Ofqual’s grading algorithm, represented by the equation Pkj = (1-rj)Ckj + rj(Ckj + qkj – pkj), may be short but it reveals much about what went wrong with the grading system. One apparent issue is the algorithm’s simplicity, as it only involves the four variables Ckj, qkj, pkj, and rj, indicating a lack of breadth in the input data. Contrary to popular big data solutions, Ofqual chose to gather the smallest possible amount of data to enable efficient and accurate standardisation.
The first three terms, Ckj, qkj, and pkj, represent various distributions of grades at schools j. Ckj simply involves historical grade distribution over the past three years; hence it is crucial to Ofqual. Regrettably, it may lead to students receiving grades based on the abilities of pupils they have never met, becoming an inherently unfair assumption. For qkj, it uses the predicted grade distribution based on students’ previous attainment at GCSEs, allowing a subsequent prediction of A-Level grades. As for pkj, it offers a predicted grade distribution of previous years based on GCSEs, whose accuracy is essential in determining present grades’ reliability.
Finally, the term rj differs in that it denotes how many students in a class have historical data available, with a value of 1 for perfect tracking and 0 for insufficient data. Multiplying 1-rj and rj by the equation yields two halves that prioritises either historical A-level results or adding this year’s GCSE results and then downgrading them based on previous prediction accuracy. This gives Pkj, the predicted grades for the school.
Flaws within the algorithm sprout from the fact that it exclusively uses GCSE results and short, recent history to determine prior attainment. Additionally, it does not give ample consideration to a student’s genuine success, a factor that could have been necessary to produce accurate results. Further issues arose from the decision to allow small classes to receive teachers’ recommendations and modify the grade boundaries to combat overall grade inflation.
The algorithm also failed to acknowledge that the primary aim is to determine individual student grades, not a distribution for a class. While the complexity of big data solutions may lead to inaccurate grades, Ofqual’s grading algorithm may need revising to avoid transparent errors.
