zacharyst/scale-free-protest_superlinearscaling

Fork to start a new project to model the emergence of superlinear scaling in protest

scale-free-protest

--------------- NEW -------------

This project investigates the scaling properties of protest, specifically how protest size changes as a function of network size.

Scripts/simulations.py is from the shanest/scale_free_protest repository.

See Steinert-Threlkeld and Steinert-Threlkeld (2021), "How social networks affect the repression-dissent puzzle" in PLOS ONE, for descriptions of original modeling decisions. For this project, focus on the Holme-Kim model across its ranges of m_t, with particular attention to m_t=.315.

Motivation/OLD_Protest_SuperlinearScaling.pdf summarizes the state of this project as of November 1, 2018. This paper documents superlinear scaling in actual protest data and shows initial modeling that failed to generate superlinear scaling. The models used for those results are earlier, less sophisticated versions of what was used in the PLOS ONE paper, so inability to generate scaling should not be overinterpreted. In addition to providing motivation, the key use of this paper is to show the analysis to perform for the simulation results. The x-axis will be network size, the y-axis is size of a simulation's protest, and the line of best fit is the resulting scaling relationship.

--------------- OLD ----------------

Companion repository to the paper ``How Social Networks Affect the Repression-Dissent Puzzle''.

Runs network simulations of the spread of protest with varying network structure and repression mechanisms.

Data for some empirical results can also be found at: http://dx.doi.org/10.17605/OSF.IO/6U9RW

Setup:

• Make a new venv/conda environment, with python >= 3.6
• pip install -r requirements.txt

Basic usage:

python simulations.py --exp {config} --num_procs 1 --out_dir OUTPUT_DIR

The experiments from the paper can be found as specified in YAML files in the exps directory. The argument should be the root of the file (e.g. exp1-1).

Analysis scripts

The simulations will output a large amount of raw data about the dynamics of the protest across trials. There are several scripts in analysis for analyzing this data, briefly described here:

1. 01_ProcessSimulationData_v1.R

   • Merges datasets
   • Input:
     • exp3a-graph_type=powerlaw_cluster_graph-num_nodes=1000-threshold_type=uniform-m=3-repression_type=node_removal.csv
     • exp2a-threshold_type=uniform-repression_type=node_removal-graph_type=watts_strogatz_graph-k=15-num_nodes=1000.csv
     • exp1a-1-num_nodes=1000-graph_type=scale_free_graph-threshold_type=uniform-repression_type=node_removal.csv
     • exp1a-2-num_nodes=1000-graph_type=scale_free_graph-repression_type=node_removal-threshold_type=uniform.csv
     • exp1a-3-repression_type=node_removal-num_nodes=1000-graph_type=scale_free_graph-threshold_type=uniform.csv
     • exp1a-4-repression_type=node_removal-num_nodes=1000-graph_type=scale_free_graph-threshold_type=uniform.csv
   • Output:
     • processedData/01_experimentsCombined.csv

2. 02_MakeFigures_v1.R

   • Make plots to explore the data. Main goal is to facet results by network type, DV, one facet plot per variable of interest.
   • Input:
     • processedData/01_experimentsCombined.csv
   • Output:
     • A ton of figures

3. 03_RandomForestRegression_v1.R

4. 04_Simulations_Regression_InvestigatedScaleFreeAndRepression.R

   • The purpose of this script is to see if regression inferences change based on regression on different subsets of the scale-free dataset. I have found in its parent script that repression_rate can sometimes have heterogenous effects, when low enough and sample is small enough.
   • Input:
     • processedData/01_experimentsCombined_eigenCore.csv
   • Output:
     • Lots of figures

5. 05_Simulations_RepressionFigures_v1.R

   • The purpose of this script is to make the figures detailed on Page 132 of my Moleskine journal.
   • Input:
     • processedData/01_experimentsCombined_eigenCore.csv
   • Output:
     • Lots of faceted figures

6. 06_tStatSampleSizeChange_v1.R

   • The purpose of this script is to see how a t-statistic changes as increase sample size.

7. 07_FatalityRate_v1.R

   • The purpose of this script is to generate an estimate of the fatality rate at protests. It uses ACLED and NAVCO.
   • Input (in OSF project linked above):
     • ACLED_protests_1900-01-01-2019-08-07.csv
     • navco3.csv
   • Output:
     • Just an average value, no saved output.

8. 08_RegressionResultsOverTime_v1.R

   • The purpose of this script is to see if regression results change as we get new data about an event. The idea is that it replicates how our knowledge changes as new data comes in with history.
   • Input:
     • Data/MassMobilization_Binghampton/mm_public_120615.csv
     • Data/MassMobilizationsAutocracyDataset/EventLevel/events.csv
   • Output: