Certified Legendary Thread Squiggle 2017

[Vintage]

I'm very sceptical about this increased veneration for scoring shots in new versions.

Sure, a stack of behinds can be indicative of poor luck when kicking for goal, but it can also be a consequence of

(1) Rushed behinds
(2) Poor penetration in the forward line. Taking shots from bad angles or distances because they can't get into better positions
(3) Defensive pressure

If you're going to go down this path, you really want to make calculations based on the shots which teams manage to set up, and what their chances were. This will include the easy shots which were sprayed OOB, and exclude the non-chances which dribbled through for a behind.

Bottom line - converting chances to goals is good - it's what teams are trying to achieve.

Surely FS has earned our trust?

If he doesn't have it then I'm not sure what it takes!?

 

[Hobbes]

Surely FS has earned our trust?

If he doesn't have it then I'm not sure what it takes!?

I trust FS probably more than most - he even runs simulations on my algorithm. And he has a loyalty to truth - the squiggle was fine-tuned to maximise predictive accuracy (subject to the constraints of the data used - scores and home stage). I assume that changes won't be instituted unless he's tested them thoroughly. But I do worry that in rewarding the teams which had unlucky runs of shooting for goal, we also reward the teams who set themselves up for failure with wild stabs from 48m on an angle. And, when it comes down to it, converting chances to goals is what the game is about.

 

[SJ]

I trust FS probably more than most - he even runs simulations on my algorithm. And he has a loyalty to truth - the squiggle was fine-tuned to maximise predictive accuracy (subject to the constraints of the data used - scores and home stage). I assume that changes won't be instituted unless he's tested them thoroughly. But I do worry that in rewarding the teams which had unlucky runs of shooting for goal, we also reward the teams who set themselves up for failure with wild stabs from 48m on an angle. And, when it comes down to it, converting chances to goals is what the game is about.

But Final Siren doesn't decide who gets rewarded. If the algorithm beats the others, then it works. It doesn't matter if it's not perfect.

 

[hutchy31]

Scoring shots is a complicated one. First of all, there's a difference between scoring shots (the total amount of goals and behinds scored) and shots on goal, which is the total amount of times a team shoots towards goal hoping to score (and remember sometimes if a team can't find a target up forward they'll kick it to the boundary hoping to win a clearance and kick a goal).

Secondly, if we use Richmond as an example because it's the side I'm most familiar with, Richmond's poor kicking is partly due to shallow entries, because it has a small forwardline which needs space, and space is not often offered close to goal, so Richmond try and hit up shallower entries which are less likely to score. To get entries close to goal, you often need a strong marking genuine tall, like Daniher. Richmond only has one of those, and even then, Riewoldt hasn't been clunking goalsquare marks for a while now. You'll often see players like Edwards, Caddy and Castagna taking fairly low percentage shots on goal, but the mindset of the team seems to be that taking a shot on goal would be better than trying to find another target up forward which could result in a turnover. Better to have a shot, if you score a goal, great, if not, reset the forward press and hope for another entry (they are also three hungry players just quietly).

So is Richmond's poor accuracy pure bad luck, or a fault of the side? I'd say it's the latter. Richmond get a lot of scores per inside 50 entry, which shows that the midfield and backline is working okay. But in terms of the forwardline, it's not. So it's hard to evaluate whether that makes Richmond a good side or a bad side.

I'd compare kicking accuracy in football to reception percentages in the NFL. Teams with a poor receiving group will drop catches whereas teams with a good receiving group wont. Whereas including scoring shots in a model would imply that scoring shots is like field goal accuracy in the NFL, which does fluctuate randomly, with kickers going through weird hot and cold spots. I think a team which generates a lot of shots on goal but not a lot of accuracy may be a side which has a poor forwardline. To prove this (and we'll just use scoring shots), who are the four worst sides for accuracy this year? In order Footscray, St Kilda, Collingwood and Footscray. St Kilda's forwardline is fine, that's bizarre. But the other three's forwardlines are buttons. The best for accuracy Geelong, Melbourne, Adelaide, Essendon and Brisbane, who all have fairly good forwardlines. In fact the difference between a side like say Collingwood and Essendon this year has often just been about who can kick straight this season.

I was going to protest that you had the dogs twice in the bottom four, but having watched them all season they've probably kicked badly enough to be in there twice just for emphasis.

 

[Pessimistic]

Not sure where to post this for discussion, but does anyone think a good number of teams have looked at the bye between H&A and finals (bulldogs winning the finals) and modified their fitness schedule accordingly, which has contributed to all the strange results in H&A this year?

