Tutoring is almost always evaluated by randomizing students within schools, a design that differences out any spillover onto classmates. We instead randomize 81 Danish schools and 2,583 kindergartners, tutoring a few children in each and testing all. Decoding improved 0.38 standard deviations… more →