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

[Simon_Nesbit]

Thanks, I did a bad. Fixed now.

I'm going to add a thing where you can enter scores and see what would happen.

Thank you thank you thank you. I've been asking for the "what if" story for ever.

If possible, make it like a tipping - tipping the current round moves all squiggles as if real results, recalculates season and next results, etc.

Would also love the option to use "actual score/points" rather than a % for wins (even if this is less accurate mathematically, it helps with what if scenarios)

 

[JayJ20]

The squiggle predicts a Showdown Grand Final 28 days out from the start of the 2018 AFL season.

Eddie's reaction will be priceless if this happens.

You interstate supporters have this warped view about Victorians and thinking we hate everything outside Victoria.

Showdown grand final would be great. I'd have it 10/10 times over some Victorian match ups.

 

[BigVic]

I love a Showdown decider on neutral territory. No home ground advantage and it will be anyone's game

 

[Billy ray]

6-16, 69.9%
i'll take that

 

[Dirty Bird]

Subscribing

 

[dazbroncos]

That would explain Geelong, a very ordinary team who has been found out in finals recently being ranked so high at 5th. Home ground advantage is great during the H&A but when the real shit starts, the Cats fall into their own shitboxes.

Clearly from a team that knows its shitboxes.

GO Catters

 

[RedmanWasHere]

Squiggle is back.

YAY!

New host but same goodness.

 

[Final Siren]

Got some more updates ahead of season 2018:

(1) Pre-season tracking: We had something resembling a normal preseason this year, so Squiggle is using those results in its calculations for the season ahead.

(2) Round-based Sensitivity: Squiggle is now more sensitive in the early rounds and in finals, to better capture the important signalling of surprising results at these times. In particular, an unexpected Round 1 result can cause a team to move a very long way.

(3) Goal-kicking Accuracy: As flagged last year, since goalkicking accuracy seems to be non-reproducible -- teams that are highly accurate one week aren't more likely to be highly accurate the next -- Squiggle will interpret a scoreline like 13.20 as a team that was unlucky to not win by more, and thus will rate them more highly, while a scoreline like 12.3 will be interpreted as a team that got closer than they probably deserved.

(4) Home Ground Advantage: Gone is the practice of assigning 12 points of home advantage to interstate hosts and ignoring everything else. Squiggle now calculates HGA based on relative ground & state familiarity: It scores each team on how many times they've played at the same venue and (less significantly) in the same state over recent years, and assigns HGA points based on how lopsided the numbers are. So at venues where both teams have played fairly often, ground familiarity is balanced and there's not much HGA, if any. At venues where one teams plays a lot and the other team rarely even visits the state, HGA is high.

The main difference in practice is better modelling of Geelong in Melbourne, along with uncommon situations like Melbourne playing in the Northern Territory. For example, for the Round 1 Melbourne vs Geelong game at the MCG, under the old model Squiggle would give Melbourne 12 points of HGA, but under the new model, it's only 2.5 points, due to the Cats' fairly good familiarity with that ground.

Actually here are the current HGA values for all R1 matches:

• Richmond +1.1 vs Carlton @ MCG
• Essendon +7.9 vs Adelaide @ Docklands
• St. Kilda +7.6 vs Brisbane @ Docklands
• Port Adelaide +10.0 vs Fremantle @ Adelaide Oval
• Gold Coast +10.3 vs North Melbourne @ Cazalys
• Hawthorn -1.0 vs Collingwood @ MCG
• GWS +9.0 vs Bulldogs @ UNSW Canberra
• Melbourne +2.5 vs Geelong @ MCG
• West Coast +10.9 vs Sydney @ Perth Stadium

Hawthorn are the only home team to have negative HGA. West Coast still get a bunch of HGA despite playing on a brand new ground due to the state-based numbers.

 

[treefingers]

Do you think you could run a parallel squiggle that keeps things the old way as well? It would help for comparison and to see if there's any flaws in the new methodology.

I like the ideas though.

 

[threenewpadlocks]

Got some more updates ahead of season 2018:

(1) Pre-season tracking: We had something resembling a normal preseason this year, so Squiggle is using those results in its calculations for the season ahead.

(2) Round-based Sensitivity: Squiggle is now more sensitive in the early rounds and in finals, to better capture the important signalling of surprising results at these times. In particular, an unexpected Round 1 result can cause a team to move a very long way.

