Full Changelog: v0.22.2...v0.23.0

v0.22.2

16 Jun 16:32
30dd48d
Compare
Choose a tag to compare
v0.22.2 Pre-release
Pre-release

Changes

Maintenance

v0.22.0

13 Jun 20:52
cd2e08c
Compare
Choose a tag to compare

Bug Fixes 🐛

Changes

Infrastructure

  • infrastructure: fix minor release drafter bugs @s3alfisc (#504)