Certified Legendary Thread Race for the flag, in squiggly lines

[Final Siren]

Can we just accept that the Squiggle is great!
It's not every year we have a Bulldog level event to question our faith. Reaffirm your Squiggle faith by looking at every other squiggly year. This is currently the best football prediction model there is. It's clear, follows mathematical rules and isn't made up like power rankings.

In all seriousness, the best (public) football prediction model is probably the one operated by Darren O'Shaughnessy, who works for Hawthorn and also runs a thing called Rankings Software. That's the only model I've seen beat the Squiggle consistently enough to feel fairly confident it's better.

There are several other models with a stronger mathematical foundation than Squiggle, but don't beat it often, for whatever reason.

Plus of course there's BigFooty's own mad genius Roby, who seems to be particularly good at picking upsets, but doesn't publish in a way that's easy to verify.

Darren O'Shaughnessy does some kind of subscriber-only betting thing, which I don't recommend because I think the road to Hell is paved with the empty wallets of people who thought they could beat the system, but you can follow him on Twitter, where he posts a lot of interesting analytics stuff, including articles like this.

 

[LittleG]

In all seriousness, the best (public) football prediction model is probably the one operated by Darren O'Shaughnessy, who works for Hawthorn and also runs a thing called Rankings Software. That's the only model I've seen beat the Squiggle consistently enough to feel fairly confident it's better.

There are several other models with a stronger mathematical foundation than Squiggle, but don't beat it often, for whatever reason.

Plus of course there's BigFooty's own mad genius Roby, who seems to be particularly good at picking upsets, but doesn't publish in a way that's easy to verify.

Darren O'Shaughnessy does some kind of subscriber-only betting thing, which I don't recommend because I think the road to Hell is paved with the empty wallets of people who thought they could beat the system, but you can follow him on Twitter, where he posts a lot of interesting analytics stuff, including articles like this.

Will look up his stuff. But not a football punter, not interested in that aspect. Am more a software designer and love the new data models you can create.
I have read Roby. It's kind of sad, rather than informative. He goes for upsets too often, but IF you have large pockets he will produce a return IF you don't follow every bet.
There was no identifying the Bulldog event, unless you have weighted contested possession differential and player injuries. But there was no historical precedent, so why would you account for variables that have not been predictive in the past.
Looking forward to the Bulldogs v Giants v Swans fight this year. Can see 4 new teams in the eight this coming year, as the evenness of the competition makes it harder to stay up for longer.

 

[AlibiMonday]

In all seriousness, the best (public) football prediction model is probably the one operated by Darren O'Shaughnessy, who works for Hawthorn and also runs a thing called Rankings Software. That's the only model I've seen beat the Squiggle consistently enough to feel fairly confident it's better.

There are several other models with a stronger mathematical foundation than Squiggle, but don't beat it often, for whatever reason.

Plus of course there's BigFooty's own mad genius Roby, who seems to be particularly good at picking upsets, but doesn't publish in a way that's easy to verify.

Darren O'Shaughnessy does some kind of subscriber-only betting thing, which I don't recommend because I think the road to Hell is paved with the empty wallets of people who thought they could beat the system, but you can follow him on Twitter, where he posts a lot of interesting analytics stuff, including articles like this.

Darren is Rourke for those interested.

 

[Belnakor]

Darren O'Shaughnessy does some kinds of subscriber-only betting thing, which I don't recommend because I think the road to Hell is paved with the empty wallets of people who thought they could beat the system, but you can follow him on Twitter, where he posts a lot of interesting analytics stuff, including articles like this.

its a bit like some of those horse racing syndicates, if the results were so good why would you share them... most of them are just schemes for people who are front running.

 

[Final Siren]

Darren is Rourke for those interested.

Oh I didn't know that. That's cool. Hi Darren!

 

[old55]

Computers didn't fail, the programmers did. They made so many false assumptions, that they were calculating the popular vote and not the regional vote. These are some of their terribly inept assumptions:
-Uniform national swing
-all votes are equal
-African Americans will all turn out as strongly for Hillary as for Obama
-Likely voters will vote in a uniform percentage for each candidate
-polls in all states matter
-equal waiting of polls in major cities, compared to minor cities

Trump targeted just 4 states and the 64 votes he needed to win. Hillary targeted every vote and tried to appeal to everyone.
I could write another 40+ pages on this but in conclusion, computers are only as good as their programming. Until we make AI that program themselves.