(3) Goal-kicking Accuracy: As flagged last year, since goalkicking accuracy seems to be non-reproducible -- teams that are highly accurate one week aren't more likely to be highly accurate the next -- Squiggle will interpret a scoreline like 13.20 as a team that was unlucky to not win by more, and thus will rate them more highly, while a scoreline like 12.3 will be interpreted as a team that got closer than they probably deserved.

(4) Home Ground Advantage: Gone is the practice of assigning 12 points of home advantage to interstate hosts and ignoring everything else. Squiggle now calculates HGA based on relative ground & state familiarity: It scores each team on how many times they've played at the same venue and (less significantly) in the same state over recent years, and assigns HGA points based on how lopsided the numbers are. So at venues where both teams have played fairly often, ground familiarity is balanced and there's not much HGA, if any. At venues where one teams plays a lot and the other team rarely even visits the state, HGA is high.

The main difference in practice is better modelling of Geelong in Melbourne, along with uncommon situations like Melbourne playing in the Northern Territory. For example, for the Round 1 Melbourne vs Geelong game at the MCG, under the old model Squiggle would give Melbourne 12 points of HGA, but under the new model, it's only 2.5 points, due to the Cats' fairly good familiarity with that ground.

Actually here are the current HGA values for all R1 matches:

• Richmond +1.1 vs Carlton @ MCG
• Essendon +7.9 vs Adelaide @ Docklands
• St. Kilda +7.6 vs Brisbane @ Docklands
• Port Adelaide +10.0 vs Fremantle @ Adelaide Oval
• Gold Coast +10.3 vs North Melbourne @ Cazalys
• Hawthorn -1.0 vs Collingwood @ MCG
• GWS +9.0 vs Bulldogs @ UNSW Canberra
• Melbourne +2.5 vs Geelong @ MCG
• West Coast +10.9 vs Sydney @ Perth Stadium

Hawthorn are the only home team to have negative HGA. West Coast still get a bunch of HGA despite playing on a brand new ground due to the state-based numbers.

If you applied these retrospectively on the squiggle, how many more tips would you have predicted in the regular season?

 

[Final Siren]

If you applied these retrospectively on the squiggle, how many more tips would you have predicted in the regular season?

I may tweak a few more things but on current numbers:

Year|Squiggle v1|Squiggle 2.0|Diff\2017| 127 (61.4%)| 134 (64.7%)| +7\2016| 143 (69.1%)| 155 (74.9%)|+12\2015| 146 (70.9%)| 152 (73.8%)|+6\2014| 149 (72.0%)| 149 (72.0%)|-\2013| 150 (72.5%)| 151 (73.0%)|+1\2012| 161 (77.8%)| 152 (73.4%)|-9\2011| 152 (77.6%)| 151 (77.0%)|-1

This kind of thing is always slightly dodgy, though, because the Squiggle 2.0 numbers are "retro-dictions" -- they've been generated after the fact. It's easier to build a model that fits past data than to predict the future.

Also there's a fair bit of luck involved, especially when you measure by Tips instead of something like Mean Average Error, so a bad model can beat a good one in any given year.

 

[Andre]

The main difference in practice is better modelling of Geelong in Melbourne, along with uncommon situations like Melbourne playing in the Northern Territory. For example, for the Round 1 Melbourne vs Geelong game at the MCG, under the old model Squiggle would give Melbourne 12 points of HGA, but under the new model, it's only 2.5 points, due to the Cats' fairly good familiarity with that ground.

For HGA is this now modelling Shanghai as a 0 point HGA, since both Port and GC will have played there just the once and it's not home state for either?

 

[Final Siren]

For HGA is this now modelling Shanghai as a 0 point HGA, since both Port and GC will have played there just the once and it's not home state for either?

Yep, exactly.

 

[Papa G]

Does a large Vic team, who's home ground is the MCG, lets say Richmond for example, playing a minnow Vic team who's home ground isn't the MCG, lets say North for example count for any type of home ground advantage? Lets say the Tigers supporters outnumber the North Supporters 4 to 1 and play at the ground 14 times a year and North play there 3 times. Surely there has to be a home ground advantage of sorts and should have a weighting?

