In his report, "Testing the Test", out this week, Professor John Gardner, head of the school of education, includes statistics showing that 12,500 of the 17,600 pupils who sat the test last year could be wrongly classified.
This includes more than two-thirds of the 6,500 candidates who gained a grade A. Professor Gardner also complained that the process was shrouded in secrecy.
He found that "bunching" results meant that it was impossible to distinguish between an A-grade and a bottom grade. Only 18 marks out of 150 straddle the six grades whereas the normal statistical error is 20 marks. In one test, for example, 74 per cent of the candidates correctly answered at least 70 per cent of the questions.
As a result, a cadidate who scored 105 out of 150 in English, mathematics, science and technology would still have received the bottom grade, D, while those who scored 123 gained an A.
"This means that children with 70 per cent of the answers correct would have 'failed'. To be given a 'failing' grade with such a high proportion of correct answers is simply unheard of and is very difficult to justify," he said.
The current test system has been in place since 1993. Professor Gardner also complained of secrecy, with no information on candidates' scores, rank order, for example. "It is not clear why this has been the case, but doubt about the effectiveness of the various tests must at least be an element."
The work pre-empts research on selection commissioned by the Department of Education from a team at Queen's and the University of Ulster, which excluded study of the tests themselves.