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

[Dr Tigris]

Yes, there's one game in particular where the Hawks kept the Giants to 4.5 in Sydney - Squiggle looooved that and it's responsible for a lot of the Hawks' high rating.

I've gone back and tinkered with it a bit, but I just can't find a way to stop that kind of movement without harming the performance of the model. That is, anything I try to do to cap or scale down extraordinarily large movement off a single game makes the model worse as a predictor. So for now, at least, the Hawks keep their rating.

A model is just that, a model of reality. it isn't reality. So that the Hawks are so highly rated by the Squiggle says that they underperformed in 2019. But it might also be a false positive - because we know more about what happened than the computer we know that the Hawks in general weren't that good. We'll see next year. I'm not writing them off, but I'm also not seeing them as top 4 certainties.

 

[Dr Tigris]

Yeah keen to get a grasp on it further. I normally graph this stuff out on paper and it takes a while and is not as in depth so thank god I found this. Going to save me a bit of hassle.

Graphing on paper!!!

OK, I am now very impressed.

The Squiggle website is pretty cool. You get some idea of what's going on and to some extent why. The man, is the Man, Mr Max Barry.

 

[Final Siren]

A model is just that, a model of reality. it isn't reality. So that the Hawks are so highly rated by the Squiggle says that they underperformed in 2019. But it might also be a false positive - because we know more about what happened than the computer we know that the Hawks in general weren't that good. We'll see next year. I'm not writing them off, but I'm also not seeing them as top 4 certainties.

Nothing's a certainty, that's for sure.

I do expect most models to rate the Hawks pretty high - in the Top 4 or close to it - because they had a ripper finish to the season. From Round 16, the Hawks defeated Collingwood, Geelong, West Coast, and GWS - a Grand Finalist, two prelim finalists, and a semi-finalist - plus delivered a 5-goal win over Fremantle and a 12-goal win over Gold Coast. They dropped games to Brisbane and North Melbourne in that stretch, but even so, that's a quality packet of games.

edit: Also the Hawks should have beaten the Eagles again in Round 15 - they kicked 9.17 and lost by 6. I'm still salty about that because I tipped them.

 

[Dr Tigris]

Nothing's a certainty, that's for sure.

I do expect most models to rate the Hawks pretty high - in the Top 4 or close to it - because they had a ripper finish to the season. From Round 16, the Hawks defeated Collingwood, Geelong, West Coast, and GWS - a Grand Finalist, two prelim finalists, and a semi-finalist - plus delivered a 5-goal win over Fremantle and a 12-goal win over Gold Coast. They dropped games to Brisbane and North Melbourne in that stretch, but even so, that's a quality packet of games.

edit: Also the Hawks should have beaten the Eagles again in Round 15 - they kicked 9.17 and lost by 6. I'm still salty about that because I tipped them.

The Hawks are a bit of a mystery to me. At their best they are so very good. But so often they have been ordinary. Not sure if Clarko got them into his preferred shape at the end of the season and that's what we'll see in 2020. Or they hit great form for a short period, and that's it. Still they are the best performed non-finalist by far. That is often a good sign of things to come.

And on models. I'm always a sceptic unless I can understand the actual logic of why things are happening. Maths can easily bamboozle, even the people doing the maths.

 

[the harry]

The Hawks are a bit of a mystery to me. At their best they are so very good. But so often they have been ordinary. Not sure if Clarko got them into his preferred shape at the end of the season and that's what we'll see in 2020. Or they hit great form for a short period, and that's it. Still they are the best performed non-finalist by far. That is often a good sign of things to come.

And on models. I'm always a sceptic unless I can understand the actual logic of why things are happening. Maths can easily bamboozle, even the people doing the maths.

The best thing about models I think is that they'll come to the same conclusion if they have the same data each time. People don't do that. They might be tired, have a bias against someone, have lost money on a team previously ect. People are just unreliable when assessing large amounts of data

 

[Dr Tigris]

The best thing about models I think is that they'll come to the same conclusion if they have the same data each time. People don't do that. They might be tired, have a bias against someone, have lost money on a team previously ect. People are just unreliable when assessing large amounts of data

Absolutely.

