Data Analysis: Quantitative Data

Overview

In this open-ended session, we will try to get ChatGPT, Claude, and other LLMs to generate code and graphics  — both classically in R and Bayesianly in Stan — for some of the most common models run in the social and physical sciences, starting with standard regression and working our way up to various nonlinear models (e.g., GAMs, GPs), time-series / panels, and discrete choices (e.g., binary and multinomial logit). The emphasis will be on data exploration, generalizing code to relax assumptions underlying more computationally simple models, and locating bottlenecks in existing code.

Note for participants: Bring your laptop! This session will include use of custom GPTs, accessible through ChatGPT Plus ($20/monthly). Test data will be distributed in advance, but participants can bring some of their own to see if we can analyze it collectively in real-time. It is recommended, but optional, to have RStudio installed on your own laptop or to be able to run it remotely (e.g., on Great Lakes).