Instructors at 2- and 4-year colleges! Give your students free access to a virtual tutor optimized for learning by taking part in our study on using AI to improve education. Instructors receive a $1,000 honorarium.

Apps received by 8/8 receive priority: pingpong.hks.harvard.edu/eduaccess

(1/2) Our own Johann Gaebler and @5harad.com, w/coauthors @seanjwestwood.bsky.social & Shanto Iyengar, analyzed 50+ years of US TV news using a novel LLM-based classification system. They find a steep decline in in-depth political coverage & substantive reporting, while soft news & commercials rise.

NEW in Management Science!

My coauthors and I came up with a new consequentialist approach to designing equitable algorithms.

Instead of imposing fairness criteria on an algorithm (like equal false negative rates), we aim for good outcomes.

More in the 🧵 below! (1/)

🚨 Excited to share our new article in @annualreviews.bsky.social. Working with Kristin Linn, @5harad.com, Amol Navathe, and Ravi Parikh, we examine the fairness debates of seven prominent and controversial healthcare algorithms.🧵 madisoncoots.com/files/racial...

There are, of course, lots of important caveats to this work. For example, the results change if we move to a resource-constrained setting. But we hope our work illustrates the importance of looking at utility, not simply model miscalibration, in these debates.

As a result, the overall clinical utility of using race in these models is often surprisingly small. It still might make sense to include race, but the benefits of doing so have probably been overstated.

And because people for whom decisions flip are necessarily close to the decision boundary, their *utility* for intervening vs. not is comparable. (In the shared decision-making settings we consider, the utility is 0 at the boundary, i.e., the boundary is set to be the point of indifference.)

But despite this miscalibration, *decisions* (e.g., to intervene in some way) based on race-aware vs. race-unaware risk estimates are largely the same, as very few individuals are close to the decision boundary.

We find that race-unaware risk estimates are indeed often *miscalibrated*, systematically over or underestimating risk for different groups. Such miscalibration is often cited as evidence that including race improves the quality of predictions for all groups.

Using race in medical risk assessments is hotly debated, with some arguing that doing so improves accuracy while others worry it reinforce pernicious attitudes. With @madisoncoots.com and colleagues, we identify a statistical twist that's been largely overlooked. 5harad.com/papers/race-...

Calling instructors at 2- and 4-year colleges!

Give your students free access to a virtual tutor optimized for learning by taking part in our study on using AI to improve education. Learn more and sign up by Dec. 16: zurl.co/WitA

Our virtual tutor, called PingPong, is based on ChatGPT. It's designed to help students learn, not simply give them answers to homework problems.

We'll help customize PingPong to your course content.

Instructors can view de-identified student interactions.

PingPong is multilingual.

Instructors at 2- and 4-year colleges! Give your students free access to a virtual tutor optimized for learning by taking part in our study on using AI to improve education. Instructors receive a $1,000 honorarium.

Sign up by Dec. 16! pingpong.hks.harvard.edu/eduaccess

Naviance is widely used by high school students to help them decide where to apply to college. In our new PNAS paper, Sabina Tomkins, Josh Grossman, Lindsay Page and I show it can inadvertently dissuade qualified students from applying to selective schools. www.pnas.org/doi/10.1073/...

Our paper "The Measure and Mismeasure of Fairness" — long in draft form — is now out in JMLR! We show that common error rate measures are often misleading indicators of algorithmic bias, and argue it's better to evaluate algorithms by looking at their effects. 5harad.com/papers/fair-...

Call for papers! Behavioral Science & Policy is running a special issue on behavioral insights for AI policy. Great opportunity to reach policymakers.

Initial submissions just require a 500-word abstract by Dec. 1.

Please help us get the word out!

behavioralpolicy.org/wp-content/u...

In a new randomized experiment at the Santa Clara County Public Defender Office, my colleagues and I found that text message reminders reduce *incarceration* for missed court dates by over 20%! More in the 🧵 below. alexchohlaswood.com/assets/paper... 1/11

We find that Asian American applicants — especially South Asians — are much less likely to be admitted to selective colleges than white students with similar test scores, GPAs, and extracurricular activities.

Much of the gap is due to geography and preferential treatment of legacy applicants.

New faculty position at the intersection of politics and computing at MIT
https://academicjobsonline.org/ajo/jobs/25219