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

[Final Siren]

What would be the cats rating under the old squiggle where it just counted total points and didn't care about accuracy?

It would look like this:



... while the new version looks like this:



So yes, the Cats are being heavily marked down for their unusually high goalkicking accuracy, which Squiggle assumes is temporary. (So too are the Eagles.) In particular, Squiggle v4 is far less impressed by games like these:

• R7: GEE 13.8 def ESS 7.12 - a 32pt victory, but only 21 shots to 19
• R8: NTH 11.14 def by GEE 16.8 - a 24pt victory, but with fewer scoring shots
• R9: GEE 21.7 def BUL 13.11 - a 44pt thumping but only 29 shots to 24

 

[Duskfire]

While I understand the point of the goal accuracy being seen as an anomaly from a pure statistics point of view, surely going for a half a season casts doubt in the anomaly aspect?

Not to mention, I don’t think Geelong take many crazy shots. A lot of the time they mark it directly in front, or the forward pressure around the ball allows a free player to gather and steady. Ablett and Miers have often been able to take their time getting a crumb.

It also helps they have a couple of really good kickers; Ablett Hawkins and Miers are all great kicks. A lot of Rohan goals come from directly in front etc etc.

 

[Final Siren]

While I understand the point of the goal accuracy being seen as an anomaly from a pure statistics point of view, surely going for a half a season casts doubt in the anomaly aspect?

Not to mention, I don’t think Geelong take many crazy shots. A lot of the time they mark it directly in front, or the forward pressure around the ball allows a free player to gather and steady. Ablett and Miers have often been able to take their time getting a crumb.

It also helps they have a couple of really good kickers; Ablett Hawkins and Miers are all great kicks. A lot of Rohan goals come from directly in front etc etc.

That would be fine with me if it would place Geelong in amongst the top pack of teams this century in terms of goalkicking accuracy. But what it would actually mean is that Geelong 2019 are the second-most accurate team in VFL/AFL history (behind St Kilda 2004).

It's not impossible, but it's up against an enormous weight of history, which contains an awful lot of very good teams with very good goalkickers who still couldn't achieve a 60% conversion rate over a season. There must be many lucky seasons in there as well: still no-one with Geelong's current numbers, except St Kilda.

To be fair, accuracy has gradually improved since 1897, so it's not really worth considering data before this century. And Sydney 1998 got close (59.6%). But even so, that's around 350 teams, so I need a lot of convincing that Geelong have found a way to return a more accurate season than all but one of them.

 

[catstorm]

I have a question, if the algorithm has changed has that also caused a change in where the previous premiership cups are being placed?

Because it wouldn't give an accurate representation of a teams position compared to the past if they didn't move as well.

 

[numbwombat]

That would be fine with me if it would place Geelong in amongst the top pack of teams this century in terms of goalkicking accuracy. But what it would actually mean is that Geelong 2019 are the second-most accurate team in VFL/AFL history (behind St Kilda 2004).

It's not impossible, but it's up against an enormous weight of history, which contains an awful lot of very good teams with very good goalkickers who still couldn't achieve a 60% conversion rate over a season. There must be many lucky seasons in there as well: still no-one with Geelong's current numbers, except St Kilda.

To be fair, accuracy has gradually improved since 1897, so it's not really worth considering data before this century. And Sydney 1998 got close (59.6%). But even so, that's around 350 teams, so I need a lot of convincing that Geelong have found a way to return a more accurate season than all but one of them.

That makes sense, but if Geelong kick for goal slightly less accurately over the second half of the season and they drop to 58%, they would still be much better than the league average, and would no longer be an outlier and not even be in the top 2 most accurate teams of the past 5 seasons, behind Hawthorn. Sure it's unlikely that they'll stay at an all-time record 60%, but there are plenty of examples of good teams kicking well above 51% for extended periods - we had a few years over 55% in our good run, too.

