This repository is a work in progress.

Respondent Driven Sampling is an epidemiological sampling method used to estimate the size of different 'hidden' populations.

This project demonstrates four simulations of RDS as random walks on networks with different topological structures:

1. Degree Bias: On a scale-free network, a random walk over-samples high-degree 'hubs', leading to biased estimates if uncorrelated attributes (like Narcan possession) are associated with popularity.
2. Bottlenecks: On a Stochastic Block Model, a single recruiter gets trapped in a local community, failing to explore the 'hidden' population across the bottleneck.
3. Homophily: When recruiters prefer peers with similar status, the 'effective' bottleneck tightens, biasing the estimate further.
4. Multi-Seed: By initiating 10 parallel chains, we reduce variance and overcome bottlenecks via the Central Limit Theorem.

Generated for EPS 109 Final Project.