Austin van Loon

Software & Teaching

Software

mixedsubjects (R package)

Design and analyze experiments that combine human subjects with LLM predictions. mixedsubjects implements seven unbiased estimators for mixed-subjects randomized experiments — including the difference-in-predictions estimators introduced in our work — with delta-method and bootstrap inference, plus tools that take pilot data, a budget, and relative costs and recommend an estimator and the optimal split between human and silicon subjects.

install.packages("mixedsubjects")

CRAN·Documentation·GitHub

To cite: run citation("mixedsubjects") in R, and cite Broska, Howes & Van Loon (2025) for the design itself.

Contributions to ppi_py (Python)

The PPI correlation and prediction-powered power analysis introduced in "The Mixed Subjects Design" are implemented in the open-source ppi_py library.

GitHub

Affective meaning ratings app (R/Shiny)

An open-source web application for collecting evaluation–potency–activity (EPA) ratings of concepts, built for "Word embeddings reveal how fundamental sentiments structure natural language" and reusable for extending affect control theory dictionaries.

GitHub

Replication packages

Code and data for my papers are public wherever possible: Imagined Otherness · The Mixed Subjects Design · What Divides Us (measuring schemas of America — companion to the imagined-otherness line) · more on GitHub and OSF.

Teaching

At MIT Sloan

Current (2026–27)

  • AI Foundations — required sessions that build AI literacy for MIT Sloan's MBA and EMBA students: how frontier AI systems work, what they can and cannot do, and how to evaluate them for real use.
  • 15.S23 — AI Builder Space (Fall 2026) — a new course I co-developed and lead, giving students the space, community, and support to build products, ventures, and productivity tools with AI. No technical background required.
  • 15.347 — Designing Empirical Research in the Social Sciences (PhD) — the epistemological and meta-theoretical foundations of social science research, for doctoral students across MIT: What is a theory? How do different methods build on and test theories? What are science's goals, and why should we subscribe to them? What makes an experiment good?
  • Generative AI Hackathon for Social Good (Sloan Innovation Period, 2027) — MBA students build AI-enabled prototypes addressing social challenges, receive mentorship from Sloan alumni, and present to a panel of faculty and practitioners.

For MBA students interested in my research, AI Builder Space and the hackathon below are the best entry points.

Recent

  • 15.280 — Communication for Leaders (core MBA course; Fall 2024, Fall 2025) — persuasive, strategic communication for driving organizational change. The course drew directly on my research: which arguments actually change minds, why disagreement triggers defensiveness, and how to communicate when stakes are high.
  • Generative AI Hackathon for Social Good (co-created with Michiel Bakker; Sloan Innovation Period, 2026) — MBA students build AI-enabled prototypes addressing social challenges, receive mentorship from Sloan alumni, and present to a panel of faculty and practitioners.

Workshops & open materials

  • Introduction to Text Analysis in Python — hands-on tutorial from the SICSS Festival 2021. Video
  • People Analytics — I designed and taught People Analytics courses at Stanford (2022) and Duke (2024); course materials are open. Materials
  • Guest lectures on causal inference and machine learning (Stanford SOC 382). Materials