Statistical constraints on the use of single cases in neuropsychological research