Geelong's opponents have kicked for goal at 45% this year, after a season average of 45.5% in 2018, so even if the squiggle doesn't trust their current accuracy, a comparable number of shots (e.g. round 8) should indicate a Geelong win.

 

[cryptor]

I have a question, if the algorithm has changed has that also caused a change in where the previous premiership cups are being placed?

Because it wouldn't give an accurate representation of a teams position compared to the past if they didn't move as well.

They've indeed moved. Compare the placements in his two pics above.

 

[laam]

Similar question. But given the premiers all won at least 3 games against good opposition to win the flag and the location shows the final spot, ther is an inbuilt increase in position for the premier. Would it be more reasonable to show the premiers position at round 22 that would show where you need to be prior to finals? As typical all premiership teams shoot into the zone after their 3 wins.

On SM-G960F using BigFooty.com mobile app

 

[Mad_Hatter]

That would be fine with me if it would place Geelong in amongst the top pack of teams this century in terms of goalkicking accuracy. But what it would actually mean is that Geelong 2019 are the second-most accurate team in VFL/AFL history (behind St Kilda 2004).

It's not impossible, but it's up against an enormous weight of history, which contains an awful lot of very good teams with very good goalkickers who still couldn't achieve a 60% conversion rate over a season. There must be many lucky seasons in there as well: still no-one with Geelong's current numbers, except St Kilda.

To be fair, accuracy has gradually improved since 1897, so it's not really worth considering data before this century. And Sydney 1998 got close (59.6%). But even so, that's around 350 teams, so I need a lot of convincing that Geelong have found a way to return a more accurate season than all but one of them.

Just because it's the norm doesn't mean it HAS to happen.
Outliers are outliers for a reason, perhaps we are seeing one this year.

 

[the harry]

Just because it's the norm doesn't mean it HAS to happen.
Outliers are outliers for a reason, perhaps we are seeing one this year.

But you would be silly to expect something other than the norm to happen.

 

[Final Siren]

That makes sense, but if Geelong kick for goal slightly less accurately over the second half of the season and they drop to 58%, they would still be much better than the league average, and would no longer be an outlier and not even be in the top 2 most accurate teams of the past 5 seasons, behind Hawthorn. Sure it's unlikely that they'll stay at an all-time record 60%, but there are plenty of examples of good teams kicking well above 51% for extended periods - we had a few years over 55% in our good run, too.

Geelong's opponents have kicked for goal at 45% this year, after a season average of 45.5% in 2018, so even if the squiggle doesn't trust their current accuracy, a comparable number of shots (e.g. round 8) should indicate a Geelong win.

Yep, this is all correct, and you are accurately describing how "reversion to the mean" works. The principle doesn't mean we should expect Geelong to start converting at a below-average rate in order to get back to the average - that would be a form of the Gambler's Fallacy, the false belief that something is "due" because it hasn't happened for a while. What it means is we should expect Geelong to convert at about an average rate from this point forward - which, since they're starting from such a high base, will have the effect of bringing their total rate down.

Also, as you say, it's fairly common for good teams to finish with above-average conversion, like around 55%.

I do believe Geelong will finish the season with an above-average conversion rate - probably well above average - since (a) they are a very good team, and (b) already off to a flier. I don't believe they will maintain their current conversion rate (60.2%).

 

[Final Siren]

But you would be silly to expect something other than the norm to happen.

Yes! Humans look to outliers far too often. We can always think up a reason why this time is special, or that extraordinary once-in-a-decade event might be about to be repeated. We forget the dozens or hundreds of similar times it didn't happen.

 

[Hobbes]

Richmond v Geelong +21
Carlton v Brisbane +22
Gold Coast v North Melbourne +18
Adelaide v GWS +4
Sydney v West Coast +3
Collingwood +25 v Melbourne

Three upsets, so my 3/6 will be pretty normal. 66/105.

