This research team analyzed the administrative data of 12 million Texas students—including more than a billion item responses—to understand which questions best predicted earnings at age 25. By pioneering multiple machine-learning techniques, the researchers package questions into measures of nuanced skills.