It's not programmers or computers who failed. Programmers encode instructions provided by subject matter experts. It's the research demographers who failed in this case.

 

[Yakker]

It's not programmers or computers who failed. Programmers encode instructions provided by subject matter experts. It's the research demographers who failed in this case.

No no no..... it's not the research demographers who failed. It was the footy teams that failed to win/lose as predicted that failed!!

 

[Rotayjay]

Jeez, people are quick to jump off the Squiggle bandwagon when Sydney chokes again and the Bulldogs pull off one of the best finals campaigns in the history of footy. Squiggle is a model that shows roughly how well teams attack and defend. It doesn't account for the humanity of this great sport, and it is not a get-rich-quick betting tool.

 

[Dogs_R_Us]

Also, just a suggestion for next year, if it's not too much trouble for Final Siren , to put a link to each squiggle in the OP? I know that means 27 links but saves trawling back looking for certain squiggles. Please?

 

[Rourke]

In all seriousness, the best (public) football prediction model is probably the one operated by Darren O'Shaughnessy, who works for Hawthorn and also runs a thing called Rankings Software. That's the only model I've seen beat the Squiggle consistently enough to feel fairly confident it's better.

There are several other models with a stronger mathematical foundation than Squiggle, but don't beat it often, for whatever reason.

Plus of course there's BigFooty's own mad genius Roby, who seems to be particularly good at picking upsets, but doesn't publish in a way that's easy to verify.

Darren O'Shaughnessy does some kind of subscriber-only betting thing, which I don't recommend because I think the road to Hell is paved with the empty wallets of people who thought they could beat the system, but you can follow him on Twitter, where he posts a lot of interesting analytics stuff, including articles like this.

That's very kind, thank you. I find your visualisations the best of their kind, can't get enough of the Squiggle! Just to clarify, I don't bet on football and have no real interest in doing so in most sports. Consulting to Hawthorn since 2012 I am prohibited from betting anyway. The models that I've used with reasonable success in the Monash tipping comp for the past 20 years have these principles:

• don't overreact to recent form
• don't overreact to player ins & outs
• don't try fitting specific team/ground home advantage because you'll overfit
• don't use the scores, use a limited mix of relevant indicators that correlate with scores but have more information and less luck

The subscriber site you're talking about is TedSport / TedBet, run by Ted Hopkins. My involvement is on the technical side. Its main point of difference is that it has a dynamic stake calculator to manage risk in-play, but mostly people just want footy tips and decent advice about staking strategy. The other advantage is a trained crew that collect observations live on each game.

I agree with your statement: "the road to Hell is paved with the empty wallets of people who thought they could beat the system". Your model is only as good as its assumptions, and even if it's right it can be wrong often enough for you to lose money. You would often do better investing in regular stocks, but most people bet for the thrill of it.

 

[Rourke]

its a bit like some of those horse racing syndicates, if the results were so good why would you share them... most of them are just schemes for people who are front running.

Yes, most of the people with successful strategies are keeping them secret. I don't see much evidence of front running from the betting touts out there though, I think they're just fudging their profit figures when it suits them (or going out of business).
TedSport's approach is to share the staking algorithms, which should encourage people to bet less on their wild instincts. Potentially not a money earner like the Bill Vlahos / Bernie Madoff "business model", but worth a shot.

 

[threenewpadlocks]

• don't use the scores, use a limited mix of relevant indicators that correlate with scores but have more information and less luck

I've always wondered this. Accuracy given a shot's location and pressure is generally considered "lucky" so I wonder if the Squiggle would be more accurate if it did everything exactly the same but used expected score rather than actual score.

 

[Final Siren]

It's an interesting concept. For most sports, it's pretty clear that there are more accurate ways to measure team strength than raw scores. But Aussie Rules is so high-scoring, it's a more reliable indicator, and the relative effect of luck (or whatever you want to call random-like unpredictability) is smaller. So you can go a long way with scores alone.

You can also feel confident that every team is trying to maximize its score, whereas with other metrics you have to be careful that they apply equally well to all teams and all game plans.

The idea with "expected score" is that whether a player kicks a goal or a behind is purely a matter of probability based on where they are and how much pressure they're under. So then you can figure out what a team should have scored, based on their types of shots, if luck was equal. Figuring Footy produce a lot of great visualizations on this.

