Teaching Nonreaders to Read

Kermoian 1962, Feshbach, Adelman, & Fuller 1974), the fear of mislabeling may increase the incidence of false-negative errors, particularly for the severe high-risk child. This problem was seen recently in an unpublished study from our longitudinal project which compared teacher predictions (end of kindergarten) to test predictions, based on the abbreviated test battery (beginning of kindergarten), to reading outcomes at the end of second grade. The results showed that while the overall accuracy of kindergarten teacher predictions was as high as the tests (approximately 80$), the detection of the severe high-risk child was much lower when predicted by the teachers. The teachers identified only 19$ of these children, whereas the tests detected 75$ of them. In other words, the overall teacher predictions were spuriously inflated by 'good outcome' forecasts when the base rates favored such outcomes (by 4:1). However, when they predicted severe outcomes, which was rare, their accuracy was extremely high (approximately 90$).