Replication Materials

This repository contains the code and data used in the paper "CHORUS: A new dataset of state interest group policy positions in the United States," forthcoming in State Politics & Policy Quarterly. If you use our code and/or data, please cite the paper as:

Hall, Galen, Joshua Basseches, Rebecca Bromley-Trujillo, and Trevor Culhane. 2023. "CHORUS: A new dataset of state interest group policy positions in the United States." State Politics & Policy Quarterly. Forthcoming 2023.

Data

The dataset used in the related paper is available online in the SPPQ Dataverse: https://dataverse.unc.edu/dataverse/sppq. To download it locally, run the code/download.py file.

Note that we will maintain updated versions of this dataset at a different location. For replicating the results in the paper, please use the version of the dataset available in the SPPQ Dataverse.

File structure

• data/: Contains the data used in the paper.
• data/positions.parquet: The positions recorded in testimony and lobbying records.
• data/bills.parquet: The bills on which positions were recorded, merged with data from LegiScan and the National Conference of State Legislatures for the states in CHORUS.
• data/clients.parquet: The organizations that recorded positions on bills in CHORUS.
• data/block_assignments.parquet: The block assignments for each organization in each state, from our hierarchical bayesian stochastic block model.
• data/{IL, TX, MA, CO}_network_figure_clusters_named.xlsx: Data used to create the network figures in the paper, with the clusters named by the authors.
• data/hbsbm/{state}_{record_type}_corrected_categorical_blockstate.pkl: This folder contains pickle (".pkl") files for each BlockModel object generated using the data corresponding to a unique {state, record_type} combination from positions.parquet (for example, one file contains the blockmodel for the testimony records in Arizona). The corrected in the filename indicates that the blockmodel incorporated degree correction, and the categorical indicates that the blockmodel used categorical edge covariates as opposed to the layered model.

Code

The python code used to generate the data and figures presented in the paper is available in the code folder. Code used to create the CHORUS dataset is available for review upon reasonable request.

File structure

• code/download.py: Functions to download the data from the SPPQ Dataverse or from Google Drive.
• code/load.py: Functions to load the data into memory as pandas dataframes.
• code/figures.py: Functions to generate the figures presented in the paper.
• code/utils.py: Utility functions for data analysis and plotting.
• code/hbsbm.py : Functions to create the hierarchical bayesian stochastic block models. When recreating results from scratch using run_all_blockmodels_from_scratch(), note that since the blockmodels are stochastic, the results will not be identical to those presented in the paper.
• code/main.py: Main file to run the code, via main.main().

Figures

The figures presented in the paper are available in the figures folder. The code used to generate them is available in the code/figures.py file. Note that the figures in the paper have been edited for clarity and aesthetics.

Requirements

The code was written in Python 3.10 Most required packages are listed in requirements.txt. To install them, run the following command in the terminal:

pip install -r requirements.txt.

The hbsbm code also requires the graph-tool package, which can be installed via Conda but is not available on PyPI. To install it, follow the instructions on the graph-tool website.

Runtime

We ran the code on a premium Google Colab instance with 51 GB of RAM and a Python 3 Google Compute Engine backend. The code took approximately 15 minutes to run. Note that the maximum RAM actually used was about 13GB, so the code should run on a machine with 16GB of RAM.

License

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

Contact

For questions or comments, please contact Galen Hall at <galen.p.hall [at] gmail.com>.

Acknowledgments

This project was supported with funding from the Climate Social Science Network (CSSN); see https://cssn.org for more information. We thank the CSSN for their support.

The blockmodeling code is based on the hbsbm package by Tiago Peixoto.

The authors are responsible for all errors.