It's just blind faith in models that bothers me. Many models are fantasy land, not tethered to reality in any real way. e.g. most (almost all) predictions of the future are wrong. They usually predict the future will look like a continuation of the past, and when a change happens, even a predictable one, boom. The Squiggle is deeply embedded in the reality of the AFL. So I trust it to tell me something useful. Many other models - Hmmmm

 

[Ok Boomer]

A real-quick first pass at a 2020 forecast, accounting for fixture and off-season player movement:

View attachment 772668​

This still needs work and I haven't checked it for grievous errors! Buuuuut:

2019 was a super-even year, surpassing 2016 and 2017, which were themselves considered to be very even years, compared to the dark Hawthorn/Collingwood/Geelong eras that came before. This means that a whole bunch of teams are rated quite closely to one another, and it will be more of a crapshoot than usual as to who winds up on top.

To measure player movement, Squiggle relies on AFL Player Ratings, which are pretty good but also definitely worse than expert human observation. For example, I have watched Richmond ruckman Ivan Soldo quite a bit, and I think he's a much better player than his terrible AFL Player Rating score suggests. But Player Ratings are objective and cover every single player.

Squiggle rates Fremantle as having the most new talent to pack into its Round 1 side compared to what they were putting out late last year. (This considers players returning from injury, retirements, delistings, and trades in/out). Next are Gold Coast, and then, further back, Melbourne, the Bulldogs, and St Kilda.

At the other end of the scale, Squiggle sees Adelaide as having a worse Best 22 in Round 1 next year than they put out late this year. They are the only team rated in the negatives that way. Other teams with relatively little scope for improvement are Brisbane, North Melbourne, Richmond, and - despite Tim Kelly -West Coast.

West Coast were never really rated by Squiggle in 2019, so that's what the the mid-table finish reflects, rather than anything unusual from player movement or the fixture.

Similarly, Essendon were rated very poorly by the end of 2019, and Squiggle doesn't see why they'll get better.

Conversely, Squiggle's weird love affair with Port Adelaide continues.

Hawthorn and the Bulldogs both finished off 2019 very well, leaving them well-placed for an assault on 2020.

How long have you been taking into account player ratings with the squiggle?

Do you think it is possible that it has lead to some teams, including West Coast, being expected to perform much better week to week than reality?

 

[Pessimistic]

A model is just that, a model of reality. it isn't reality. So that the Hawks are so highly rated by the Squiggle says that they underperformed in 2019. But it might also be a false positive - because we know more about what happened than the computer we know that the Hawks in general weren't that good. We'll see next year. I'm not writing them off, but I'm also not seeing them as top 4 certainties.

Tbh the hawks havent gone that well against richmond for a while now

They finished 2019 well in results, but most importanly didnt fade in quarters and in games, a problem earlier on

This with a hamstrung midfield which incidentally gets Tom Mitchell back

 

[the harry]

How long have you been taking into account player ratings with the squiggle?

Do you think it is possible that it has lead to some teams, including West Coast, being expected to perform much better week to week than reality?

This year it was our accuracy that won us several games & round 23 partly validated the doubts that people had with us.

 

[Final Siren]

How long have you been taking into account player ratings with the squiggle?

Do you think it is possible that it has lead to some teams, including West Coast, being expected to perform much better week to week than reality?

Since mid-2018. It doesn't make as much difference as you might think, but it does seem to be mildly positive in terms of tipping.

Mainly what it does is:

• Nudge a tip from one team to the other in games that are very hard to split but one team has big Ins or Outs; and
• Inflate or deflate the significance of a result based on the opponent's Ins & Outs - e.g. beating a team with several stars missing won't deliver the same improvement to your Squiggle rating as if you'd beaten them at full strength.

Squiggle's pre-2019 ladder prediction took off-season player movement into account for the first time, and was this. Without player movement, it would have been this - similar but a bit worse. In particular, the player movement correctly led it to shift West Coast down from 2nd to 4th (actual: 5th) and Sydney from 12th to 15th (actual: 15th). It did incorrectly push up Melbourne from 5th to 2nd (actual: 17th), which bumped Geelong out of the top 4 (actual: 1st), but was better overall even so.

