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

[Pessimistic]

I haven't in the past but will start doing so this year.

The modeling community has been split on this issue, with half of us using normal HGA and the other half dropping it for the Grand Final, since until recently, non-Victorian teams regularly overpowered Victorian ones for the flag, which lead to the idea that more neutral crowds and the greater relative importance of form-based factors meant HGA was reduced or eliminated. But we've had a spate of surprisingly strong GF performances from Victorian teams over the last five years, so the tide is turning.

In the absence of good evidence to the contrary, we should assume that Grand Finals are like any other game, so I'll be using the usual method of calculating home ground advantage for the 2018 GF.

Also hawthorns threepeat was the second third and fourth grand final in succession.
Not only a home ground but a home grand final ground.

 

[HTT]

Oh I agree. You too.

Wonder what Squiggle would do if it was a North/Eagles GF?

The squiggle would make a tip for it

The squiggle is not a sentimental being
It does not care if it is wrong
It does not try to troll certain clubs
It is an algorithm
It just makes predictions based on past scores

I, however, am a different case. If North Melbourne make the Grand Final, I will curl up in a ball of shame, and never make a pre season prediction ever again.

 

[Final Siren]

Final Siren

May have been noted/explained before, but the finals have Stadium Australia in there for the Swans.

Ah, thanks. I mucked up the Swans while fixing a bug earlier whereby it expected Giants' home finals to be at the SCG.

 

[Final Siren]

Go back to the start of the season and Squiggle was saying that you were bugger all chance of finishing any higher than 11th.

So it's changed its prediction based on observations over the first 10 weeks of the season. As have the media, and mug punters.

Everyone got you guys wrong. And us. It's a nice feeling, enjoy the ride.

North this year is a good example of the value of a model like Squiggle, in that it's often a week or two ahead of common perception. Sometimes it's wrong and sometimes it's right very early, but most of the time, you get about 1-2 weeks head start. That's not heaps, but it's handy.

Nearly everyone now recognizes that the Roo revival is real, but Squiggle got there a bit quicker than most. There was one hiccup, when it dumped them down to 18th for their Round 1 loss to Gold Coast, when North only managed 5 goals in the wet, but after Round 2, they were predicted to finish 12th, then 10th after Round 4, and since then they've hung around either just inside or just outside the Top 8. They're now 6th.

I see that kind of thing a lot, especially in the first half of seasons. People either get carried away too early or take too long to realize a bad team has come good.

 

[Forward Press]

North this year is a good example of the value of a model like Squiggle, in that it's often a week or two ahead of common perception. Sometimes it's wrong and sometimes it's right very early, but most of the time, you get about 1-2 weeks head start. That's not heaps, but it's handy.

Nearly everyone now recognizes that the Roo revival is real, but Squiggle got there a bit quicker than most. There was one hiccup, when it dumped them down to 18th for their Round 1 loss to Gold Coast, when North only managed 5 goals in the wet, but after Round 2, they were predicted to finish 12th, then 10th after Round 4, and since then they've hung around either just inside or just outside the Top 8. They're now 6th.

I see that kind of thing a lot, especially in the first half of seasons. People either get carried away too early or take too long to realize a bad team has come good.

From a biased personal point of view how did Squiggle rate the Eagles from Round 1?

 

[Final Siren]

From a biased personal point of view how did Squiggle rate the Eagles from Round 1?

Pre-season: 12th
R1: 12th
R2: 10th
R3: 9th
R4: 8th
R5: 10th
R6: 8th
R7: 4th
R8: 2nd
R9: 2nd
R10: 2nd

You can rewind/play through their ratings over time in Tower etc at https://live.squiggle.com.au/

 

[Forward Press]

Thanks.

I guess like most of us on the club board it was only after the Port game that we started believing (there's a big thread on predicting how we'd go in the Port(home) - GWS(away) - Richmond(home) - Hawthorn(away) run which we thought was a true barometer of what we really are and no-one predicted 4-0).

 

[Suma Magic]

Ah, thanks. I mucked up the Swans while fixing a bug earlier whereby it expected Giants' home finals to be at the SCG.

No worries.

And a question about Champion Data's model:



Do you know the differences between the Squiggle and your model? Not sure if this has been answered before.

They give the Tigers a 19% chance of finishing on top whereas Squiggle gives them about a 70% chance.

 

[Final Siren]

No worries.

And a question about Champion Data's model:

View attachment 504985

Do you know the differences between the Squiggle and your model? Not sure if this has been answered before.

They give the Tigers a 19% chance of finishing on top whereas Squiggle gives them about a 70% chance.

The CD model is a black box so I don't know anything about its internals, or how well it tends to do historically.

They have Richmond unusually low - only one other well-known model has the Tigers finishing 3rd at the moment, with four pegging them at 2nd and three (including Squiggle) at 1st.

By contrast Squiggle is more bullish on the Tigers than most (all?) models, who tend to have Richmond, West Coast, and Melbourne all close together.

These projections can change easily at this point in the season, though, as they're dependent on a high number of unknown results.

 

[Pessimistic]

Are all rounds equal? of course they are, 4 points only.