It seems intuitively like a good idea to me, even though it requires a bit of crudity to categorize "pressure." I'd expect it to produce better results than using raw scores. But not much better results, since high scores contain a lot of information already.

 

[puntking]

gws think has it in the bag swans dogs will be there but gws will rally shine next season

 

[Rourke]

You can also feel confident that every team is trying to maximize its score, whereas with other metrics you have to be careful that they apply equally well to all teams and all game plans.

This is absolutely true, you have to pick and choose your metrics and hope they are not too contaminated by the state of play / game plans. Things like being able to stop continuous opposition ball movement, create high-quality attacking opportunities, and consistently put pressure on the opponents --- not teammates --- are robust in many invasion sports.

Expected score is good in that it adjusts for the most obvious source of luck. There are others you can adjust for as well (my paper from this year has one idea). You're looking at a standard deviation of at least five goals in the final margin compared with inherent ability, so you can grind out an edge by assessing it. Just not the ad hoc way Roby does!

 

[SwampCreature]

None of the last page or so explains to me how Roby can so consistently predict the future.

I maintain the view he is a wizard.

 

[Final Siren]

This is absolutely true, you have to pick and choose your metrics and hope they are not too contaminated by the state of play / game plans. Things like being able to stop continuous opposition ball movement, create high-quality attacking opportunities, and consistently put pressure on the opponents --- not teammates --- are robust in many invasion sports.

Expected score is good in that it adjusts for the most obvious source of luck. There are others you can adjust for as well (my paper from this year has one idea). You're looking at a standard deviation of at least five goals in the final margin compared with inherent ability, so you can grind out an edge by assessing it. Just not the ad hoc way Roby does!

That paper is fascinating.

Luck is really interesting to me, because I used to think you make your own luck in football and there are no such things as undeserving winners. (Unless the umps rob you. Like that goal umpire who stood in front of the ball in Perth. HOW HARD IS IT TO STAY BEHIND THE LINE. WE COULD HAVE BEEN TOP 4 YOU PRICK.)

I'd guess that's what you're told if you're a player, too, since it's motivational. It would be hard to push your body to its limit if the outcome is largely dependent on random factors beyond your control.

And the media will craft a totally different story about a match depending on whether it was a 1-pt win or 1-pt loss. You actually hear them changing the narrative of the match every time a team gets a run on.

And as human beings we naturally seek out meaning. We want everything to happen for a reason, and to figure out what that reason is, because if life is just things happening randomly for no reason, that's scary.

But as I've gotten into stats, it really does look like better teams lose all the time. And sometimes luck determines premiers.

I dunno if this is good or not. On the one hand, I'd hate for football to be so predictable that you can tell who's going to be premier from Round 5. On the other, it kind of destroys the romance if success is luck.

Either way, though, it's definitely a big part of the grand delusion that is football. It really is a kind of mass insanity even on the basic level that we all invest so much hope and fear and thought into a made-up contest that has absolutely no relevance to anything outside itself at all. I think a lot of footy is actually story-telling.

 

[Dr Tigris]

I love the philosophical tact this has taken.

Plus the discussion on modeling and luck.

Good work guys, keep it up.

 

[lamby]

That paper is fascinating.

Yep:

The graph in Figure 2 caused surprise at Hawthorn FC, and requires substantial thought by experts in the sport
as well as statisticians. The coaching group discussed what it meant if the centre clearance was effectively a
coin flip, and how they might structure their defence, midfield and attack to respond to an event they have very
little control over. The strong hint in the data that teams might over-invest in winning the clearance at the
expense of resources in more impactful locations and roles is also a lesson to consider.​

The Hawks had the 2nd lowest centre square clearance rate for the year, but finished first.

 

[GreyCrow]

Yep:

The graph in Figure 2 caused surprise at Hawthorn FC, and requires substantial thought by experts in the sport
as well as statisticians. The coaching group discussed what it meant if the centre clearance was effectively a
coin flip, and how they might structure their defence, midfield and attack to respond to an event they have very
little control over. The strong hint in the data that teams might over-invest in winning the clearance at the
expense of resources in more impactful locations and roles is also a lesson to consider.​

The Hawks had the 2nd lowest centre square clearance rate for the year, but finished first.

That was also the time Clarkson was chided ( by me and others) for coming out and saying '' Clearances don't matter''

 

[Roovolution]

Additionally, Hillary didnt excite her base to vote generally (contrast Bill and Obama), Trump had hidden voters not picked up in polls, and contrary to opinion Trump did well with richer folk.

