Unstructured bargaining analysis

Code for analyzing the results of the unstructured bargaining experiment

Setup

Software requirements

The project is set up so that pixi handles the installation of the required dependencies into a local virtual environment (Except for latex. If you'd like to compile the paper, make sure you have a tex distribution with the necessary packages and latexmk available on the path.). First, install pixi as described in the documentation. Then, you can run the commands as described in the replicating the paper section.

Using pixi (advanced)

With pixi installed, you have three main commands at your disposal:

• pixi run [TASK] runs the pixi task [TASK]. For a list of available tasks, run pixi task list.
• pixi run [COMMAND] runs the command [COMMAND] in the pixi environment. For example, pixi run python starts a Python shell in the pixi environment.
• pixi shell starts a shell in the pixi environment. It is analogous to conda activate. Note, that there is no need to activate the environment before using the run command. Also, there is no deactivate command. To exit the shell, simply type exit.

All of these commands take care of setting up the virtual environment and installing the required dependencies. If you want to add a new dependency, simply run pixi add [PACKAGE]=[VERSION]. It will then be added to the pixi.toml file and installed in the virtual environment.

Pixi also uses a lockfile. This lockfile is updated automatically when you add a new dependency. If you want to update the lockfile manually, you can delete the pixi.lock file and run pixi install to recreate it. The lockfile should normally be committed to the repository to make sure that everyone uses the same versions of the dependencies.

Data requirements

Data exported from oTree is already placed into the data/raw directory. Please do not modify it under any circumstances. Automated checks are in place to ensure that the data is not modified.

Exporting data from oTree (for documentation purposes)

The data in the `data/raw` directory is exported from the otree admin interface. The following steps describe how it can be obtained. First, navigate to the `Data` tab in the otree admin interface and export the following files to the indicated location. Always chose the plain format.

• All apps → data/raw/wide_data.csv
• Per-app: live_bargaining (custom_export) → data/raw/live_data.csv
• Per app: introduction → data/raw/intro_data.csv
• Per-app: live_bargaining → data/raw/bargaining_data.csv
• Per-app: survey → data/raw/survey_data.csv
• Per-app: sliders: → data/raw/slider_data.csv
• Chat logs → data/raw/chat_data.csv
• Page times → data/raw/page_time_data.csv

Then, use pixi run anonymize to create data/raw/wide_data_nonpersonal.csv, data/raw/survey_data_nonpersonal.csv, and the anonymezed survey data data/raw/survey_data_personal.csv (note that the columns are individually reshuffled in the latter file, therefore it is only suitable for single-variable desciptives). Make sure to remove the original data files after this step.

Checksums for the raw data are stored in the RAW_DATA_CHECKSUMS repository variable.

Replicating the paper

First, clone the repository and navigate to the project directory.

git clone [email protected]:stanmart/unstructured-bargaining-analysis.git
cd unstructured-bargaining-analysis

Then, make sure that the software requirements are met as described in the software requirements section.

Finally, the following command will set up the environment, run the analysis, and create the paper.

pixi run paper

Other pixi commands are also available. E.g.:

• To create the collected datasets in data/clean_collected, run: pixi run create-datasets
• To run the statistical tests, run pixi run run-analysis
• To create all figures in pdf format, run pixi run figures
• To rerun the power analysis at src/power_analysis/power.ipynb, run pixi run power-analysis
• To create a chart of the analysis steps at build-graphs/filegraph.svg, run pixi run filegraph (or similar commands for the dag or the rulegraph)

Automatic replication

The project is set up with GitHub Actions to automatically replicate the paper on every push to the main branch. See the publish.yml workflow for details. The compiled paper is available at stanmart.github.io/unstructured-bargaining-analysis/paper.pdf.

Contributing

Automated checks

The project is set up with GitHub Actions to run automated checks on every push and pull request to the main branch. The checks include:

• ruff check for Python code style
• ruff format for Python code formatting
• codespell for spell checking
• pyright for static type checking
• Data integrity checks for the raw data based on the RAW_DATA_CHECKSUMS repository variable

These checks can also be run locally using pixi. Simply run pixi run checks to run all checks. If you only want to run a specific check, you can do so by running pixi run [CHECK], where [CHECK] is one of lint, format, spellcheck, typecheck or data-check.

Alternatively, ruff and codespell checks can also be installed as pre-commit hooks. First, install pre-commit, e.g. using pipx:

pipx install pre-commit

Then, you can install the hooks by running

pre-commit install

After this, the checks will be performed automatically before every commit.

Latex dependencies

A list of the required latex packages is stored in the `tl_packeges.txt` file. Make sure to update the file if you add new packages to the paper:

pixi run update-latex-deps

After updating the file, commit it to the repository so that the CI can install the required packages.