 

[Dr Tigris]

Not sure where to post this for discussion, but does anyone think a good number of teams have looked at the bye between H&A and finals (bulldogs winning the finals) and modified their fitness schedule accordingly, which has contributed to all the strange results in H&A this year?

Not really. It'd impact on teams in the run home. That's starting about now. In a usual historic season teams in the top 4 might plan to 'freshen' up players for the finals. Having a rest stuffs that up. You get that rest to freshen up anyway. What I would expect the difference to be is that previously teams securely in the top 4 might rest someone, alter training, to be full on come the finals. much less need to do that now. So some of the strange results in the last 2 rounds won't happen (or much less likely). Up til now it's all about winning as much as possible.

this year is just weird.

 

[Gralin]

Not sure where to post this for discussion, but does anyone think a good number of teams have looked at the bye between H&A and finals (bulldogs winning the finals) and modified their fitness schedule accordingly, which has contributed to all the strange results in H&A this year?

i think as it stands with the current rules and spread of talent along with game style changes that teams are just a lot closer to each other and a bad day or injuries make that even closer

everyone is still adapting to the new world and as it stands we are very close to that American saying on any given sunday

Gill would be happy as a pig in shit over it he is on record as stating he'd love every team to go 11-11

 

[Final Siren]

I'm very sceptical about this increased veneration for scoring shots in new versions.

Sure, a stack of behinds can be indicative of poor luck when kicking for goal, but it can also be a consequence of

(1) Rushed behinds
(2) Poor penetration in the forward line. Taking shots from bad angles or distances because they can't get into better positions
(3) Defensive pressure

It can, yep. The question is which is the more common cause of inaccuracy: genuine team weakness, like being forced to take shots from the boundary line because your midfield delivery sucks, or transient factors that won't apply the next week, like luck or temporary form. If it's more often the latter, you should look at scoring shots, even if that's not happening all the time.

For years I've noted that my scoring shot-based algorithms perform pretty well. These don't actually rely soley on scoring shots; they weight goals and behinds differently. So they still value goals more highly than behinds, just not 6 times more highly.

The best algorithms over the long term value goals at about 4 points each and behinds at 3 points. For example, on Saturday GWS 13.20 98 defeated Fremantle 13.8 86, which was a 12 point victory. A simple SHOTS-4:3 algorithm would score GWS at 112 points and Fremantle at 76 points, considering that a 36 point win. (SHOTS-4:3:ISTATE-91:12 would also adjust for interstate advantage, deeming it to be a 48 point win.) A significant adjustment, but it's not huge, even in a game where the Giants had 57% more scoring shots (33 to 21).

I'm now about convinced that more often than not, scoring shots are a better indicator of underlying team strength than final scores. This is especially the case for squiggle in matches where there's an unusually high or low number of scoring shots, because it's sensitive to extremes like that.

It doesn't always work out better, but the new algorithm has beaten ISTATE-91:12 in tips in 65% of years since 1975 and 72% of years in mean average error. Since 2000, it's even better, at 72% and 83%. This year it's 4 tips ahead with a much better MAE.

I do agree it's a little dangerous to shift away from actual scores, because you become vulnerable to gameplan variations, like Hawthorn deliberately rushing a million behinds against Geelong in the 2008 GF. But at this point, the evidence is too compelling for me not to use it.

If you're going to go down this path, you really want to make calculations based on the shots which teams manage to set up, and what their chances were. This will include the easy shots which were sprayed OOB, and exclude the non-chances which dribbled through for a behind.

Yes, ideally you would calculate the exact likelihood of a shot becoming a goal based on a range of factors like that, if the data was available. But it doesn't matter whether it's perfect, only whether it's better.

 

[Final Siren]

Final Siren there are some people on Twitter doing some good work with scoring shot, FiguringFooty one of them - would strongly recommend looking at some of his work first before going to a blunt scoring shot model, as not all defences/forward 50 entries are similar

Yep, he's one of the gang at squiggle.com.au! Robert has access to non-public data and produces some really great score chance visualizations, as well as xScores, which is his algorithm's output after crunching the numbers on things like distance, position, and the pressure the kicker was under at the time.

This is a good approach and one I'd probably take if I had access to the same data. However, it's worth noting that there's still a lot missing; for example, there's no data on shots that went out of bounds or failed to make the distance. And "pressure" is a pretty crude categorization, as I understand it. Plus, of course, there's never going to be any data on things like whether there were three other blokes running forward free at the time.

Importantly, the data that is available only goes back a handful of years, so can't be tested over anything like the long term. That makes it very hard to prove that anything is a statistically significant improvement.

