PyFixest

0.50.1 verified Fri May 01 auth: no python

Fast high-dimensional fixed effect estimation library for Python, following the syntax of the R fixest package. Current version: 0.50.1. Released ~monthly. Supports OLS, GLM, IV, quantile regression, and robust/clustered inference with multiple demeaner backends (including Rust).

pip install pyfixest

Common errors

error ImportError: cannot import name 'feols' from 'pyfixest' ↓

cause Old import path from internal modules; or pyfixest not installed.

fix

Use correct import: 'from pyfixest import feols' or 'import pyfixest as pf; pf.feols(...)'

error ValueError: The `vcov` argument must be a dict with keys 'CRV1', 'CRV3', or 'iid'. ↓

cause Passing a string like 'cluster' instead of dict.

fix

Use vcov={'CRV1': 'cluster_variable_name'}.

error TypeError: 'NoneType' object is not iterable when using demeaner_backend='rust-cg' ↓

cause Bug in v0.30.1 on Windows due to Rust ffi typing.

fix

Upgrade to >=0.30.2a or use different backend.

Warnings

breaking Default inference changed in v0.40.0: 'iid' is now default (previously it was cluster by first fixed effect). This replicates R fixest 0.13 behavior. Existing scripts that relied on clustered standard errors without explicitly specifying vcov will silently get different results. ↓

fix Set vcov='cluster' explicitly: feols(..., vcov={'CRV1': 'firm'})

deprecated The 'demean_backend' argument was renamed to 'demeaner_backend' in v0.50.0. Old argument still works but will be removed. ↓

fix Use 'demeaner_backend' instead of 'demean_backend'.

gotcha Formula handling changed in v0.50.0: the 'i()' operator for interactions now behaves differently. Old usage may break or yield unexpected results. ↓

fix Review the new formula syntax docs: https://pyfixest.org/formula.html

gotcha Windows users with Rust backend may encounter a TypeError in v0.30.2. Fixed in v0.30.2a. ↓

fix Upgrade to >=0.30.2a or use 'demeaner_backend='pyhdfe''.

Install

pip install 'pyfixest[dev]'

Imports

feols
wrong
```
import pyfixest as pf; pf.feols(...)
```
correct
```
from pyfixest import feols
```
This is actually correct - both work, but 'from pyfixest import feols' is clearer for single function use.
fepois
wrong
```
from pyfixest.estimation import fepois
```
correct
```
from pyfixest import fepois
```
Direct import from top-level is preferred; internal module paths may change.

Quickstart

Estimate a fixed effects linear regression using a sample dataset. Replace 'distance' and 'property_type' with actual columns from the dataset.

import pandas as pd
from pyfixest import feols

# Load sample data
df = pd.read_csv('https://github.com/py-econometrics/pyfixest/raw/master/data/airbnb.csv')

# Estimate fixed effects model
fit = feols('price ~ distance | property_type', data=df)
print(fit.summary())