pyreghdfe 0.3.0

DEPRECATED — use StatsPAI (pip install statspai). Final release of the Python reghdfe port.

pyreghdfe

⚠️ DEPRECATED — use StatsPAI instead

pyreghdfe 0.3.0 is the final release. The package is no longer maintained. Its entire feature surface — multi-way FE OLS, robust / multi-way cluster SEs, singleton dropping, weights, LSMR/LSQR Krylov solvers — is now a strict subset of StatsPAI (v1.6.6+).

pip install statspai

# pyreghdfe (legacy)
from pyreghdfe import reghdfe
res = reghdfe(data=df, y="wage", x=["experience"],
              fe=["firm_id", "year"], cluster=["firm_id"])

# StatsPAI equivalent
import statspai as sp
res = sp.absorb_ols(
    y=df["wage"].values,
    X=df[["experience"]].values,
    fe=df[["firm_id", "year"]],
    cluster=df["firm_id"].values,
    solver="lsmr",   # or "map" (faster default) / "lsqr"
)

StatsPAI ships additional capabilities pyreghdfe never provided: PPML-HDFE (sp.ppmlhdfe), a formula interface (sp.hdfe_ols("y ~ x | firm + year", data=df)), a Rust-accelerated mean-sweep kernel, and a unified result object with .summary() / .to_latex() / .to_excel().

Migration guide: https://github.com/brycewang-stanford/StatsPAI/blob/main/MIGRATION.md#migrating-from-pyreghdfe

What the 0.3.0 release does: it still installs and works. Importing pyreghdfe now emits a DeprecationWarning with a migration link; no API was broken. Future bugs and feature requests should be filed against StatsPAI — this repository is archived.

---

Python implementation of Stata's reghdfe for high-dimensional fixed effects regression

pyreghdfe is a fast and efficient Python package that replicates the functionality of Stata's popular reghdfe command. It provides high-dimensional fixed effects estimation, cluster-robust standard errors, and seamless integration with pandas DataFrames.

Quick Installation

pip install pyreghdfe

Quick Start

import pandas as pd
import numpy as np
from pyreghdfe import reghdfe

# Create sample data
np.random.seed(42)
n = 1000
data = pd.DataFrame({
    'wage': np.random.normal(10, 2, n),
    'experience': np.random.normal(5, 2, n),
    'education': np.random.normal(12, 3, n),
    'firm_id': np.random.choice(range(100), n),
    'year': np.random.choice(range(2010, 2020), n)
})

# Run regression with firm fixed effects
result = reghdfe(
    data=data,
    y='wage',
    x=['experience', 'education'],
    fe=['firm_id']
)

# Display results
print(result.summary())

📋 Key Features

• ✅ High-dimensional fixed effects - Efficiently absorb multiple fixed effect dimensions
• ✅ Cluster-robust standard errors - Support for one-way and multi-way clustering
• ✅ Weighted regression - Handle sampling weights and frequency weights
• ✅ Singleton dropping - Automatically handle singleton groups
• ✅ Fast computation - Optimized algorithms for large datasets
• ✅ Stata compatibility - Results match Stata's reghdfe command
• ✅ Pandas integration - Seamless DataFrame compatibility
• ✅ Flexible output - Rich statistical results and summary tables

Usage Examples

1. Multiple Fixed Effects

# Regression with firm and year fixed effects
result = reghdfe(
    data=data,
    y='wage',
    x=['experience', 'education'],
    fe=['firm_id', 'year']  # Multiple dimensions
)
print(result.summary())

2. Cluster-Robust Standard Errors

# One-way clustering
result = reghdfe(
    data=data,
    y='wage',
    x=['experience', 'education'],
    fe=['firm_id'],
    cluster=['firm_id']  # Cluster by firm
)

# Two-way clustering
result = reghdfe(
    data=data,
    y='wage',
    x=['experience', 'education'],
    fe=['firm_id'],
    cluster=['firm_id', 'year']  # Cluster by firm and year
)

3. Weighted Regression

# Add weights to your data
data['weight'] = np.random.uniform(0.5, 2.0, len(data))

# Run weighted regression
result = reghdfe(
    data=data,
    y='wage',
    x=['experience', 'education'],
    fe=['firm_id'],
    weights='weight'
)

4. OLS Regression (No Fixed Effects)

# Simple OLS regression
result = reghdfe(
    data=data,
    y='wage',
    x=['experience', 'education'],
    fe=None  # No fixed effects
)