 

[Final Siren]

FYI this is how Squiggle sees the 2020 fixture in terms of Home Ground Advantage.

No consideration to strength of double-up opponents here -- just advantage derived from who plays where.

​

Breaking it down by individual games:



Note:

This is based on Squiggle's Ground Familiarity Model. See further down for details. It naturally tends to assign greater HGA to non-Victorian teams because they are more familiar with their opponents' home grounds than their opponents are with theirs. You may or may not agree with this methodology.

Factual observations:

Every team plays at least one game in Victoria and Western Australia.

Collingwood are the only team to not play in South Australia.

Melbourne and West Coast don't visit NSW.

The most games in a single state is 17, with Collingwood, Carlton, and the Bulldogs in Victoria. In the Bulldogs' case, one of these is in Geelong and two more are in Ballarat; Collingwood and Carlton only play in Melbourne. Two teams have 16 games in Victoria (all in Melbourne): Richmond and Essendon.

The least number of games played in a home state is Gold Coast with 11 in Queensland. Unless you want to treat Geelong as its own state, in which case it's Geelong with 9. All other teams have at least 12 games in their home state.

The most number of games at a single venue is 14, shared by three teams: Richmond (MCG), Collingwood (MCG), and St Kilda (Docklands). Melbourne have 13 MCG games. Five teams have 12 games at a single venue: West Coast (Perth), Adelaide (AO), the Bulldogs (Docklands), Port Adelaide (AO), and Fremantle (Perth).

The least number of games at a primary home venue is 8: GWS at Sydney Showgrounds and Hawthorn at the MCG. Three teams have only 9 games at their primary venue: Geelong (Kardinia), Essendon (Docklands), and North Melbourne (Docklands).

The team playing at the most number of different venues (by far) is Gold Coast with 12. Most teams play at 8 or 9 different venues. Two teams play at only 7 venues: Richmond and Essendon. Two teams play at only 6 different venues: Collingwood and Carlton.

The most number of different areas (treating China, Geelong, and the NT as separate areas to the states) is 8, shared by three teams: Gold Coast, Adelaide, and St Kilda. Gold Coast and Adelaide are the only teams to play in all 6 states plus the NT and Geelong. The least number of different areas is 4, with Collingwood only playing in Melbourne, Perth, Sydney, and Brisbane.

Model observations:

North, like most Docklands tenants, are hampered by the fact that when they play non-Victorian teams away, they travel to venues and states they only rarely visit, but when the roles are reversed, the interstate side is often quite familiar with Docklands (and travelling to Melbourne in general). Also, the Kangaroos spread their home games across two states, Victoria and Tasmania, which means they never build up the kind of overwhelming bank of familiarity that is enjoyed by most other teams.

Melbourne give away a stack of HGA by hosting Adelaide in the Northern Territory instead of at the MCG, and are additionally dragged to Tasmania by North in what would otherwise be a game at a fairly neutral venue.

Similarly, Gold Coast give away home advantage by hosting St Kilda in the NT.

In all of Geelong's games, they are either at some kind of disadvantage (MCG, Docklands, non-Vic) or else have a very large advantage (Kardinia).

Adelaide dodge a game of major disadvantage by playing Melbourne in the NT.

Similarly, Port Adelaide play away to St Kilda in China, which is a relatively neutral venue for what would other be a major disadvantage.

West Coast, like all non-Victorian teams, enjoy a mild benefit from playing multiple away games at somewhat familiar Victorian venues. Additionally they are hosted by Carlton at the MCG, which isn't Carlton's home ground.

GWS only have to travel to play non-Victorian sides twice. That is, they have 20 games in NSW and Victoria. This is a very good thing, under a ground familiarity model, since it allows them to build up familiarity with the same venues.

Ground Familiarity Model

There are a few different well-respected methods of calculating Home Ground Advantage; this is Squiggle's Ground Familiarity model, which scores teams based on how frequently they've played at the same ground and the same state relative to each other in recent years, plus a small modifier for interstate travel. It's a conservative and reliable method that treats all teams the same, and doesn't attempt to measure how well individual teams play individual grounds (which is fraught with danger).