But does the position of a round in the season have a bigger weight on the predicted outcome

eg in round 11, you have won 4 from 6, and are 3 wins behind contention.
So to get 14 wins, you need 10 from the remaining 12. 83% your return up till now is 40% - a difference of 44%
If you had won 5 from 5, you need only 9, which is 75%, and your return is 50% - a much less difference of 25%

the win in round 10 is over a 10% difference in outcomes. which round is the most decisive?

 

[Pessimistic]

The CD model is a black box so I don't know anything about its internals, or how well it tends to do historically.

They have Richmond unusually low - only one other well-known model has the Tigers finishing 3rd at the moment, with four pegging them at 2nd and three (including Squiggle) at 1st.

By contrast Squiggle is more bullish on the Tigers than most (all?) models, who tend to have Richmond, West Coast, and Melbourne all close together.

These projections can change easily at this point in the season, though, as they're dependent on a high number of unknown results.

CD provide the premiership quarter - an attack defence x-y distribution similar to squiggle but much simplified. this could be similar

 

[Hobbes]

Collingwood +20 v Western Bulldogs
Richmond +51 v St Kilda
Brisbane v Sydney +12
Geelong +38 v Carlton
GWS +7 v Essendon
Hawthorn v West Coast +14
Melbourne +12 v Adelaide
Fremantle v North Melbourne +11

1. Melbourne 29.7 (+2)
2. West Coast 28.5
3. Richmond 24.8 (-2)
4. North Melbourne 15.8 (+1)
5. Sydney 13.4 (-1)
6. Port Adelaide 12.2 (+1)
7. Geelong 10.3 (+1)
8. Collingwood 6.3 (+1)
9. Adelaide 4.5 (-3)
10. Hawthorn 3.0
11. Essendon -1.5 (+3)
12. Brisbane -9.2 (+1)
13. Fremantle -9.6 (-2)
14. GWS -12.8 (-2)
15. Gold Coast -18.2 (+1)
16. Western Bulldogs -19.0 (-1)
17. St Kilda -20.3
18. Carlton -24.7

Melbourne on top thanks to the massacre in Alice Springs.

Sydney +43 v Carlton
Western Bulldogs v Melbourne +49
Hawthorn v Port Adelaide +3
Gold Coast v Geelong +22
Essendon v Richmond +26
West Coast +58 v St Kilda
North Melbourne +31 v Brisbane
Colingwood +25 v Fremantle
Adelaide +26 v GWS

Only one match looks even close this week.

1. West Coast 18.2 (+1)
2. Richmond 17.1 (-1)
3. Melbourne 16.8
4. North Melbourne 14.37 (+2)
5. Sydney 14.36
6. Port Adelaide 14.33 (-2)
7. Geelong 12.9 (+1)
8. Collingwood 12.3 (+2)
9. Hawthorn 12.2
10. Adelaide 11.9 (-3)
11. Essendon 9.0 (+2)
12. Fremantle 8.7
13. GWS 8.3 (-2)
14. Gold Coast 6.6 (+2)
15. Western Bulldogs 6.3 (-1)
16. Brisbane 5.9 (-1)
17. Carlton 4.6
18. St Kilda 4.3

 

[Forward Press]

Interesting that North is only rated a 35% chance of making finals. I think the rest of their fixture isn't that hard and should be better than a one third chance. Seems CD has written off the Dockers and Dogs and rates Essendon as a small chance.

Looking forward to revisiting in three months...

 

[Hobbes]

Interesting that North is only rated a 35% chance of making finals. I think the rest of their fixture isn't that hard and should be better than a one third chance. Seems CD has written off the Dockers and Dogs and rates Essendon as a small chance.

Looking forward to revisiting in three months...

I looked up the bookie odds and they're about 4/6. Definitely better than 50% to play finals.

 

[Forward Press]

I looked up the bookie odds and they're about 4/6. Definitely better than 50% to play finals.

I wonder if you can bet against North making finals because if so I think there's some money to be made if we think Champion Data are making sense.

 

[richoatthedisco]

I wonder if you can bet against North making finals because if so I think there's some money to be made if we think Champion Data are making sense.

Betfair. 'Lay' option.

 

[Janus]

As of R10 (actual ladder position in italics):

1. (1) West Coast (1.704)
2. (3) Melbourne (1.583)
3. (5) North Melbourne (1.442)
4. (2) Richmond (1.431)
5. (6) Geelong (1.388)
6. (8) Collingwood (1.339)
7. (4) Sydney (1.338)
8. (7) Adelaide (1.294)

9. (9) Port Adelaide* (1.170)
10. (10) Hawthorn (1.098)
11. (11) GWS (1.086)
12. (12) Essendon (1.082)
13. (13) Fremantle (0.990)
14. (17) Brisbane (0.916)
15. (15) Gold Coast* (0.815)
16. (16) St Kilda (0.691)
17. (14) Western Bulldogs (0.681)
18. (18) Carlton (0.458)

 

[A Swallow]

Interesting that North is only rated a 35% chance of making finals. I think the rest of their fixture isn't that hard and should be better than a one third chance. Seems CD has written off the Dockers and Dogs and rates Essendon as a small chance.