 

[citizen-erased]

I would predict Sydney to win a QF even in Adelaide.
Forcing a showdown prelim of epic proportions followed by a GF replay from 2017, or perhaps a different SA challenger

Richmond unlikely to struggle against gws in Melbourne and a swans v tiges prelim would be rather awesome too

Even better - we beat gws and face Richmond in a semi. We’ll lose but I’ll take a finals win and the atmosphere in Melbourne leading up to it

 

[threenewpadlocks]

Does a large Vic team, who's home ground is the MCG, lets say Richmond for example, playing a minnow Vic team who's home ground isn't the MCG, lets say North for example count for any type of home ground advantage? Lets say the Tigers supporters outnumber the North Supporters 4 to 1 and play at the ground 14 times a year and North play there 3 times. Surely there has to be a home ground advantage of sorts and should have a weighting?

The problem with that line of thinking is two things: it's incredibly reductionist, so data gets messy and noisy, but also you have to consider this:
Nobody can really explain how the various factors for home ground advantage exist.

Is it travel and being "away from home", sleeping in hotels and so on? Is it familiarity with the ground itself? Is it familiarity with the weather and altitude of the ground? There's been scientific studies which have shown greater testosterone in the saliva of ice hockey players playing at home - is it the primal nature of "defending our territory" that improves testosterone and therefore performance? Crowds. Do crowds impact performance, or just quality of umpiring/refereeing as some studies have shown? If you factor in crowds to a model, do you go with total crowds, or density of crowds? Then you've got footy, a sport where the fields vary in size, and how that relates to tactics: I'd say teams that rely on good defensive structure behind the ball do better on narrow grounds as there's less opportunity to slip through the ground (Geelong's success at Kardinia vs the MCG last season, for example), but even if we inherently understand that, how to you put it into a prediction model?

How do all these factors balance against each other in what we understand to be HGA? And even if we understood that, how do we put it into a model?

Past results at venues specifically, and experience at a venue, are easily measured and applied, and probably create more accuracy than a simple uniform HGA. But beyond that it starts getting messy, and in the case of the Squiggle which was created as a kind of "less is more" prediction model, is kind of against its principle.

 

[Papa G]

The problem with that line of thinking is two things: it's incredibly reductionist, so data gets messy and noisy, but also you have to consider this:
Nobody can really explain how the various factors for home ground advantage exist.

Is it travel and being "away from home", sleeping in hotels and so on? Is it familiarity with the ground itself? Is it familiarity with the weather and altitude of the ground? There's been scientific studies which have shown greater testosterone in the saliva of ice hockey players playing at home - is it the primal nature of "defending our territory" that improves testosterone and therefore performance? Crowds. Do crowds impact performance, or just quality of umpiring/refereeing as some studies have shown? If you factor in crowds to a model, do you go with total crowds, or density of crowds? Then you've got footy, a sport where the fields vary in size, and how that relates to tactics: I'd say teams that rely on good defensive structure behind the ball do better on narrow grounds as there's less opportunity to slip through the ground (Geelong's success at Kardinia vs the MCG last season, for example), but even if we inherently understand that, how to you put it into a prediction model?

How do all these factors balance against each other in what we understand to be HGA? And even if we understood that, how do we put it into a model?

Past results at venues specifically, and experience at a venue, are easily measured and applied, and probably create more accuracy than a simple uniform HGA. But beyond that it starts getting messy, and in the case of the Squiggle which was created as a kind of "less is more" prediction model, is kind of against its principle.

Yes it is difficult, but was it considered an advantage back in the day when the Dogs played another VFL team at the Western Oval? Carlton at Princes Park? Essendon at Windy Hill? Yes it was. Is my example not a similar principle?

 

[threenewpadlocks]

Yes it is difficult, but was it considered an advantage back in the day when the Dogs played another VFL team at the Western Oval? Carlton at Princes Park? Essendon at Windy Hill? Yes it was. Is my example not a similar principle?

Yes, but that's factored into the "experience at ground" statistical part of the new formula.
Once you get into things like weighing proportion of crowd support against the opposition according to some formula, but that's different for each stadium because the density and acoustics of each stadium is different which is another formula, or the while where as we don't know whether or not crowds influence umpires to be biased which leads to better home performance, or just makes the home team work harder, or a combination of both, and factoring in that, it gets really really really messy.

Even an NBA study has shown that in clutch situations, whilst a home crowd helps effort activities like offensive rebounding, players actually shoot worse at free throws. Does that apply to the AFL?

 

[Gralin]

Got some more updates ahead of season 2018:

(1) Pre-season tracking: We had something resembling a normal preseason this year, so Squiggle is using those results in its calculations for the season ahead.

