Aveek Bhattacharya

aveek18.bsky.social

20 September 2026 17:01

You hit trouble if the model is systematically biased one way or another - for example @michaelcaley.bsky.social has expressed worries that his model undervalues set piece chances because it is trained on pre-meat wall era data - that won't be fixed with bigger sample.

Aveek Bhattacharya

aveek18.bsky.social

20 September 2026 17:01

Sorry "random error" is a somewhat technical term, but the key point is that no model can capture every detail, what makes it a model is that it simplifies. If the error is random - if you're overestimating for some shots and underestimating others, then it should even out with more shots/matches

Aveek Bhattacharya

aveek18.bsky.social

20 September 2026 14:26

Would be interesting to test this, but my guess is that people systematically overestimate how likely players are to score on the eye test. That said, we should recognise there's a lot of random error in these models which is why analysts complain about reading too much into single game xG...

barry novibes

iaindowie.bsky.social

19 September 2026 11:47

look forward to reading this. not enough of the talk this week has been around the safeguarding. we shouldn't know who this lad is, or indeed any academy player before they're seventeen, even if that means changes to the youth cup, which i love.

Tom

tomvictor.bsky.social

08 September 2026 09:15

[explaining midfield reinforcements to an american owner] imagine a berge