Working with Results

Accessing Coefficients and Statistics

result = reghdfe(data=data, y='wage', x=['experience', 'education'], fe=['firm_id'])

# Get coefficients
coefficients = result.coef
print("Coefficients:", coefficients)

# Get standard errors
std_errors = result.se
print("Standard Errors:", std_errors)

# Get t-statistics and p-values
t_stats = result.tstat
p_values = result.pvalue
print("T-statistics:", t_stats)
print("P-values:", p_values)

# Get confidence intervals
conf_int = result.conf_int()
print("95% Confidence Intervals:", conf_int)

# Get R-squared
print(f"R-squared: {result.rsquared:.4f}")
print(f"Adjusted R-squared: {result.rsquared_adj:.4f}")

Summary Statistics

# Full regression summary
print(result.summary())

# Detailed summary with additional statistics
print(result.summary(show_dof=True))

Advanced Configuration

Custom Absorption Options

result = reghdfe(
    data=data,
    y='wage',
    x=['experience', 'education'],
    fe=['firm_id'],
    absorb_tolerance=1e-10,  # Higher precision for absorption
    drop_singletons=True,    # Drop singleton groups
    absorb_method='lsmr'     # Alternative solver
)

Different Covariance Types

# Robust standard errors (default)
result = reghdfe(
    data=data, 
    y='wage', 
    x=['experience'], 
    fe=['firm_id'], 
    cov_type='robust'
)

# Clustered standard errors
result = reghdfe(
    data=data, 
    y='wage', 
    x=['experience'], 
    fe=['firm_id'], 
    cov_type='cluster', 
    cluster=['firm_id']
)

Comparison with Stata

This package aims to replicate Stata's reghdfe command. Here's how the syntax translates:

Stata:

reghdfe wage experience education, absorb(firm_id year) cluster(firm_id)

Python (PyRegHDFE):

result = reghdfe(
    data=data,
    y='wage',
    x=['experience', 'education'],
    fe=['firm_id', 'year'],
    cluster=['firm_id']
)

Integration Options

This package is actively maintained as a standalone library. For users who prefer a unified ecosystem with additional econometric and statistical tools, reghdfe functionality is also available through:

• StatsPAI - Comprehensive Stats + Econometrics + ML + AI + LLMs toolkit

Related Projects

• PyStataR - Unified Stata-equivalent commands and R functions in Python

API Reference

Main Function: reghdfe()

reghdfe(data, y, x, fe=None, cluster=None, weights=None, 
        cov_type='robust', absorb_tolerance=1e-8, 
        drop_singletons=True, absorb_method='lsmr')

Parameters:

• data (DataFrame): Input data
• y (str): Dependent variable name
• x (list): List of independent variable names
• fe (list, optional): List of fixed effect variable names
• cluster (list, optional): List of clustering variable names
• weights (str, optional): Weight variable name
• cov_type (str): Covariance type ('robust', 'cluster')
• absorb_tolerance (float): Tolerance for fixed effect absorption
• drop_singletons (bool): Whether to drop singleton groups
• absorb_method (str): Absorption method ('lsmr', 'lsqr')

Returns:

• RegressionResults: Object containing regression results

Results Object

The RegressionResults object provides:

• .coef: Coefficients
• .se: Standard errors
• .tstat: T-statistics
• .pvalue: P-values
• .rsquared: R-squared
• .rsquared_adj: Adjusted R-squared
• .conf_int(): Confidence intervals
• .summary(): Formatted summary table

Requirements

• Python ≥ 3.9
• NumPy ≥ 1.20.0
• SciPy ≥ 1.7.0
• Pandas ≥ 1.3.0
• PyHDFE ≥ 0.1.0
• Tabulate ≥ 0.8.0

Contributing

We welcome contributions! Please feel free to:

• Report bugs or request features via GitHub Issues
• Submit pull requests for improvements
• Share your use cases and examples
• Improve documentation and add examples

Development Setup

git clone https://github.com/brycewang-stanford/pyreghdfe.git
cd pyreghdfe
pip install -e ".[dev]"
pytest tests/

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

Support

• Documentation: GitHub Repository
• Issues: GitHub Issues
• Discussions: GitHub Discussions

---

⭐ This package is actively maintained. If you find it useful, please consider giving it a star on GitHub!

Questions, bug reports, or feature requests? Please open an issue on GitHub.