1. Geelong 32.9
2. GWS 16.7
3. Collingwood 16.2
4. Adelaide 8.7 (+2)
5. North Melbourne 8.3 (+4)
6. Port Adelaide 7.7 (-1)
7. West Coast 6.7 (-3)
8. Sydney 5.8 (+5)
9. Essendon 4.9 (+1)
10. Fremantle 3.7 (-3)
11. Richmond 0.7 (-3)
12. Hawthorn -0.1
13. Brisbane -1.3 (-2)
14. Western Bulldogs -8.2
15. Melbourne -13.5
16. St Kilda -16.6
17. Carlton -19.6 (+1)
18. Gold Coast -22.8 (-1)

A bit of a positive skew here, for reasons which aren't completely obvious to me. It doesn't affect the functionality of the algorithm, but you could subtract roughly 1.7 from every rating to level it out.

The keys here:
(1) Geelong are streaking ahead of the field. At this point, there's no match in which the algorithm wouldn't tip them.
(2) Collingwood and GWS are next.
(3) There's a big cluster of teams underneath them.
(4) North, Sydney and Fremantle are in better form than is suggested by the ladder.
(5) Brisbane and Richmond are slipping.
(6) The teams at the bottom are staying at the bottom.

Adelaide +14 v Richmond
Essendon +5 v Hawthorn
Gold Coast v St Kilda +1 (Virtually tipping a tie)
Fremantle +9 v Port Adelaide
Carlton v Western Bulldogs +11
North Melbourne v GWS +3

 

[numbwombat]

I do believe Geelong will finish the season with an above-average conversion rate - probably well above average - since (a) they are a very good team, and (b) already off to a flier. I don't believe they will maintain their current conversion rate (60.2%).

Fair enough! Maybe I misunderstood - I don't mind how the squiggle rates Geelong, we either make the granny or we don't (& I don't think we're as good as a lot of teams of the past ten years) - but I got the impression you were saying the squiggle was rating us particularly poorly because it was expecting scoring shots from both teams to be close to 52%. Would it be viable to have the model accept any accuracy between say 49-55% as sustainable, and anything outside of that as potentially sustainable depending on the accuracy a team and their opponents have been kicking at over the last 1-2 seasons?

What it means is we should expect Geelong to convert at about an average rate from this point forward - which, since they're starting from such a high base, will have the effect of bringing their total rate down.

This part doesn't necessarily seem like a reasonable expectation, given that some teams have converted well above the average rate over multiple seasons. We've only been kicking well for 10 games, so might be down to luck and convert near 52% going forward, but would the squiggle have looked at Hawthorn mid-2015 and say their conversion was unlikely to continue to be so good, even though they'd been doing it for 2 1/2 seasons already? That seems like an easily fixed flaw in the model, if so.

 

[caloschwaby]

Despite Pies beating Dees by 30 more points ,keeping them 20 points less and scoring 10 points more than Squiggle predicted, it seems they haven't moved at all on the squiggle?

 

[HTT]

Despite Pies beating Dees by 30 more points ,keeping them 20 points less and scoring 10 points more than Squiggle predicted, it seems they haven't moved at all on the squiggle?

Only 1 more scoring shot so the squiggle's view is that the scoreboard flattered you a bit.
Adjusted scores are 86.1-70.6 so almost as predicted for the squiggle.

 

[caloschwaby]

Only 1 more scoring shot so the squiggle's view is that the scoreboard flattered you a bit.
Adjusted scores are 86.1-70.6 so almost as predicted for the squiggle.

Oh didn't realise that was taken into account as heavily. Fair enough

 

[Hobbes]

Oh didn't realise that was taken into account as heavily. Fair enough

I think the Squiggle works on 4 points for a goal, one for a point to reduce goalkicking luck.

 

[Final Siren]

I think the Squiggle works on 4 points for a goal, one for a point to reduce goalkicking luck.

It's about 4.5 per goal and 2.5 per behind.

