More than half of teachers' forecasts for their pupils' GCSE and A-level grades are wrong, and staff have a tendency to be too optimistic about results, exam board research indicates.
The findings from the OCR board, shared exclusively with TES, show that accuracy is worse now than in previous years. More importance is set to be placed on teachers' predictions of student performance in the wake of the government's overhaul of A-levels.
Fewer schools and colleges are expected to offer AS-levels in the next few years, because the reforms mean the awards will no longer count towards A-level grades. Universities will therefore have less information on which to base offers to students, so are expected to place greater emphasis on teachers' predictions.
Headteachers' leaders are warning that the "decoupling" of AS- and A-levels, which begins in September, will lead to even less accurate teacher forecasts.
"In the past, teachers could use AS-level grades to help predict A-level results, because they counted for half the marks," said Malcolm Trobe, deputy general secretary of the Association of School and College Leaders. "The new A-levels are likely to be more difficult to predict accurately in the first few years."
Some school leaders fear that the reduced use of AS-levels will result in more parents pressurising teachers to give their children better predicted grades than they deserve, in order to attract university offers.
One headteacher, who asked not to be named, said: "Youngsters and their parents will put pressure on teachers to give the highest possible estimate. At the moment a teacher can [refuse to predict an A] by telling a student, `You've only got a B at AS-level'. That will be lost when the AS-levels have gone."
The OCR study of almost 170,000 teacher forecasts - which were submitted to the exam board to inform its grade-awarding process - finds that the latest forecasts are less accurate than in previous years.
The research shows that just 43 per cent of A-level forecasts were correct in 2014, down from 48 per cent in 2012 when previous figures were collected. Forty-four per cent of GCSE forecasts were correct in 2014, down from 46 per cent in 2013.
Mr Trobe said he was "surprised" that the proportion of inaccurate forecasts was so high, adding that, in his experience as a headteacher, just one in every six or seven predictions had been wrong. He added that universities' increasing reliance on teacher predictions could lead to more students missing out on their first choice of university because their final results were lower than predicted.
However, Mr Trobe said, new rules allowing English universities to recruit an unlimited number of undergraduate students from the UK and the EU could mean that students who fell short of their predicted grades were still offered a place.
The OCR researchers said the drop in accuracy could have been caused by the scrapping of January exams, which used to inform the forecasts that teachers submitted in May.
Their study shows that 43 per cent of the A-level forecasts in 2014 were higher than students' final grades, while 14 per cent were lower. For GCSEs, 42 per cent of forecasts were higher than the actual results and 14 per cent were lower.
The study finds that grammar and independent schools are better than non-selective schools at forecasting A-level grades, and are also the least optimistic (see panel, below). The researchers said this could be because pupils at selective schools tended to achieve higher grades, and higher grades were easier to predict.
Dr Tom Benton, principal research officer at Cambridge Assessment, told TES that the study highlighted how difficult it was to predict grades.
"It isn't that teachers are particularly poor at predicting; it's that grades are difficult to predict," he said. Statisticians' best efforts at forecasts for individual students - based on GCSE results - had only a 40 per cent accuracy rate, he added.
"One reason [for the difficulty in forecasting] is that grade boundaries are not that far apart," Dr Benton said. "Being able to predict which side of the boundary a student might fall is just a difficult thing to do."
Sylke Scheiner, director of assessment standards at OCR, told TES that the findings "underline the need for more teachers to become examiners so they can really understand the nuts and bolts of how their students can perform".
Separate figures published last year by Ucas, which processes students' university applications, show that only 21 per cent of pupils who were expected by teachers to get grades ABB ultimately matched or bettered that prediction in 2014, compared with 32.2 per cent four years previously.
Playing the guessing game
Independent schools are the best at predicting their students' A-level grades, the OCR research shows. In 2014 they produced correct forecasts for 50 per cent of their pupils.
Grammar schools came second, with 47 per cent of their forecasts proving to be correct.
Next were academies at 42 per cent, followed by comprehensives and sixthform colleges, which were right in 41 per cent of cases.
FE colleges came last, with just 36 per cent of their forecasts proving correct. FE colleges were also the most likely to be overly optimistic about their students' performance: 53 per cent of predictions were higher than the grades ultimately achieved.