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

[Final Siren]

Very rough guide, there are too many variables to dissect if you look at a broader picture.

For example, Collingwood last year despite winning the same amount of games never really hit form largely due to a tweak in game plan. So from a fundamental perspective the change was visually so but the outcome very similar.

So how does one measure 'fundamentals' in numbers? Well one could argue disciplined aspects of the game like contested possession, pressure, tackles, DE etc.

I don't know if these things are measured in the squiggle but my point proves it is possible to drop in form for similar outcome, it's also possible to improve for similar or less outcome. To further open the can of worms you'd need to measure things like draw difficulty and opposition form against those 'fundamentals' also.

So how does one correlate those measurements of fundamentals to things like goal kicking accuracy? The can of worms gets deeper doesn't it.

Deeper still, how does one correlate them to other metrics like I50, marks I50, metres gained etc. etc. ?

At some point you have to consider basic fundamentals or 'effort' on how they effect the numbers or the squiggle. But you can't really because good effort doesn't automatically equal good game metrics outcomes.

Models don't take a "bottom-up" approach like you describe - because of the problems that you also describe.

A very simple AFL model that takes only scores and venue as inputs will have no idea about any of these things like disposals, metres gained, etc. It doesn't even know that it's measuring a football match. But it will still reliably outperform most human tipsters.

That's because the scores are the bottom-line. Regardless of how a team gets there, we can safely assume that everyone is trying their best to tip the scores in their favour.

Once a model starts moving away from scores to track things like I50s and disposals, it becomes more prone to misreading a team because their gameplan doesn't align with everyone else's. A model that adjusts for disposals, for example, will underrate Richmond, because Richmond's gameplan is about forcing turnovers and moving very directly to attack, with as few possessions as possible.

So models don't need to account for exactly how a team produces scores - and, in fact, usually shouldn't.

 

[Final Siren]

I would take Cats coming 3rd again this year. I think we'll make the 8, but miss the top 4.

The Cats are interesting, because I wonder if people are extrapolating. Geelong were awesome halfway through last year, then got worse, then lost Tim Kelly... so looking ahead, you might expect that decline to continue, as a couple of wobbly wheels come fully off in 2020.

This might be true - because humans are good at sensing trends in data - but might also not be true, because humans are also good at sensing false trends in data. We detect a lot of false positives.

I don't think any Squiggle-tracked models extrapolate like this - instead, they might make an adjustment for Kelly but otherwise expect the Cats to continue on as they left off last year. And Geelong's finals campaign was not terrible: losses to Collingwood (by 10) and Richmond (by 19), and a win over the Eagles (by 20).

 

[Dr Tigris]

The Cats are interesting, because I wonder if people are extrapolating. Geelong were awesome halfway through last year, then got worse, then lost Tim Kelly... so looking ahead, you might expect that decline to continue, as a couple of wobbly wheels come fully off in 2020.

This might be true - because humans are good at sensing trends in data - but might also not be true, because humans are also good at sensing false trends in data. We detect a lot of false positives.

I don't think any Squiggle-tracked models extrapolate like this - instead, they might make an adjustment for Kelly but otherwise expect the Cats to continue on as they left off last year. And Geelong's finals campaign was not terrible: losses to Collingwood (by 10) and Richmond (by 19), and a win over the Eagles (by 20).

All true. But Geelong's core of top end talent is getting well old, and Kelly was instrumental to their performance. There's always that strong argument, from experience, that teams with an older core suddenly drop off a cliff. Is 2020 that cliff? I have no idea. But that extra thing we all know is something that the models don't take into account. Don't know how they would. Unless something changes dramatically the guys that make a big difference to Geelong will be gone, or much reduced in impact fairly soon. Kelly really changed that by adding depth in quality and probably being their equal best player, with Danger.

 

[HTT]

Tigers surging up and left

 

[caloschwaby]

16.5 minute quarters going to affect offensive/defensive output of teams and therefore their squiggle ratings? Or will it be okay because all the ratings are relative to all the other teams in the single season (who also have lower output)?

 

[HTT]

16.5 minute quarters going to affect offensive/defensive output of teams and therefore their squiggle ratings? Or will it be okay because all the ratings are relative to all the other teams in the single season (who also have lower output)?

Assuming it doesn’t get calibrated to this (and scoring goes down as you’d expect it would) teams go down and right and score predictions will be too high for the whole year because 2019 scores are still being considered, but it shouldn’t affect the ability of the squiggle to pick winners.

 

[HTT]

Assuming it doesn’t get calibrated to this (and scoring goes down as you’d expect it would) teams go down and right and score predictions will be too high for the whole year because 2019 scores are still being considered, but it shouldn’t affect the ability of the squiggle to pick winners.

With that being said, looks like predicted scores have been scaled down, which also explains Richmond’s insane movement.

 

[BlueTucker456]

Squiggle movements wild so far this year.

 

[Jaffa Fan]

Tigers and Pies practicing social distancing.

 

[Final Siren]

With that being said, looks like predicted scores have been scaled down, which also explains Richmond’s insane movement.

Expected scores are scaled down by 20% to match the reduced game time, yeah.

Squiggle is very sensitive to strong defensive performances, so without this (and maybe even with it), it would overreact to low scores.

The big movement though is mainly because Squiggle is very reactive in the early rounds! Especially R1.

 

[Millky95]

Expected scores are scaled down by 20% to match the reduced game time, yeah.

Squiggle is very sensitive to strong defensive performances, so without this (and maybe even with it), it would overreact to low scores.

The big movement though is mainly because Squiggle is very reactive in the early rounds! Especially R1.

Squiggle loves Port now because of that defensive display

 

[HTT]

Hefty amount of down and right action on the squiggle tonight

 

[RW]

Expected scores are scaled down by 20% to match the reduced game time, yeah.