 

[Final Siren]

Fair enough! Maybe I misunderstood - I don't mind how the squiggle rates Geelong, we either make the granny or we don't (& I don't think we're as good as a lot of teams of the past ten years) - but I got the impression you were saying the squiggle was rating us particularly poorly because it was expecting scoring shots from both teams to be close to 52%. Would it be viable to have the model accept any accuracy between say 49-55% as sustainable, and anything outside of that as potentially sustainable depending on the accuracy a team and their opponents have been kicking at over the last 1-2 seasons?


This part doesn't necessarily seem like a reasonable expectation, given that some teams have converted well above the average rate over multiple seasons. We've only been kicking well for 10 games, so might be down to luck and convert near 52% going forward, but would the squiggle have looked at Hawthorn mid-2015 and say their conversion was unlikely to continue to be so good, even though they'd been doing it for 2 1/2 seasons already? That seems like an easily fixed flaw in the model, if so.

Mmm, really good thoughts, and worth investigating further, but I'm a little skeptical. Sure, some teams have recorded high accuracy rates over a few seasons, but plenty of others have fluctuated from season to season (and indeed game to game). So the first question is whether there really is a genuine correlation between past & future conversion accuracy.

I suspect there is, but only because both correlate with something else even more strongly: team performance. Good teams regularly record above-average conversion rates, while bad teams record poor ones. So I reckon it's easier to guess a team's conversion rate if you know whether they're having a good year or not (e.g. Collingwood 2018: top 4, 55%) rather than knowing their past conversion rate (e.g. Collingwood 2017: bottom 6, 50%).

Squiggle already rates team performance, and values a team's ability to kick more goals than behinds, so it's not missing anything unless there is an extra independent effect whereby teams can have unusually good or poor conversion over an extended period of time regardless of how well they're playing. That's something I would like to look into, but my initial numbers didn't throw up anything.

By the way, "we should expect Geelong to convert at about an average rate" isn't supposed to mean the strict average of 52.86% - it's likely to be more like the average of a good team, which is about 55%. That would drag the season-long number down to somewhere around 57-58%.

Here are the current top 6 all-time best conversion rates after Round 12, showing where they wound up:

1. HAW 2014 61.8% -> 58.9% (-2.9)
2. SYD 2003 61.1% -> 59.0% (-2.1)
3. MEL 2016 60.9% -> 55.6% (-5.3)
4. BUL 2000 60.8% -> 59.2% (-1.6)
5. GEE 1996 60.3% -> 56.8% (-3.5)
6. GEE 2019 60.3% -> ?%

So in none of those cases did the team manage to maintain their outstanding conversion rates. Instead, they converted at a roughly average rate in the second half of the season, which brought them down to very-good-but-not-amazing season numbers.

 

[Billy ray]

Suns continuing our juvenile behaviour with the penis drawing

 

[HTT]

Squiggle sliding doors WTF moment #1748 - this one from 2017 (good year, that one)
(Alternate reality #1,632,958,846, luck set to 5%, for reference

 

[Hobbes]

I'm having trouble doing sliding doors for this season. Am I doing something wrong? Every time I try it, some teams have fewer losses for the year than they have already lost.

 

[HTT]

I'm having trouble doing sliding doors for this season. Am I doing something wrong? Every time I try it, some teams have fewer losses for the year than they have already lost.

Sliding doors seems to still be set up for the 2018 season. Final Siren pls fix I’m sick of seeing alternate realities where we get murdered by Collingwood at some point in finals.

 

[shintemaster]

Similar question. But given the premiers all won at least 3 games against good opposition to win the flag and the location shows the final spot, ther is an inbuilt increase in position for the premier. Would it be more reasonable to show the premiers position at round 22 that would show where you need to be prior to finals? As typical all premiership teams shoot into the zone after their 3 wins.

On SM-G960F using BigFooty.com mobile app

Great comment. Fascinated to see the answer to this.

 

[Final Siren]

Sliding doors seems to still be set up for the 2018 season. Final Siren pls fix I’m sick of seeing alternate realities where we get murdered by Collingwood at some point in finals.

Fixed! Thanks for letting me know.