Reciprocity under conflict: Replication Materials

This repository provides all the materials necessary to replicate our study "Reciprocity as a social strategy evolves more readily under conflict rather than cooperation" by A. Romano, A.S. Saral and C.K.W. De Dreu.

Contents of this repository

The repository contains the following items:

• code/simulation: This directory contains the simulation code implemented in Python 3.
• code/analysis: This directory contains the code for the analysis and plots written in R.
• output/: This directory will contain the output for the simulation (data).
• plots/: This directory will contain the plots generated by R.

Ways to reproduce or test

1. Simulate the study on your machine You can reproduce our results by running the simulation code on your machine. Please note that running the simulation code with every combination in the paper may takes considerable time. (Several days on a standard machine with eight cores). However, you can still run it with fewer iterations and without the robustness check parameters. By default, the parameters in this repository are a reduced version. You can find those parameters and change them in code/simulation/simulate.py and comment on the full parameter set.

2. Use Our Precomputed Data We also provide the precomputed data that we used in our study, which you can download from the following link (805 MB in total): https://www.dropbox.com/sh/zu6czzbl1jiscmq/AACE0-A3NiDN2Q98zk9R6QQ9a?dl=0

3. Just demo the analysis/plots We already have a small simulation data in the repository for demonstration. You can test the analysis code by producing the plots using this small data.

Instructions to Reproduce

To reproduce the analysis and plots presented in our study, you should run the R scripts provided in the code/ directory. These scripts use the data generated by the simulation code or the precomputed data, depending on your chosen option. Note that you must install the necessary R packages listed in the code/requirements.R file before running the analysis scripts. The resulting plots and analysis will be saved in the plots/ directory.

Reproducing using our pre-simulated data

1- Clone the repository

git clone https://github.com/aseyq/reciprocity_under_conflict.git

2- Download .parquet files from the Dropbox link (805 MB) and extract the .parquet files (df_aggregate_payoffs.parquet, df_coop.parquet, df_long.parquet) in the folder output/processeddata https://www.dropbox.com/sh/zu6czzbl1jiscmq/AACE0-A3NiDN2Q98zk9R6QQ9a?dl=0 Overwrite the files that are already there.

Alternatively, you can use wget to download the files using the command line to output/processeddata:

cd reciprocity_under_conflict

cd output/processeddata

Remove the demo files that are already there.

rm *.parquet

Download the files

wget "https://www.dropbox.com/scl/fi/62bn7bziqjq5k8spocf96/df_aggregate_payoffs.parquet?rlkey=bg5k5yddm3ln5ceza1c3y8qxk&dl=1" -O df_aggregate_payoffs.parquet

wget "https://www.dropbox.com/scl/fi/ob4bba2s41ao015l9cf4d/df_coop.parquet?rlkey=f5pbpps68i1u4lmzvm2qzpcib&dl=1" -O df_coop.parquet

wget "https://www.dropbox.com/scl/fi/z0ninjglk72kh4jiz0ds9/df_long.parquet?rlkey=y0zbhe0m37o0dr0nosmoevd1i&dl=1" -O df_long.parquet

3- Install the required R packages (if necessary)

# (go to the main folder `reciprocity_under_conflict` if you are not)
cd code/analysis

RScript requirements.R

4- Run the following scripts

• Plots in the paper: `paper_plots
• SI: si_plots.R, si_basefit.R, si_moran.R

Rscript paper_plots.R

Rscript si_plots.R

Rscript si_basefit.R

Rscript si_moran.R

Reproducing by running the simulations

1- Clone the repository

git clone https://github.com/aseyq/reciprocity_under_conflict.git

2- Go to the downloaded folder

cd reciprocity_under_conflict

3- (Optional but recommended) Create a Python virtual environment. This will let you install packages only in the virtual environment in an isolated manner.

python3 -m venv venv

and activate it

(MacOS and Ubuntu)

source venv/bin/activate

(Windows)

venv\Scripts\activate

4- Install required packages

cd code/simulation

pip install -r requirements.txt

5- Run simulations The paper uses 1000 generations and 1000 iterations, while the reproduction code has 300 generations and 12 iterations. This is sufficient to test to code but not to replicate the results. You set these numbers closer to the original number of generations and iterations by modifying lines 23 and 24 in simulate.py.

Please note that it might take some time, even with the default reduced number of parameters. With the computer we tested, it takes about 17 minutes. We run the study in a HPC cluster with multiple cores over 75 hours, but one should expect to wait 80-240 hours for the complete simulations to run.

python3 simulate.py

6 - Install the required R packages (if necessary)

# (go to the main folder `reciprocity_under_conflict` if you are not)
cd ../analysis

Rscript requirements.R

7- Go to code/analysis folder and combine CSV files created by the simulation (might take a couple of minutes)

cd ../analysis

Rscript 1_combine_csvs.R

This should combine the CSV files generated by simulations, and you should have combined_data.csv in output/combineddata folder.

7- Convert CSV files to a parquet file

Rscript 2_convert_parquet.R

This should convert CSV file to parquet data and create combined.parquet in output/combineddata folder.

8- Reshape data into a long format and generate additional tables (might take a couple of minutes)

Rscript 3_reshape_data.R

This should create df_aggregate_payoffs.parquet, df_coop.parquet and df_long.parquetin the processed data folderoutput/processeddata. Then you can run the scripts to regenerate the analyses in the paper. At this point, you can delete the previous data generated and keep only the data in processeddata` folder.

9- Go to code/analysis folder and execute the following scripts to create plots in plots/ folder. (Takes a few minutes)

• Plots in the paper: paper_plots.R

Rscript paper_plots.R

• Plots in Supplementary Information: si_plots.R, si_basefit.R, si_moran.R (Note that you should have also simulated additional parameter set of robustness checks.

Rscript si_plots.R

Rscript si_basefit.R

Rscript si_moran.R

Just demo the code generating the plots in the paper with the example data in the repository

1- Clone the repository

git clone https://github.com/aseyq/reciprocity_under_conflict.git

2 - Install the required R packages (if necessary)

# (go to the main folder `reciprocity_under_conflict` if you are not)
cd ../analysis

Rscript requirements.R

3- Execute the following scripts to create demo plots in plots/ folder.

• Plots in the paper: paper_plots.R

Rscript paper_plots.R

System Requirements

Hardware requirements

Any modern personal computer or server that supports Windows, MacOS, or Linux operating system with enough RAM (at least 4 GB recommended.) The original simulations take a long time and generate data around 2GBs. Some operations, like combining the data files after the simulations, might require larger RAM depending on the number of simulation parameters. But if you once converted it to parquet (or downloaded the parquet files we provided), you wouldn't need to use memory as big as the dataset as it uses Apache arrow (arrow) to solve this issue.

We tested this repository with the following machine: Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz Processor 32,0 GB RAM

Software Requirements

You need to have Python 3 (used 3.8) for simulations and R (used 4.2.3) for the analysis.

Additionally, we used the following packages (and their dependencies), which can be installed automatically using requirements files in the repository:

Python packages:

• Numpy 1.22
• Pandas 2.0.3
• joblib 1.3.1
• Pytest 6.2.5

R packages

• dplyr (1.1.0)
• purr (1.0.0)
• forcats (1.0.0)
• string (1.5.0)

We tested this repository on Windows 11 and Ubuntu 18 Linux operating systems.

Citation

If you use our materials for your own research, please cite our original paper using the following reference:

[TBA]