Connor T. Jerzak

Excited to teach two grad courses @UTAustin Spring '26!

📊 Gov 385La: Making Big Data – Learn web scraping, crowd-sourcing, & data curation to build high-quality datasets for social sci impact.

🌍 Gov 385Lb: Causal ML & EO for Social Sci – Deploy satellite data & ML for causal inf at planetary scale

Connor T. Jerzak, Stephen A. Jessee: Attenuation Bias with Latent Predictors https://arxiv.org/abs/2507.22218 https://arxiv.org/pdf/2507.22218 https://arxiv.org/html/2507.22218

Join a cool group at UC Berkeley, March 7-8, 2026, for "Clouds, Streams, and Ground (Truths)"—exploring methods to study algorithmic music ecosystems. Details: www.algorithmicmusicmethods.com

YouTube video by Planetary Causal Inference - Academic Content A First Course in Planetary Causal Inference: Confounding - Adel Daoud at IC2S2 2025

A First Course in Planetary Causal Inference: Confounding. Adel presents.

www.youtube.com/watch?v=oD7D...

Connor T. Jerzak, Priyanshi Chandra, Rishi Hazra
Selecting Optimal Candidate Profiles in Adversarial Environments Using Conjoint Analysis and Machine Learning
https://arxiv.org/abs/2504.19043

🚨 New preprint! Excited to share our work on extracting and evaluating the potentially many feature descriptions of language models

👉 arxiv.org/abs/2506.15538

Nicolas Audinet de Pieuchon, Adel Daoud, Connor T. Jerzak, Moa Johansson, Richard Johansson
Benchmarking Debiasing Methods for LLM-based Parameter Estimates
https://arxiv.org/abs/2506.09627

Norrköping campus at sunset

We are hiring postdocs in Computational Social Science
📍SweCSS, Norrköping, Sweden
⏰Deadline June 3
🔗https://liu.se/en/work-at-liu/vacancies/26854
Please apply // help us spread the word

For how long? 2-3 yr contract & flexible remote.

Where? Remote options possible; core mentor, Adel, based in Sweden.

AI & Global Dev Lab: aidevlab.org

New postdoc position posted at the AI & Global Development Lab!

What would you do? Use ML & causal inference to evaluate African development programs—build remote-sensing pipelines for geo-temporal poverty estimates & impact evals, among other things.

Details: liu.se/en/work-at-l...

Cool stuff!

Equal parts Mars, Venus, and Titan.

globalleadershipproject.net

On scaling laws for muli-scale dynamics in effect heterogeneity estimation.

Paper: arxiv.org/abs/2411.02134

More: planetarycausalinference.org

Rankings of neighborhoods/provinces in Africa with the greatest improvement in poverty between 1990 and 2019. Estimation methodology is EO-ML-based.

Rankings: aidevlab.org/rankings/

Raw data now on Hugging Face: huggingface.co/datasets/cje...

Earth Observation Meets Adam Smith: Earth Observation AI in Policy Innovation for Global Development👇
planetarycausalinference.org/earth-observ...

YouTube video by Connor T. Jerzak's Academic Content Can Large Language Models (or Humans) Disentangle Text? - Official Video

Can LLMs (or humans) Disentangle Text? Nice video by one of our PhD students Nicholas Audinet de Pieucheron + collaborators (thanks @jerzakconnor.bsky.social!): www.youtube.com/watch?v=CW7H...

Course intro vid:

More: connorjerzak.com/teaching/

New graduate course for Fall 2025 in ATX:

Topics in Causal Machine Learning and Earth Observation for Social Science

Brian Libgober and Connor T. Jerzak. Linking datasets on organizations using half a billion open-collaborated records. Political Science Research and Methods. Published online 2024:1-20. doi:10.1017/psrm.2024.55 Abstract Scholars studying organizations often work with multiple datasets lacking shared identifiers or covariates. In such situations, researchers usually use approximate string (“fuzzy”) matching methods to combine datasets. String matching, although useful, faces fundamental challenges. Even where two strings appear similar to humans, fuzzy matching often struggles because it fails to adapt to the informativeness of the character combinations. In response, a number of machine learning methods have been developed to refine string matching. Yet, the effectiveness of these methods is limited by the size and diversity of training data. This paper introduces data from a prominent employment networking site (LinkedIn) as a massive training corpus to address these limitations. By leveraging information from the LinkedIn corpus regarding organizational name-to-name links, we incorporate trillions of name pair examples into various methods to enhance existing matching benchmarks and performance by explicitly maximizing match probabilities. We also show how relationships between organization names can be modeled using a network representation of the LinkedIn data. In illustrative merging tasks involving lobbying firms, we document improvements when using the LinkedIn corpus in matching calibration and make all data and methods open source.

Linking datasets on organizations using half a billion open-collaborated records

Brian Libgober (@blibgober.bsky.social) & Connor T. Jerzak (@jerzakconnor.bsky.social) via PRSM @psrm.bsky.social doi.org/10.1017/psrm... #polisky #NUResearch

APSR will start inviting replications for a random subset of accepted papers. I really like this as well as the constructive tone around it 👏 (from the latest Notes from the Editors)

planetarycausalinference.org

📊Do you have multiple datasets lacking shared identifiers and do you want to merge them?

➡️ @blibgober.bsky.social & @jerzakconnor.bsky.social introduce a massive training corpus and discuss various methods to enhance existing matching benchmarks www.cambridge.org/core/journal... #FirstView

A new method helps us combine satellite data at different zoom levels (from house to village) to better understand how different people will respond to anti-poverty aid: arxiv.org/abs/2411.02134

If you're a #HarvardGov concentrator, consider taking our Gov 94 seminar (Making Big Data) this spring. The tools covered should be useful to those interested in research, policy, and the intersection of tech and governance 📈