[asplake]

Ofqual Chairman Says It Was A "Fundamental Mistake" To Believe Algorithm Grades "Would Ever Be Acceptable" https://www.politicshome.com/news/article/ofqual-algorithm-m...

[rbecker]

"Students will now receive grades based on their teacher’s estimate of what their grade would have been"

But if those estimates are improved using statistics, there's a political fallout. There's some valid criticism of the algorithm used (far less than that BBC article tries to imply), but there's no question the algorithm's estimates were more accurate.

So much ink was spilled calling the algorithm biased for its 4% increase in A-grades for independent schools, yet teacher's 40% increase of grades above the expected average is... what? Unbiased?

On any other topic, such a position would be called "anti-science".

[nicoburns]

> there's no question the algorithm's estimates were more accurate.

Some people were predicted A's and given U's by the algorithm. It might have been less biased as an average. But it's results were nevertheless completely unacceptable.

[asplake]

There’s a fundamental asymmetry here. Fail a student unfairly, and the harm to them is potentially irreparable. Pass more than usual, and you increase competition for places and while there’s certainly some unfairness there, the system will ultimately compensate through interviewing, delayed starts, etc.

[rbecker]

> the system will ultimately compensate through interviewing, delayed starts, etc.

A roundabout way of saying that some students that would have been accepted to their chosen college, won't be, because their grades weren't as inflated as their competitions. Isn't that also potentially irreparable harm, not just "some unfairness"?

[asplake]

There’s a massive difference between not getting your preferred place and being denied the opportunity even to apply

[rbecker]

So you would consider using the algorithmicaly adjusted grades an improvement, if in case of a failing grade, the teacher's estimate was used instead?

And a follow-up question: How many such cases were there? Where the adjusted estimate failed a student, but the teacher's estimate didn't?

[jsmith99]

Well, yes. By a curious coincidence he is also chair of the Centre for Data Ethics and Innovation which is publishing a paper on bias in algorithmic decision making.