The ethics of algorithmic fairness | by Marta Ziosi | AI for ...

In Scotland:

Moderation was used to identify schools that were “significantly outwith” their historical attainment profile. It was deemed necessary because the SQA found when it compared teacher estimates to the past four years of attainment data there was “significant overestimation” – for example, at Higher, 2020 estimates would have resulted in an A to C pass rate of 88.8 per cent, 14 percentage points higher than 2019, and a significantly higher pass rate than had been seen for each of the past four years. Moderation is used to ensure the outcome of the grade is “fair, valid and reliable” and to ensure consistency of judgements across schools and colleges. Ultimately roughly a quarter (26.2 per cent) of teacher estimates were adjusted [based on an algorithm].

In England:

Teacher assessed grades have not been used to calculate the “vast majority” of GCSE results that students receive later this month, Tes can reveal.  And the same will be true for large numbers of the A-level results being published next week – one exam board has told Tes that in large-entry subjects around 60 per cent of grades will be based purely on statistical modelling [algorithm]. This is because of the methodology being used by Ofqual for this year’s results in both sets of qualifications. Tes has learned that the regulator has decided that where a subject has more than 15 entries in a school, teachers’ predicted grades will not be used as part of the final grade calculation.

Scottish MSM – geddit?