The children would be asked to talk about topics that interested them--experiences they had had, movies they had seen, or anything that would lead to genuine, lively communication and discussion. This, he thought, would improve their abilities to reason and communicate logically. He also asked the teachers to give their pupils some practice in measuring and counting things, to assure that they would have some practical experience with numbers.
The results were incredible:
At the beginning of their sixth grade year, the children in the experimental classes, who had not been taught any arithmetic, performed much better than those in the traditional classes on story problems that could be solved by common sense and a general understanding of numbers and measurement. Of course, at the beginning of sixth grade, those in the experimental classes performed worse on the standard school arithmetic tests, where the problems were set up in the usual school manner and could be solved simply by applying the rote-learned algorithms. But by the end of sixth grade those in the experimental classes had completely caught up on this and were still way ahead of the others on story problems.
As far as anyone knows, this is the only experimental evidence on math education that exists. And it was a proper experiment, with different classes randomly placed in different teaching strategies. We know that the experiment was not contaminated by outside tutoring, because all of the experimental classes, both the testing and control groups, were of poor children whose parents were not paying attention to their education. That was the only way he could get away with the experiment, even in the 1930's.
So basically, all available experimental evidence shows that teaching math in elementary school is worse than useless. As a good scientist, I will abandon my preconceived ideas and accept this as fact until more evidence comes in.