This repository contains experimental data and statistical analysis scripts for replicating the results in the paper:

Priniski, et al. (2025). Network structure shapes consensus dynamics through individual decisions. PNAS.

Please refer to the following data files, Jupyter Notebooks, and R scripts listed. Subdirectories are formatted as headers, with important files in that directory listed as bullet points. The main replication files are marked with a ⭐️, and are InteractionDynamics.R, Focal Narrative Alignment.ipynb, NarrativeShifts.R (for modeling experimental data), and SimulationCode.ipynb (running the agent-based models). For more detailed instructions on running these scripts, consult comments in the notebooks and R scripts.

• all_interaction_data.csv: long-formatted dataframe of all interactions. (Analysis scripts based on this dataframe)
• all_tweets.csv: long-formatted dataframe of all pre- and post-interaction personal narratives. (Analysis scripts based on this dataframe.)
• Network Interactions: long-formatted .csv files for individual network runs. File names: f20h1.csv means Face N = 20 Homogeneous Run 1.
• Pre- and Post Data: wide-formatted .csv files of personal narratives and pre-/post- interaction hashtags. Pre- and Post-interaction hashtags were not analyzed in this experiment. File names: f20h1.csv means Face N = 20 Homogeneous Run 1.

• Network Interaction
  • ⭐️ InteractionDynamics.R: R script for analyzing interaction data (Figure 2)
  • models.zip: Statistical models fit in R for interaction analysis. Load these models in when replicating InteractionDynamics.R to reduce run-time for fitting Bayesian models.
  • ⭐️ Focal Narrative Alignment.ipynb: Jupyter Notebook for running the narrative alignment analysis in the paper (Figure 3A)
• Personal Narratives
  • all_tweets.csv: same file as above, just saved locally for running script.
  • claims.csv: causal claims extracted by the Causal Claims Transformer. The Causal Claims Transformer can be accessed at the following Hugging Face repo
  • ⭐️ NarrativeShifts.R: R Script for analyzing the personal narrative changes.

• Pre- and Post Qualtrics Wrappers. This directory contains the qualtrics wrappers around the OTree network interaction experiments that are used for Phase 1 and Phase 3 of the experiment.
• Network Interaction OTree Software.
  • face_experiment: OTree Code for the Name Game condition
  • hashtag_experiment: OTree code for the Hashtag Game condition
  • generate_trials.py: Python code for generating interaction pairs. The trials used in the experiment are in custom_networks directories of the face and hashtag experiment.

• ⭐️ SimulationCode.ipynb: Run this Jupyter Notebook to simulate the Context Aware Agents. All other code and files are helper functions.