Project for MSc. in Statistics and Data Science
The iglm package in R provides a unified statistical framework for modeling individual responses and social network structures simultaneously in connected populations.
In real-world social and economic systems, individual behaviors and relationship ties co-depend on each other, meaning classical regression models that assume independent observations yield biased results.
Modeling the joint probability distribution of node-level attributes and connections, iglm allows researchers to capture peer influence, contagion, and selection effects in one unifying framework.
Bayesian analysis of network data is notoriously difficult because intractable normalizing constants prevent standard parameter sampling.
This project implements Bayesian inference utilizing the Exchange Algorithm alongside auxiliary simulation steps to cancel out intractable terms.
This contribution gives researchers uncertainty quantification and the option to encode prior information during the estimation.
This thesis constitutes both a rigorous statistical modeling endeavor and a core software engineering contribution within an actively developed R package.
Students will join a scientific programming team whose tools are directly used by empirical social scientists, epidemiologists, and economists to analyze human behavior and evaluate policy interventions.