Joshua was awarded the Deans' Summer Research Fellowship for summer 2025. Learn more about his time below.
Author: Joshua Dutton
Is there detectable political bias in Large Language Models (LLMs)? And if so, what is the correlation between the political valence of the input and the political valence of the model output? And how does this differ by model? I learned quite a bit about text-as-data methods, and as part of my project, am updating a dataset and reproducing a method from a famous econometric paper (Gentzkow, Shapiro, Taddy 2019). This basically is a method that allows me to predict the political party of a person through their speech. I also discovered several other methods that I am now looking more into, including sentiment analysis as well. After doing lots of data cleaning, my next step is going to be to finally run some regressions!
Part of the process involved a significant amount of webscraping in order to get a complete set of the Congressional Record. Luckily, I had experience with webscraping as part of my job as a research assistant in DEAL, and I had help from others in the DEAL lab whenever I got stuck.
I started getting interested in economics research after taking Econ 101, and hearing some professors come in to speak about their research. After one speaking event, I emailed the professor in question and landed a research job through the Duke Economic Analytics Lab (DEAL).
This project was made possible through the Deans’ Summer Research Fellowship (DSRF)—a unique opportunity for rising juniors and seniors in the Trinity College of Arts & Sciences to pursue funded summer research full-time!
Effective this 2025-2026 Academic year, we are rebranding DSRF to be the Trinity Summer Undergraduate Mentored Research Fellowship (T-SUMR)! If you're pursuing graduation with distinction and are passionate about research, consider applying to T-SUMR in December of your Sophomore or Junior year.