Why we need diverse data to know what works in education

By focusing on what works for students on average, education research risks mistaking important anomalies for noise in the data.