Squiggle is very sensitive to strong defensive performances, so without this (and maybe even with it), it would overreact to low scores.

The big movement though is mainly because Squiggle is very reactive in the early rounds! Especially R1.

At least our 2005 flag side will have some company... it's pretty lonely being a defensive outlier

 

[HTT]

West Coast now predicted to be second last, with a 12.6% chance of missing finals. Squiggle ahead of the curve again or off the mark?

 

[Billy ray]

Love my squiggle movement up and to the right

 

[manicmagpie]

Collingwood’s defence basically off the chart. Excellent.

 

[Final Siren]

West Coast now predicted to be second last, with a 12.6% chance of missing finals. Squiggle ahead of the curve again or off the mark?

That was a temporary technical issue - it has the Eagles at 12th now.

Was a good weekend's results for the preseason tips! I copped the most stick over the offseason for having Port high, West Coast mid-tier, and Adelaide 17th, but those all look a bit more plausible now.

 

[Final Siren]

Collingwood’s defence basically off the chart. Excellent.

Two games and their opposition has scored 34 and 32 points, so yeah, that's pretty exceptional - even in a season where the average score is only 68.

 

[BrainOfMorbius]

So lets say there are no games to be played in Victoria for the rest of the year, including finals. Does that impact Squiggle, and its finals tips?

 

[Final Siren]

So lets say there are no games to be played in Victoria for the rest of the year, including finals. Does that impact Squiggle, and its finals tips?

Yes, the venue matters. Although there's still so much uncertainty around how strong teams are; that's the real question. Home advantage is a thing, but it's not everything. I'd care a lot more about figuring out exactly how good the Bulldogs are than where they'll play.

So far this year, Squiggle has been operating with home advantage set to only 30% of regular levels, to account for the lack of crowds. (Crowds being the major source of home advantage, as far as I know.) That will dial back up as crowds return. It still won't make a difference to games at neutral venues, like Geelong v Brisbane @ SCG tonight, but will be a factor once big crowds attend true home games.

Of course, we still don't know what the fixture will look like! At the moment, Squiggle assumes all unannounced games will be played at neutral venues, although it seems more likely that non-Victorian sides will get some games with proper home advantage.

 

[richoatthedisco]

Yes, the venue matters. Although there's still so much uncertainty around how strong teams are; that's the real question. Home advantage is a thing, but it's not everything. I'd care a lot more about figuring out exactly how good the Bulldogs are than where they'll play.

So far this year, Squiggle has been operating with home advantage set to only 30% of regular levels, to account for the lack of crowds. (Crowds being the major source of home advantage, as far as I know.) That will dial back up as crowds return. It still won't make a difference to games at neutral venues, like Geelong v Brisbane @ SCG tonight, but will be a factor once big crowds attend true home games.

Of course, we still don't know what the fixture will look like! At the moment, Squiggle assumes all unannounced games will be played at neutral venues, although it seems more likely that non-Victorian sides will get some games with proper home advantage.

Will you assign home advantage to Richmond at the Gabba, where we haven't lost for 16 years, and are likely to have far more supporters than the Swans?

 

[Final Siren]

Will you assign home advantage to Richmond at the Gabba, where we haven't lost for 16 years, and are likely to have far more supporters than the Swans?

No, even though that might be correct.

I don't do team+venue-specific numbers because I'm not confident I can separate signal from noise. Some other good computer models do, though, including Matter of Stats and Swinburne. So they're responsive to trends like the one you mention, but Squiggle isn't.

At the Gabba, for example, the Swans and Tigers have played there only twice each since 2016. So you have to put a lot of stock in few results if you're changing your tips based on the historical record.

On the crowd, I would use that as an input if I had a nice unbiased way to project relative numbers of fans.

 

[BEaston]

Yes, the venue matters. Although there's still so much uncertainty around how strong teams are; that's the real question. Home advantage is a thing, but it's not everything. I'd care a lot more about figuring out exactly how good the Bulldogs are than where they'll play.

So far this year, Squiggle has been operating with home advantage set to only 30% of regular levels, to account for the lack of crowds. (Crowds being the major source of home advantage, as far as I know.) That will dial back up as crowds return. It still won't make a difference to games at neutral venues, like Geelong v Brisbane @ SCG tonight, but will be a factor once big crowds attend true home games.

Of course, we still don't know what the fixture will look like! At the moment, Squiggle assumes all unannounced games will be played at neutral venues, although it seems more likely that non-Victorian sides will get some games with proper home advantage.

How do you know crowd is the major source of HGA as opposed to other factors like familiarity, ground dimensions, travel impact, etc?

 

[LicoriceAllsorts]

How do you know crowd is the major source of HGA as opposed to other factors like familiarity, ground dimensions, travel impact, etc?

Agreed. I think crowd is nowhere near that big of an impact. Especially in the AFL. In others sports yes, but in the AFL not really

 

[Final Siren]

How do you know crowd is the major source of HGA as opposed to other factors like familiarity, ground dimensions, travel impact, etc?

I don't, definitively, but it's what seems to be the case both in AFL and other sports around the world. There have been a lot of articles and studies, and they generally lean pretty consistently (but not overwhelmingly) toward the idea of crowd-generated HGA - probably "noise of affirmation"-style influencing umpiring decisions, as well as some psychological effects on players.

Other factors that get frequently cited don't seem to hold up when you remove crowds. For example, travel is hard to justify when Kardinia is so close to Melbourne. Unique ground dimensions don't seem to work at the SCG. Ground familiarity is very very closely tied to fan support, because teams have the most fans at venues they play the most - so I actually use this, but I don't believe it's the ground itself that makes the difference.