if your Maths A level class is predicted .2% A*s, 95% As, 4.6% Bs and .2% Us, and you have a class of 20, guess what the grade breakdown is
Maybe it'd be 19 As and a B. That's the least error, right? But it deprives that top student of their A*. So maybe it's A*, 18 As and a B. But that gives a student an A* for the same probability that no-one gets a U. So maybe it's A*, 17 As, a B, and a U? But that's too few As.
No: the actual ranking for that class would be 19 As and a U.
Even better: that class could be predicted 0.2% Us *even if no-one in previous years had ever been given a U*.
Here is a real school's data from the linked article:
Past years: 12.5% A* (…) 0% U.
Expected grades after "prior attainment adjustment": 5.71% A*, 2.3% U
Actual marks given to a class of 27, where each pupil is 3.7% of the total: one A* and one U
Your class can be predicted a U even if no previous class has ever got a U. And once your class is predicted a U, someone has to be given it, even if the proportion of the class expected to get a U is less than one student.
WHERE INDEED https://twitter.com/dsquareddigest/status/1294202164248813573?s=20
love the understatement
I've focused on the bottom end of the distribution here but you can also see how exactly the same approach fucks the pupil at the top end. If your class of 20 is predicted 4.9% A*s, then guess what? No-one gets an A*! Doesn't matter how good the top pupil in the class is.
You can follow @alexhern.
Tip: mention @twtextapp on a Twitter thread with the keyword “unroll” to get a link to it.

Latest Threads Unrolled: