Certified Legendary Thread The Squiggle is back in 2023 (and other analytics)

[The_Steadier]

10 was just a random point with a decent sample size (plus I couldnt be bothered looking up exactly how many years weve had 18 sides for), and it'd still be more accurate than 7.1 will be.

Go 15 or 20 years if you like, it'd be much the same I'd imagine.

Why stop there?

The point is that you have the answer that you want- and are then going back through the historical data to confirm what you want to hear.

What if going back 30 years contradicts the answer you want to hear?

Your solution to simply only then use 10/15/20 years of data (ie the data that supports what you want to hear) is a classic example of how you can use statistics to "prove" anything

 

[tiger_tough]

Why stop there?

The point is that you have the answer that you want- and are then going back through the historical data to confirm what you want to hear.

What if going back 30 years contradicts the answer you want to hear?

Your solution to simply only then use 10/15/20 years of data (ie the data that supports what you want to hear) is a classic example of how you can use statistics to "prove" anything

It won't, and guaranteed its still closer than 7.1.

Go back to the start of the AFL era, I bet it's still the same. At some point you have an adequate enough sample size to make a prediction, surely, rather than saying 'we know this is extremely likely to be wrong, but here it is'.

I dont 'want to hear' anything. Just curious. My logical brain doesn't comprehend using an algorithm you know is going to be off at either end of the ladder.

Not having a dig, just trying to understand.

 

[Rourke]

make a prediction, surely, rather than saying 'we know this is extremely likely to be wrong, but here it is'.

I dont 'want to hear' anything. Just curious. My logical brain doesn't comprehend using an algorithm you know is going to be off at either end of the ladder.

Not having a dig, just trying to understand.

The single Squiggle predictive ladder is geared towards individual clubs, and how well they might go in 2025. Most algorithms also "regress to the mean" by 20-30%, which allows for uncertainty but pushes the worst ones up. After that, Squiggle puts the clubs in order of their predicted wins. That looks like a win-loss ladder, but it isn't.

The display never predicts the number of wins by a club, given that it finished 18th (a conditional variable). You have to do that with simulations, and the answer --- even with the same assumed team strengths --- is closer to 4 wins than 7. We just don't ask the computer that question, because most people don't care about it.

 

[peetoo]

It’s very kind to the hawks this year. Will be unbeaten until they play brisbane away in r24.

 

[tiger_tough]

The single Squiggle predictive ladder is geared towards individual clubs, and how well they might go in 2025. Most algorithms also "regress to the mean" by 20-30%, which allows for uncertainty but pushes the worst ones up. After that, Squiggle puts the clubs in order of their predicted wins. That looks like a win-loss ladder, but it isn't.

The display never predicts the number of wins by a club, given that it finished 18th (a conditional variable). You have to do that with simulations, and the answer --- even with the same assumed team strengths --- is closer to 4 wins than 7. We just don't ask the computer that question, because most people don't care about it.

Thanks for explaining, that makes more sense. I obviously didn't realise the wins column on squiggle is basically irrelevant.

edit - I have another question. Why does it place Eagles 17th, when they have the same number of predicted wins as Richmond, and a better percentage? And if Eagles have the same predicted wins AND a better percentage, why is their finishing range of 16-18 much worse than Richmond's (12-18), and even Norths (14-18)?

2nd edit - Reading the Q&A's - I think this must be because of where other models have those 3 teams finishing?

 

[Simon_Nesbit]

It’s very kind to the hawks this year. Will be unbeaten until they play brisbane away in r24.

I've always said it was the best prediction system out there....