I did take a stab at a more refined model using shot position data alone. I found it interesting to learn how similar shot likelihood percentages are at many different spots on the ground -- because footballers basically move the ball around until they have a 60% or better chance of kicking a goal, then they take a shot! I had expected to see something much more linear, with conversion percentages going steadily up as the goals got closer, but it's not like that.

Which all means that approximations are still pretty good.

 

[Simon_Nesbit]

Hawthorn's data analytics from 2010(ish) showed that actual goal accuracy was one of the least reliable indicators of winning/premierships; whilst net shots on goal (shots taken less shots against) was one of the strongest indicators.

It was one of the reasons Hawthorn spend so little time on goal-kicking (and why we've sucked at it for a decade). It's simply not a priority - our good goal-kickers have all come to the club with great technique/mental aptitude for the skill.

 

[threenewpadlocks]

Hawthorn's data analytics from 2010(ish) showed that actual goal accuracy was one of the least reliable indicators of winning/premierships; whilst net shots on goal (shots taken less shots against) was one of the strongest indicators.

It was one of the reasons Hawthorn spend so little time on goal-kicking (and why we've sucked at it for a decade). It's simply not a priority - our good goal-kickers have all come to the club with great technique/mental aptitude for the skill.

Hawthorn are seemingly switched on with this sort of stuff - obviously we don't know (and we would hope) how the 17 other clubs go behind closed doors with this sort of thing, but at least we know the Hawks do it:



I'm dubious to think that long-term you should use 100% "expected score" given league averages in any given shot, for example, it's logically not correct to treat Buddy the same as a league-average player when he's kicking from 50m on an angle, or Levi Casboult when he's 25m directly in front. But as explained in the video, it seems to be a good tool for coaches in-game ("the sum of the efforts") for example, some coaches might make in-game changes if they're behind on the scoreboard when they shouldn't, because they've been the better team whilst being "unlucky" in front of goal.

 

[Zuta]

Adelaide have made a massive move to put themselves amongst the premiership cups. Tipped to thump Geelong in the grand final. Theirs to lose from here!

 

[The Dice Man]

Squiggle has Geelong defeating Tigers in a PF. Squiggle has been wrong about the Tigers all year, long may this continue.

 

[thephenom]

Adelaide have made a massive move to put themselves amongst the premiership cups. Tipped to thump Geelong in the grand final. Theirs to lose from here!

If only the squiggle mattered in real life hey

 

[WhiskeyTangoFoxtrot]

Fly my pretty, fly!

 

[Del165198]

Port beating Richmond in the squiggle.

Please.

Embarrassing.

 

[Jonts]

I was thinking about the squiggle way too much during the showdown

 

[Crom]

Fly my pretty, fly!

It's, just, so, beautiful..

 

[chazwazza]

Squiggle is a cruel mistress..... YEAH BABY!!!!
Oh do behave.
Cue the music.....

 

[Gydafud]

So Swans are second favourite for the flag according to the flagpole, but don't make the GF based on the tipping prediction.

Who wins between Crows and Swans at the MCG on their current stats?

 

[WhiskeyTangoFoxtrot]

So Swans are second favourite for the flag according to the flagpole, but don't make the GF based on the tipping prediction.

Who wins between Crows and Swans at the MCG on their current stats?

The website recommends using the flagpole to determine the likely GF winner, rather than the ladder predictor.

 

[Gydafud]

The website recommends using the flagpole to determine the likely GF winner, rather than the ladder predictor.

Yeah, but I just want to see those scores

 

[Roobs321]

Twas an intriguing weekend. Crows have back-to-back convincing floggings of pretenders. Swans continue their ominous form. GWS show signs of turning the corner back into formidable contention. Richmond in the mix. Geelong still have that Victoria-only run into finals which should see them regather in good stead.

Part of me wants Adelaide to runaway with this ravenous form and thereby give this season something of quality to hang its hat upon (and are deservedly lording the tower now). But my NSW teams are knocking on the door of another GF. Hell, even the Eagles are still a chance for some (brief) September action, as well as some relatively strong showings from long starved clubs.

As long as we don't get something inept like the aforementioned Geelong-Richmond prelim. Please no.

 

[Hobbes]

Yeah, but I just want to see those scores

The squiggle would tip Adelaide by 3 points. It applies a 12-point advantage to the home team, and it's tipping Adelaide to win by 15 points at Adelaide Oval.

 

[The Swans Blog]

Cats making the grand final? Unbloodylikely.

Crows turning it on at the 'G? Pigs will fly before that happens.