In reality, the major factor in Home Ground Advantage seems to be crowd noise (which may influence umpiring and player psychology). This is hard to measure directly, so a model such as Ground Familiarity acts as a proxy for this - teams tend to have larger, noisier supporters when they're playing at grounds they play a lot.

It's not perfect by any measure, though, and like any assessment of Home Ground Advantage, should be best treated as a general guide - accurate to within a goal or so, but probably not much more than that.

 

[Pessimistic]

So while the AFL give gold coast recruiting and salary cap leg up, they screw them in the fixture

 

[Final Siren]

So while the AFL give gold coast recruiting and salary cap leg up, they screw them in the fixture

It's better than last year! In 2018 the Suns didn't have a home ground for half a year, and even played a home game against Fremantle in Perth. Now that was a seriously disadvantageous fixture. (The Suns' chief exec Mark Evans said he was "delighted" and "couldn't be happier" about the commercial side of it, which gives you an idea of Gold Coast's priorities.)

Fixturing is actually pretty equitable in the AFL - at least, before you get to the Grand Final, and until you figure out how tough your double-up games really were.

 

[Dr Tigris]

FYI this is how Squiggle sees the 2020 fixture in terms of Home Ground Advantage.

No consideration to strength of double-up opponents here -- just advantage derived from who plays where.

View attachment 781340​

Breaking it down by individual games:

View attachment 781346

Note:

This is based on Squiggle's Ground Familiarity Model. See further down for details. It naturally tends to assign greater HGA to non-Victorian teams because they are more familiar with their opponents' home grounds than their opponents are with theirs. You may or may not agree with this methodology.

Factual observations:

Every team plays at least one game in Victoria and Western Australia.

Collingwood are the only team to not play in South Australia.

Melbourne and West Coast don't visit NSW.

The most games in a single state is 17, with Collingwood, Carlton, and the Bulldogs in Victoria. In the Bulldogs' case, one of these is in Geelong and two more are in Ballarat; Collingwood and Carlton only play in Melbourne. Two teams have 16 games in Victoria (all in Melbourne): Richmond and Essendon.

The least number of games played in a home state is Gold Coast with 11 in Queensland. Unless you want to treat Geelong as its own state, in which case it's Geelong with 9. All other teams have at least 12 games in their home state.

The most number of games at a single venue is 14, shared by three teams: Richmond (MCG), Collingwood (MCG), and St Kilda (Docklands). Melbourne have 13 MCG games. Three teams have 12 games at a single venue: West Coast (Perth), Adelaide (AO), the Bulldogs (Docklands), Port Adelaide (AO), and Fremantle (Perth).

The least number of games at a primary home venue is 8: GWS at Sydney Showgrounds and Hawthorn at the MCG. Three teams have only 9 games at their primary venue: Geelong (Kardinia), Essendon (Docklands), and North Melbourne (Docklands).

The team playing at the most number of different venues (by far) is Gold Coast with 12. Most teams play at 8 or 9 different venues. Two teams play at only 7 venues: Richmond and Essendon. Two teams play at only 6 different venues: Collingwood and Carlton.

The most number of different areas (treating China, Geelong, and the NT as separate areas to the states) is 8, shared by three teams: Gold Coast, Adelaide, and St Kilda. Gold Coast and Adelaide are the only teams to play in all 6 states plus the NT and Geelong. The least number of different areas is 4, with Collingwood only playing in Melbourne, Perth, Sydney, and Brisbane.

Model observations:

North, like most Docklands tenants, are hampered by the fact that when they play non-Victorian teams away, they travel to venues and states they only rarely visit, but when the roles are reversed, the interstate side is often quite familiar with Docklands (and travelling to Melbourne in general). Also, the Kangaroos spread their home games across two states, Victoria and Tasmania, which means they never build up the kind of overwhelming bank of familiarity that is enjoyed by most other teams.

Melbourne give away a stack of HGA by hosting Adelaide in the Northern Territory instead of at the MCG, and are additionally dragged to Tasmania by North in what would otherwise be a game at a fairly neutral venue.

Similarly, Gold Coast give away home advantage by hosting St Kilda in the NT.

