pysyncon

A python module for the synthetic control method that provides implementations of:

• Synthetic Control Method (Abadie & Gardeazabal 2003)
• Robust Synthetic Control Method (Amjad, Shah & Shen 2018)
• Augmented Synthetic Control Method (Ben-Michael, Feller & Rothstein 2021)
• Penalized Synthetic Control Method (Abadie & L'Hour 2021)

The package also provides methods for performing placebo tests and generating confidence intervals.

The implementation of the synthetic control method aims to be reconcilable with the R package Synth and similarly the implementation of the Augmented synthetic control method and the R package augsynth.

Installation

Install it from PyPI using pip:

python -m pip install pysyncon

Usage

Documentation is available on github-pages. In the examples folder are notebooks reproducing the weights/ATT from:

• The Economic Costs of Conflict: A Case Study of the Basque Country, Alberto Abadie and Javier Gardeazabal; The American Economic Review Vol. 93, No. 1 (Mar., 2003), pp. 113-132. (notebook here)
• The Kansas income-tax cuts example from The Augmented Synthetic Control Method, Eli Ben-Michael, Avi Feller & Jesse Rothstein; Journal of the American Statistical Association Vol. 116, No. 536 (2021), pp. 1789-1803. (notebook here)
• The worked example 'Prison construction and Black male incarceration' from the last chapter of the first edition of 'Causal Inference: The Mixtape' by Scott Cunningham. (notebook here)
• Comparative Politics and the Synthetic Control Method, Alberto Abadie, Alexis Diamond and Jens Hainmueller; American Journal of Political Science Vol. 59, No. 2 (April 2015), pp. 495-510. (notebook here)

Reproducing the results from the papers

The optimisers used (scipy's Nelder-Mead and SLSQP) are sensitive to the BLAS backend and to the installed numpy/scipy versions, so computed weights and estimates can vary slightly across environments — typically in the third decimal place of the weights and a few percent of point estimates. Reproducing the published values to full precision requires the reference environment that the notebook-parity tests (parity_tests/) are verified against on CI:

• Python 3.11
• numpy 2.0.2
• scipy 1.13.1
• pandas 2.3.3
• matplotlib 3.9.4
• ubuntu-22.04

These are pinned in the parity job (.github/workflows/parity.yml). Even with these versions, results on other hardware may still differ slightly, as the underlying BLAS libraries dispatch on the CPU.

Citation

If you use this package in your research, you can cite it as below.

@software{pysyncon,
  author = {Fordham, Stiofán},
  month = dec,
  title = {{pysyncon: a Python package for the Synthetic Control Method}},
  url = {https://github.com/sdfordham/pysyncon},
  year = {2022}
}