Social Escape from Learning Traps

Analysis pipeline for the paper "Decision rule inference limits social escape from learning traps."

Repository Structure

social-escape-learning-traps/
├── data/                    # Raw experimental data (JSON files)
├── preprocessing/           # Data preprocessing pipeline
├── results/                 # Analysis pipelines for each research question
├── simulation/              # ALCOVE-RL simulation framework
└── Root files              # Utilities, constants, and run all script. 

See individual folder READMEs for detailed documentation of each component.

Paper figures

Figure	Description	File Path
Figure 2	Decision rule distribution after initial individual learning phase.	./results/drule_proportions/outputs/first_test_overview.pdf
Figure 3	Decision rule distribution for trapped learners after second learning phase.	./results/drule_proportions/outputs/trapped_learner_comparison.pdf
Figure 4	Exploration approach bad/good heatmaps for trapped learners who learned optimal rule during second learning phase.	./results/reward_punishment_heatmap/outputs/trapped_to_optimal_grid.pdf
Figure 5	Effect of social decision rule inference success on observational learning success.	./results/partner_prediction/outputs/partner_prediction_combined_figure.pdf
Figure 6	Decision rule distributions for each learning and test block, by second learning phase condition.	./results/drule_proportions/outputs/drule-proportions-bar-plot.pdf

Root Level Files

Core Pipeline

• generate_all_outputs.py - Master script that runs the entire analysis pipeline:
  1. Simulation data generation
  2. Data preprocessing
  3. All statistical analyses and visualizations

Utilities

• constants.py - Global constants (dataset names, phase structures)
• stats.py - Statistical functions (permutation tests, bootstrap CIs)
• script_utils.py - Helper functions for running analysis scripts
• .gitignore - Standard Python/data science exclusions

Quick Start

Run Complete Pipeline

python generate_all_outputs.py

This will:

1. Generate simulated data using ALCOVE-RL model
2. Preprocess all experimental and simulated data
3. Run all analyses and generate figures

Run Individual Components

Each component can also be run independently:

# Preprocessing only
cd preprocessing/
python generate_outputs.py

# Specific analysis
cd results/drule_proportions/
python generate_outputs.py

Key Analyses

Analysis	Location	Description
Demographics	results/demographics/	Participant counts and filtering statistics
Decision Rule Proportions	results/drule_proportions/	Decision rule proportions across conditions.
Test Phase Decision Rule Stats	results/test_drule_statistics/	Decision rule proportions at test phases, across conditions, with additional stats like CIs and significance testing comparisons.
Points Analysis	results/points_per_block/	Task performance of trapped learners
Partner Prediction Phase Analyses (Exp 5)	results/partner_prediction/	Effect of partner prediction phase accuracy on learning (Exp 5)
Exploration heatmaps	results/reward_punishment_heatmap/	Exploration pattern heatmap (proportion approached safe/dangeorus stimuli)

Data Flow

Raw JSON → Preprocessing → Filtered CSVs → Individual Analyses → Figures/Tables
                ↑
         Simulated Data

Data Anonymization

⚠️ Note: The anonymization scripts in anonymization/ are included for transparency and reproducibility documentation, but they operate on the original unanonymized data files which are not included in this pub release.

This repository contains anonymized experimental data. The original raw data (containing personally identifiable information) has been processed through an anonymization pipeline that:

Dependencies

Standard Python data science stack:

• pandas, numpy
• matplotlib, seaborn
• scipy (for statistical tests)

Acknowledgments

This analysis pipeline was developed with assistance from Claude Code (Anthropic). Files with more wholesale Claude generated code are marked with authorship comments at the top of the file. I have reviewed all code and assume full responsibility for its correctness.