In all of Geelong's games, they are either at some kind of disadvantage (MCG, Docklands, non-Vic) or else have a very large advantage (Kardinia).

Adelaide dodge a game of major disadvantage by playing Melbourne in the NT.

Similarly, Port Adelaide play away to St Kilda in China, which is a relatively neutral venue for what would other be a major disadvantage.

West Coast, like all non-Victorian teams, enjoy a mild benefit from playing multiple away games at somewhat familiar Victorian venues. Additionally they are hosted by Carlton at the MCG, which isn't Carlton's home ground.

GWS only have to travel to play non-Victorian sides twice. That is, they have 20 games in NSW and Victoria. This is a very good thing, under a ground familiarity model, since it allows them to build up familiarity with the same venues.

Ground Familiarity Model

There are a few different well-respected methods of calculating Home Ground Advantage; this is Squiggle's Ground Familiarity model, which scores teams based on how frequently they've played at the same ground and the same state relative to each other in recent years, plus a small modifier for interstate travel. It's a conservative and reliable method that treats all teams the same, and doesn't attempt to measure how well individual teams play individual grounds (which is fraught with danger).

In reality, the major factor in Home Ground Advantage seems to be crowd noise (which may influence umpiring and player psychology). This is hard to measure directly, so a model such as Ground Familiarity acts as a proxy for this - teams tend to have larger, noisier supporters when they're playing at grounds they play a lot.

It's not perfect by any measure, though, and like any assessment of Home Ground Advantage, should be best treated as a general guide - accurate to within a goal or so, but probably not much more than that.

Just looking at it from the tigers perspective it's fascinating that they are negative on net points, but have a very positive number of games where they have an advantage. So depending on the lens they have a big home advantage or none at all. Similar to Collingwood, but they have an advantage both ways.

And North get screwed every which way. So their paranoia is justified.

 

[Final Siren]

Just looking at it from the tigers perspective it's fascinating that they are negative on net points, but have a very positive number of games where they have an advantage. So depending on the lens they have a big home advantage or none at all. Similar to Collingwood, but they have an advantage both ways.

And North get screwed every which way. So their paranoia is justified.

Collingwood and Richmond both have many somewhat-positive games by virtue of the fact that they play often in Melbourne (and at the MCG in particular). Even when they visit a Docklands-based team, it's often to face someone who also plays away from Docklands quite a bit. So under this model, they do pretty well, even though they have a large disadvantage when travelling interstate.

This is what Richmond 2020 looks like to Squiggle's HGA model:

RICHMOND:  -4.9

 +10.2   R4: Adelaide @ M.C.G. (VIC)
  +9.2  R11: Sydney @ M.C.G. (VIC)
  +9.2  R14: West Coast @ M.C.G. (VIC)
  +8.9  R16: Port Adelaide @ Marvel Stadium (VIC)
  +8.7   R7: Greater Western Sydney @ M.C.G. (VIC)
  +4.1  R12: North Melbourne @ M.C.G. (VIC)
  +4.0   R9: Geelong @ M.C.G. (VIC)
  +3.0  R19: Western Bulldogs @ M.C.G. (VIC)
  +2.1   R1: Carlton @ M.C.G. (VIC)
  +2.0  R22: Hawthorn @ M.C.G. (VIC)
  +1.9  R21: Carlton @ M.C.G. (VIC)
  +1.6  R10: Essendon @ M.C.G. (VIC)
  +1.3   R6: Melbourne @ M.C.G. (VIC)
  +1.1   R3: St Kilda @ Marvel Stadium (VIC)
  -0.0  R17: Collingwood @ M.C.G. (VIC)
  -0.1   R2: Collingwood @ M.C.G. (VIC)
 -11.5  R18: Gold Coast @ Carrara (QLD)
 -11.6  R20: Adelaide @ Adelaide Oval (SA)
 -12.0   R8: Brisbane Lions @ Gabba (QLD)
 -12.1  R23: Fremantle @ Perth Stadium (WA)
 -12.4   R5: West Coast @ Perth Stadium (WA)
 -12.5  R15: Greater Western Sydney @ Sydney Showground (NSW)

 

[Dr Tigris]

Collingwood and Richmond both have many somewhat-positive games by virtue of the fact that they play often in Melbourne (and at the MCG in particular). Even when they visit a Docklands-based team, it's often to face someone who also plays away from Docklands quite a bit. So under this model, they do pretty well, even though they have a large disadvantage when travelling interstate.