Hilary came with a lot of baggage which people kept downplaying. Benghazi, email scandal, her foundation, the secret handshake from within her own party to get the nom ahead of Bernie Sanders, her husband and the fact that she represented a clear indicator that she was an elitist and the system was corrupt and didn't allow outsiders who would really change things. Trumps candour and crass comments actually worked in his favour and the fact that his own party were considering dumping him just reinforced that reputation even more.

 

[LittleG]

One of the issues with the scoring and likely scoring model is the impact of large early leads. Many times a team can be 8-10 goals up and ease up. Players go into self preservation mode and time wasting mode, reducing defensive pressure and attacking intent.
I saw this in many matches in the last few seasons. Perhaps it was just the matches I attended but going defensive when a long way ahead was very common.
Not sure how you account for this.

 

[lamby]

That was also the time Clarkson was chided ( by me and others) for coming out and saying '' Clearances don't matter''

I think the key is that Clarkson set up knowing that he is a more than even chance of losing the centre clearance so setting up that way. Other coaches set up expecting to win the centre clearance so get into trouble when they start losing more clearances than they win.

So data analysis like this probably played a part in giving the Hawks enough of an edge to win a flag or 3!!

 

[Dr Tigris]

This is absolutely true, you have to pick and choose your metrics and hope they are not too contaminated by the state of play / game plans. Things like being able to stop continuous opposition ball movement, create high-quality attacking opportunities, and consistently put pressure on the opponents --- not teammates --- are robust in many invasion sports.

Expected score is good in that it adjusts for the most obvious source of luck. There are others you can adjust for as well (my paper from this year has one idea). You're looking at a standard deviation of at least five goals in the final margin compared with inherent ability, so you can grind out an edge by assessing it. Just not the ad hoc way Roby does!

Just to bounce off you.

What I took from your analysis - ignoring the larger issues the paper was actually about - was that a lot of center bounce results are about luck. So if as a coach you set up knowing that you can get a small, but consistent, advantage over an opposition that doesn't recognize this. The message isn't that contested ball, or center bounce wins are unimportant. It is quite clear from your paper that they are (+1.06). But, it is that thinking in a deterministic way provides an advantage to the coach that thinks probablistically. So what I took is that the analysis can help coaches and players think more clearly about how the game actually works, compared to how we traditionally think about it.

Thus, knowing that center bounce clearances = advantage, a coach would naturally focus a lot of attention on obtaining center bounce clearances. They would play more, and better quality players, in the center bounce area. They would set up for how best to take advantage of winning to center bounce. But, if you accept luck is probably more important you would set up to try and nullify the center bounce as much or more than trying to win it. And you would set up more and better quality players to 1) take advantage of a randomly bouncing ball, and 2) you would bias your structure a bit more toward losing the center clearance, knowing that because the ball will come in fast, the opposition will not have a good structure if you can get the ball and bring it out. You can tilt the (nominal) field of play in your direction. But, what you wouldn't do is to give up the center bounce entirely because then the opposition would change their set up knowing they would have control of the ball. Then you would essentially have a game style of playing in your defensive half and hoping to stop ball coming in fast and cleanly, then rebounding. The opposition coach would structure accordingly, boosting the center clearance and playing extra guys behind the ball. But, if you just adjusted a little bit you would still get some center clearances and force the opposition to set up to win the center clearance as priority, allowing you to tilt advantage your way.

Is that the right way to think about this sort of thing??

 

[TheMcManusNose]

One of the issues with the scoring and likely scoring model is the impact of large early leads. Many times a team can be 8-10 goals up and ease up. Players go into self preservation mode and time wasting mode, reducing defensive pressure and attacking intent.
I saw this in many matches in the last few seasons. Perhaps it was just the matches I attended but going defensive when a long way ahead was very common.
Not sure how you account for this.

Yeah the NHL has a massive issue with this. Because goals are relatively more rare, going into a defensive deployment happens very quickly.

So you have metrics like score-tied or score-adjusted Corsi/Fenwick e.g. http://www.broadstreethockey.com/2012/1/23/2722089/score-adjusted-fenwick

Be interesting to see if one could make a fist of scoring attempts or I50s (or whatever is the best metric) in AFL when the score's within and outside of an appropriate margin (whatever that is).