What counts as evidence for working memory training? Problems with correlated gains and dichotomization