This is what Richmond 2020 looks like to Squiggle's HGA model:

RICHMOND:  -4.9

+10.2   R4: Adelaide @ M.C.G. (VIC)
  +9.2  R11: Sydney @ M.C.G. (VIC)
  +9.2  R14: West Coast @ M.C.G. (VIC)
  +8.9  R16: Port Adelaide @ Marvel Stadium (VIC)
  +8.7   R7: Greater Western Sydney @ M.C.G. (VIC)
  +4.1  R12: North Melbourne @ M.C.G. (VIC)
  +4.0   R9: Geelong @ M.C.G. (VIC)
  +3.0  R19: Western Bulldogs @ M.C.G. (VIC)
  +2.1   R1: Carlton @ M.C.G. (VIC)
  +2.0  R22: Hawthorn @ M.C.G. (VIC)
  +1.9  R21: Carlton @ M.C.G. (VIC)
  +1.6  R10: Essendon @ M.C.G. (VIC)
  +1.3   R6: Melbourne @ M.C.G. (VIC)
  +1.1   R3: St Kilda @ Marvel Stadium (VIC)
  -0.0  R17: Collingwood @ M.C.G. (VIC)
  -0.1   R2: Collingwood @ M.C.G. (VIC)
-11.5  R18: Gold Coast @ Carrara (QLD)
-11.6  R20: Adelaide @ Adelaide Oval (SA)
-12.0   R8: Brisbane Lions @ Gabba (QLD)
-12.1  R23: Fremantle @ Perth Stadium (WA)
-12.4   R5: West Coast @ Perth Stadium (WA)
-12.5  R15: Greater Western Sydney @ Sydney Showground (NSW)

Results pretty much what you'd expect. But that big chunk of no real dis/advantage games is pretty large. What you would expect of an MCG tenant I guess.

It's an interesting one. Would you rather more games with a smaller advantage or less games with a larger advantage?

 

[Final Siren]

Results pretty much what you'd expect. But that big chunk of no real dis/advantage games is pretty large. What you would expect of an MCG tenant I guess.

It's an interesting one. Would you rather more games with a smaller advantage or less games with a larger advantage?

I think what you really want is home games vs teams near you on the ladder, and away games vs teams you'll probably lose to or beat no matter where you play.

That way you have a better chance of HGA converting into actual wins.

 

[Dr Tigris]

I think what you really want is home games vs teams near you on the ladder, and away games vs teams you'll probably lose to or beat no matter where you play.

That way you have a better chance of HGA converting into actual wins.

So for the Tigers how does that translate?

Being a bugger (and hoping all works out in 2020), top 4 is teams near us, and everyone else is teams we should beat. Unfortunately it never works the way you expect.

 

[Billy ray]

My poor team.

 

[tazaa]

Just looking at it from the tigers perspective it's fascinating that they are negative on net points, but have a very positive number of games where they have an advantage. So depending on the lens they have a big home advantage or none at all. Similar to Collingwood, but they have an advantage both ways.

And North get screwed every which way. So their paranoia is justified.

I want to * Tasmania off for this reason
Majorly

 

[LittleG]

I want to fu** Tasmania off for this reason
Majorly

IF you relocated there, you would get a massive home field advantage.

 

[tazaa]

IF you relocated there, you would get a massive home field advantage.

Why would we

 

[Maylandsman]

Why would we

Because your clubs is destined to die a death by a thousand cuts if you keep your head in the ground. Kangaroos are quickly becoming irrelevant and should be worried further when the suns potentially hit their straps.

 

[tazaa]

Because your clubs is destined to die a death by a thousand cuts if you keep your head in the ground. Kangaroos are quickly becoming irrelevant and should be worried further when the suns potentially hit their straps.

Ok boomer

 

[The Punter]

Ok boomer

Nah, he played for your mob.