Looking forward to revisiting in three months...

I found that very odd. We have a very favourable draw from here on out, and would need to go at 50% wins or worse to finish 9th-10th.

Brisbane x 2
Dogs x 2
GC
St Kilda
Essendon
Collingwood
Sydney (Etihad)
Geelong
WC (Hobart)
Adelaide

It’s obviously possible, but if we go 7-4 after this weekend it seems unlikely.

 

[Isaac Cumming No 1]

I found that very odd. We have a very favourable draw from here on out, and would need to go at 50% wins or worse to finish 9th-10th.

Brisbane x 2
Dogs x 2
GC
St Kilda
Essendon
Collingwood
Sydney (Etihad)
Geelong
WC (Hobart)
Adelaide

It’s obviously possible, but if we go 7-4 after this weekend it seems unlikely.

Getting the Dogs twice just doesn't seem fair

 

[Final Siren]

Interesting that North is only rated a 35% chance of making finals. I think the rest of their fixture isn't that hard and should be better than a one third chance. Seems CD has written off the Dockers and Dogs and rates Essendon as a small chance.

Looking forward to revisiting in three months...

Champion Data have been low on the Roos all year, much more pessimistic than every model out there I've seen. Maybe they'll turn out to be right, but so far they've looked like they're lagging, i.e. CD will peg the Roos where everyone else had them a couple of weeks earlier.

Maybe they just don't like North! They were also pessimistic last year, which turned out to be correct.

Here's another look at CD's pre-season predictions compared to another model. So far CD look to have been wrong on Richmond, West Coast, and North, but right on Melbourne.

 

[Final Siren]

Are all rounds equal? of course they are, 4 points only.

But does the position of a round in the season have a bigger weight on the predicted outcome

eg in round 11, you have won 4 from 6, and are 3 wins behind contention.
So to get 14 wins, you need 10 from the remaining 12. 83% your return up till now is 40% - a difference of 44%
If you had won 5 from 5, you need only 9, which is 75%, and your return is 50% - a much less difference of 25%

the win in round 10 is over a 10% difference in outcomes. which round is the most decisive?

This could be a very complicated question. In its simplest form, though, the earlier your team can deliver a strong performance, the better.

Of course, if it turns out to be a one-off, not an indicator of improving form, it doesn't really matter. But normally a good game means a better chance of winning all future games, too, so you want as many future games as possible.

A team that only pulls out the stops late in the season can turn out like Geelong 1993, who went on a post-Round 16 rampage only to miss finals by percentage -- in what was probably the most even season of the last 50 years. (The Cats finished with 12 wins and the minor premier, Essendon, with 13.5.) I believe Geelong could very well have won the flag that year if they'd only clicked into gear one game sooner.

Then again, you could be Hawthorn 2008, running into peak form just ahead of finals and riding it all the way to a premiership. A key point here, though, is that the Hawks were always assured of a finals spot - in fact, they never looked like missing the top 2. In that case, there's no benefit from winning more Home & Away games, and since you can't peak forever, you want to save your best for last.

 

[A Swallow]

This could be a very complicated question. In its simplest form, though, the earlier your team can deliver a strong performance, the better.

Of course, if it turns out to be a one-off, not an indicator of improving form, it doesn't really matter. But normally a good game means a better chance of winning all future games, too, so you want as many future games as possible.

A team that only pulls out the stops late in the season can turn out like Geelong 1993, who went on a post-Round 16 rampage only to miss finals by percentage -- in what was probably the most even season of the last 50 years. (The Cats finished with 12 wins and the minor premier, Essendon, with 13.5.) I believe Geelong could very well have won the flag that year if they'd only clicked into gear one game sooner.

Then again, you could be Hawthorn 2008, running into peak form just ahead of finals and riding it all the way to a premiership. A key point here, though, is that the Hawks were always assured of a finals spot - in fact, they never looked like missing the top 2. In that case, there's no benefit from winning more Home & Away games, and since you can't peak forever, you want to save your best for last.

Squiggle puts more weight on early season games right? If North had flogged GWS and then beaten Freo comfortably in Perth in rounds 1 and 2, would we have moved more aggressively in Squiggle’s eyes?

 

[Final Siren]

Squiggle puts more weight on early season games right? If North had flogged GWS and then beaten Freo comfortably in Perth in rounds 1 and 2, would we have moved more aggressively in Squiggle’s eyes?

Yep, that's right. Teams can change a lot over the off-season, so Squiggle is more reactive to early-season form.

 

[SJ]

The CD model is a black box so I don't know anything about its internals, or how well it tends to do historically.

They have Richmond unusually low - only one other well-known model has the Tigers finishing 3rd at the moment, with four pegging them at 2nd and three (including Squiggle) at 1st.

By contrast Squiggle is more bullish on the Tigers than most (all?) models, who tend to have Richmond, West Coast, and Melbourne all close together.

These projections can change easily at this point in the season, though, as they're dependent on a high number of unknown results.

The Champion Data model is built upon player values and forecasts injury probabilities to take into account the values of replacement players.

 

[richoatthedisco]

Champion Data uses a lot of subjectivity for a supposed objective analysis.