Teaching
Instats Webinars
A 3-day hands-on webinar that takes social scientists with no programming background from their first line of Python to a language model they fine-tune themselves. Day 1 teaches Python fundamentals through record linkage, matching messy datasets by spelling and by meaning. Day 2 models the themes in a corpus of party manifestos with LDA and the Structural Topic Model, then opens up BERT-style encoders and builds the inputs they need. Day 3 fine-tunes a left-right classifier with SetFit, tests it against a bag-of-words baseline, and saves it for reuse. All sessions run live in Google Colab and build on the R and Stata habits participants already have.
A 3-day intensive webinar that trains social scientists with no machine learning background to use large language models for measurement and data work. Day 1 introduces stance-detection encoders and generative decoders, and when to choose each. Day 2 builds a defensible quantitative proxy via hypothesis ensembles, gold-set coding, and validation. Day 3 applies generative models to crawling and data wrangling. All sessions run live in Google Colab with demos on real social science text.
Teaching Record
Instructor
Teaching Assistant
Course Proposals
International political economy has spent the past two decades rebuilding its account of trade governance around firms. This course teaches that literature and applies it to the rules being written now. It starts with the bargaining literature on host states and foreign investors and the firm-level turn that supplied the evidence. It then covers the domains where corporate influence over public rules is most visible and most contested. These include intellectual property, labor and environmental standards in supply chains, corporate taxation, subsidies and development finance, economic security and investment screening, and the export controls and industrial policy now built around semiconductors and artificial intelligence. Students work with lobbying disclosures, trade and investment data, regulatory comment dockets, and corporate filings, and test claims about corporate influence against the public record.
This course trains students to use large language models (LLMs) for real-world data workflows. Students learn to clean, link, and structure complex textual data using both discriminative and generative models. The course emphasizes selecting and applying appropriate open-source models, such as LayoutLM, SentenceTransformers, DeBERTa, and GPT, for specific analytical tasks including data cleansing, record linkage, stance detection, fine-tuning, and structured data extraction from unstructured sources. Students also gain practical experience leveraging GPU computing resources through Google Colab for efficient model training and inference.
This course is designed to familiarize students with the empirical tools widely used in political science in particular and in social science more broadly. Throughout this course, students will learn data analysis and statistical inference techniques using R and RStudio, including descriptive statistics, hypothesis testing, quasi-experimental methods, and regression analysis.