The GCSE results of all students entered for a particular subject at A-Level are used to produce prediction matrices, which are used, along with coursework results, not to predict individual student's results, but to guide examiners and set boundaries for grades.
In 2009 exam regulators OFQUAL, DFES Wales and Northern Ireland CCER commissioned research company NFER to find out whether the matrices were being used consistently across awarding bodies and to explore whether they could be improved.
The study found the matrices were being applied consistently, but recommended tolerance levels, which set the acceptable variation between predicted and actual results, should be adjusted to reflect the numbers of candidates sitting a particular subject.
Examining bodies have to report and justify any wide disparities between predicted and actual grades.
The recommended changes were made in time for this year's A-Levels.
The researchers also believed that students sitting A-Levels for CCER in Northern Ireland
were achieving better results because the examining body used a matrix based solely on the results of Northern Irish students, rather than incorporating data from across England, Scotland and Northern Ireland as the WJEC, Edexcel, AQA and OCR did.
CCER has agreed to investigate this further.
The report also concluded the existence of Welsh as an additional compulsory subject, taken by large numbers of students at GCSE-level in Wales, made no difference to the accuracy of predictions.