twowaypanel

Welcome to the documentation for twowaypanel, a Python package for bias-corrected estimation and inference in nonlinear panel data models with two-way (additive individual and time) fixed effects. This package accompanies the paper:

Yan, Zizhong, Zhengyu Zhang, Mingli Chen, Jingrong Li, Iván Fernández-Val. (2026), Robust Priors in Nonlinear Panel Models with Individual and Time Effects. arXiv e-prints, arXiv:2604.03663.

For the source code, release notes, and issue tracking, please visit the project’s GitHub homepage: https://github.com/zizhongyan/twowaypanel

twowaypanel provides bias-corrected estimation and inference for:

• model parameters, and

• average partial effects (APEs), for both continuous regressors and discrete regressors (finite changes).

In the current release, twowaypanel supports four model classes: binary logit, probit, multinomial logit, and ordered logit. The explanatory variables can be either strictly exogenous or predetermined (e.g., include lagged dependent variables to accommodate dynamic models).

If you are new to the package, we recommend starting with Installation, then working through the Tutorial and Examples sections, and finally consulting the API Reference for detailed function/class documentation.

Table of Contents

• Installation
  • Prerequisites
  • Installation options
    • 1. Install from PyPI (recommended)
    • 2. Install the development version from GitHub
    • 3. Install from a local copy
    • 4. Jupyter / notebook users

• Quick Start Tutorial
  • Background and scope
  • Supported models and methods
  • The main entry point: twowaypanel.fit
  • Inputs and arguments
    • Inputs: data arrays
    • Model selection
    • Bias correction: prior and ac
    • Serial dependence / trimming: lag
    • Estimation algorithm: algorithm
    • Output control: silent and APEs
    • Ordered logit identification: cutoff0
    • Starting values: sv
    • MCMC tuning parameters (only when algorithm="MCMC")
  • Returned object and key attributes
  • Where to go next

• Examples
  • Data used in this page
  • Import and setup
  • Example 1: Binary logit with strictly exogenous regressors (Angrist and Evans 1998)
  • Example 2: Binary logit with strictly exogenous regressors + model-specific prior (MCMC) + convergence diagnostics
  • Example 3: Analytical bias correction (Fernández-Val and Weidner, 2016)
  • Example 4: Dynamic probit with generic prior
  • Example 5: Ordered logit (simulated data)
  • Example 6: Multinomial logit (simulated data)
  • References

• API Reference
  • Data generation for examples and testing
  • Empirical data loader
  • Estimation interface

Code maintainer

Zizhong Yan, Institute for Economic and Social Research (IESR), Jinan University, Guangzhou, China. Email: helloyzz@gmail.com