New AI Tool Reveals What Small Classes Really Do for Kids

Most studies of "does this work?" give you one number: the average result. But averages hide a lot. If a new drug helps half of patients a lot and does nothing for the other half, the average says "meh." A new paper from researcher Hugo Gobato Souto introduces a method called Wasserstein Causal Forests that measures how a change shifts an entire spread of outcomes — the low performers, the middle, and the high flyers — instead of flattening everything into one figure. ("Wasserstein" is just a way of measuring how far apart two bell curves sit.)

He pointed it at Project STAR, a famous Tennessee experiment from the 1980s where thousands of students were randomly assigned to small or regular-sized classes. Earlier work found small classes lift math achievement by about 0.16 standard deviations — modest but real. The new method shows that gain is not spread evenly. It is roughly twice as large at the top of the class (+0.179 at the 90th percentile) as at the bottom (+0.091 at the 10th). And in schools with the most low-income students, the boost reached +0.262, versus +0.092 to +0.164 elsewhere. The overall spread of scores barely budged.

Why should you care if you are not a statistician? Because this is the kind of tool that decides where money goes. Smaller classes are expensive — you need more teachers and more rooms. This method helps answer the sharper question: smaller classes for whom? For parents and teachers, it suggests the biggest wins land with strong students and in struggling schools, not uniformly. The same approach could be applied to medicine, hiring, or marketing, where "the average customer" often hides the people a product actually helps.

The catch: the method stumbles when outcomes bunch into multiple separate peaks — it did worse than older tools there. It is also a single preprint, not yet vetted by other scientists, and the school-by-school numbers are descriptive patterns, not proven cause and effect. Treat it as a sharper lens, not a final verdict.