Although I am not an expert on current methods used to evaluate options for teaching children, Richard Feynman does a good job of explaining why education successes are often difficult to generalize to a broader population.
…take education. Some guy comes along and he sees the way people teach mathematics and he says, “I have a better idea. I’ll make a toy computer and teach them with it.” So he tries it with a group of children, he hasn’t got a lot of children, maybe somebody gives him a class to try it with. He loves what he’s doing. He’s excited. He understands completely what his thing is. The kids know that it’s something new, so they’re all excited. They learn very, very well and they learn the regular arithmetic better than the other kids did. So you make a test–they learn arithmetic. Then this is registered as a fact–that the teaching of arithmetic can be improved by this method. But it’s not a fact, because one of the conditions of the experiment was that the particular man who invented it was doing the teaching. What you really want to know is, if you just had this method described in a book to an average teacher (and you have to have average teachers; there are teachers all over the world and there must be many who are average), who then gets this book then tries to teach it with the method described, will it be better or not? In other words, what happens is that you get all kinds of statements of fact about education, about sociology, even psychology–all kinds of things which are, I’d say, pseudoscience…They’ve done experiments which are not really controlled experiments.
Richard Feynman, from The Pleasure of Finding Things Out.