Five laboratories set out to repeat 21 influential scientific studies of recent years. Only 13 held up.