(2) Round-based Sensitivity: Squiggle is now more sensitive in the early rounds and in finals, to better capture the important signalling of surprising results at these times. In particular, an unexpected Round 1 result can cause a team to move a very long way.

(3) Goal-kicking Accuracy: As flagged last year, since goalkicking accuracy seems to be non-reproducible -- teams that are highly accurate one week aren't more likely to be highly accurate the next -- Squiggle will interpret a scoreline like 13.20 as a team that was unlucky to not win by more, and thus will rate them more highly, while a scoreline like 12.3 will be interpreted as a team that got closer than they probably deserved.

(4) Home Ground Advantage: Gone is the practice of assigning 12 points of home advantage to interstate hosts and ignoring everything else. Squiggle now calculates HGA based on relative ground & state familiarity: It scores each team on how many times they've played at the same venue and (less significantly) in the same state over recent years, and assigns HGA points based on how lopsided the numbers are. So at venues where both teams have played fairly often, ground familiarity is balanced and there's not much HGA, if any. At venues where one teams plays a lot and the other team rarely even visits the state, HGA is high.

The main difference in practice is better modelling of Geelong in Melbourne, along with uncommon situations like Melbourne playing in the Northern Territory. For example, for the Round 1 Melbourne vs Geelong game at the MCG, under the old model Squiggle would give Melbourne 12 points of HGA, but under the new model, it's only 2.5 points, due to the Cats' fairly good familiarity with that ground.

Actually here are the current HGA values for all R1 matches:

• Richmond +1.1 vs Carlton @ MCG
• Essendon +7.9 vs Adelaide @ Docklands
• St. Kilda +7.6 vs Brisbane @ Docklands
• Port Adelaide +10.0 vs Fremantle @ Adelaide Oval
• Gold Coast +10.3 vs North Melbourne @ Cazalys
• Hawthorn -1.0 vs Collingwood @ MCG
• GWS +9.0 vs Bulldogs @ UNSW Canberra
• Melbourne +2.5 vs Geelong @ MCG
• West Coast +10.9 vs Sydney @ Perth Stadium

Hawthorn are the only home team to have negative HGA. West Coast still get a bunch of HGA despite playing on a brand new ground due to the state-based numbers.

Is Hawks getting a -1 HGA against the Pies based on records last year at the G or ?
Just curious how that works

 

[Vintage]

Man crush alert!!!!

 

[HTT]

Is Hawks getting a -1 HGA against the Pies based on records last year at the G or ?
Just curious how that works

From what I understand it’s based purely on how many games you’ve played there in recent years. Hawthorn are obviously a tenant and would feel absolutely at home on the G, but they sell 3 or 4 games to Tasmania every year, while the pies play everything they can on the G, so they’re technically more familiar with the ground.

 

[Gralin]

From what I understand it’s based purely on how many games you’ve played there in recent years. Hawthorn are obviously a tenant and would feel absolutely at home on the G, but they sell 3 or 4 games to Tasmania every year, while the pies play everything they can on the G, so they’re technically more familiar with the ground.

Makes sense

 

[Final Siren]

From what I understand it’s based purely on how many games you’ve played there in recent years. Hawthorn are obviously a tenant and would feel absolutely at home on the G, but they sell 3 or 4 games to Tasmania every year, while the pies play everything they can on the G, so they’re technically more familiar with the ground.

Yep, that's right.

|Hawks @ MCG|Hawks in Victoria|Pies @ MCG|Pies in Victoria\2018 pre-season|0|1|0|1\2017|11|13|14|17\2016|12|15|14|17\2015|11|15|14|17

After weighting the numbers, the model calculates 60.0 units of familiarity for Hawthorn vs 72.6 units for Collingwood. Put another way, the Hawks have 45% of the two teams' combined familiarity, or -5% compared to a neutral 50/50 split. Under this model, that's -1.0 points.

I've tried the other way before -- calculating whether a particular team tends to play a particular ground better or worse than expected -- but there's never enough data to form reliable conclusions.

This method (familiarity) applies a single algorithm to all teams at all venues, which means I can test it against 200 games per year and be more confident that it's tracking something real.

 

[Lions_Prid3]

Rd 1 tips:


On the back of the above, I think Fremantle has the most scope for a decent squiggle this week. As always though, i'm sure there will be some surprises.

 

[WhiskeyTangoFoxtrot]

Looking at that, I'd put money on Essendon, Fremantle, Melbourne, Brisbane and Sydney all out performing their expected Squiggle.