This project was done at the Complexity Science Hub in Vienna as part of their summer internship program, which I found through Princeton.
In this project I model FEC campaign finance data as an absorbing Markov chain in which candidates are the absorbing states. As such, to "follow the money" is to follow the random walk of money. But most committees' political affiliation are unknown. Using this model, I (1) empirically infer entity ideologies and subsequently (2) compare Democratic and Republican random walk dynamics.
My takeaways: Individual ideology is an emergent feature of the whole network. While Democratic and Republican dollars take paths of similar lengths, Republican paths end in more diverse absorbing states within the party and even outside the party. That is, Democratic and Republican dollar trajectories are equally complex but differ in shape, revealing that Republican-leaning actors use more diverse investment strategies whereas Democrat-leaning actors tend to focus on their winners.
To run my data pipeline, run data/download_data.sh (this downloads and cleans the 2000-2022 FEC data) then data/process_data.sh (this prepares the data for use by igraph). Run the analyses in analysis/analysis.ipynb. Note that the null model simulations use run_null_simulations.py which takes around 45 minutes to run.