The Forest from the Trees: A post for the stats nerds


#441

Ummm, you know how these things work right? It’s incumbent on you to understand the methodology and where there are assumptions etc before you can hold anyone “accountable” for not getting “the right answer”.

Go and review their methods and tell us where the weaknesses are, and then we can actually have a discussion. Just trying to “hold them accountable” based on whether Essendon finishes 7th or not is not understanding what they are telling you.


#442

No. We don’t have to analyse their algorithm. Squiggle profess to have a model thats works and they promulgate their prediction on the web. So all we have to do is evaluate their reputation over time based on whether it works. That gives us an idea as to whether we put them in the bin with CD or give them some credence.


#443

Or maybe not.

You don’t “put CD in the bin”, as much as it will get some easy likes. You actually understand what data they’ve collected, what is meaningful and whether they’ve asked it meaningfully, v and then you decide whether you value the conclusions they’ve drawn based on that data.

Everyone wants an easy stat that tells them the entire story. There is no such thing. Like life there’s subtlety and nuance, and you have to understand that when you’re trying to interpret these numbers. If you aren’t doing that you may as well be a HS journalist.

The simple takeaway from the squiggle stats numbers were that it’s going to be really close at the top.

The next take away is that it’s pretty unlikely that the algorithms have the precision to differentiate meaningfully at that level. The next takeaway is that you can’t help luck, either through injuries, or the vagaries of when you play other teams wrt their form, or in terms of umpiring or the bounce of the ball.

And when it’s a close season where those things will make a difference to expect any of the statistical algorithms to be accurate across a season in terms of W-L is madness.

But you know, it’s easy to “hold them accountable”. Much harder to try and understand what they’re doing and provide meaningful comment on it.


#444

I dont disagree with your points.
Its just a bit of fun. Holding them accountable? I am not going to sue them, I am not going to bet on the ladder or the premiership.

Next September it will be fun to see how the ladder looks. The Squiggle forecast has been uploaded to enable a simple comparison.


#445

One of the key features of the Squiggle’s methodology, is that, historically, dramatic rises and declines in performance are rare. It’s extrapolating based on how we finished 2018, and it gives no farkes about off season recruiting (these, again historically, rarely have the impact that your every day pundit suggests they will).

The fact that it likes us as the #7 team for 2019, with it not knowing or caring about the potential additions of Shiel and Daniher to our 2018 line up, is a very good thing.

Other observations:

  • Richmond are still the team to beat. One bad prelim performance does not a bad team make (see Essendon, circa 1999).
  • Melbourne were ticking most of the boxes that algorithms like to see (i.e. flogging teams - the margin counts according to these things) and should be a force in 2019.
  • Collingwood rated fairly low for a grand finalist, I suggest narrowish wins and a soft 2018 draw have the Squiggle wary.

Geelong might the most interesting litmus test for the Squiggle in 2019. They have been discounted as a factor by many pundits, but Squiggle likes their form line. They have a habit of winning BIG vs weak opponents and keeping their losses tight.


#446

Thanks BBB, interesting insights.


#447

Champion Data confirming that Hurley’s one on one defending (Buddy demolition job aside) has gone to hell.


#448

Does not mean what almost everyone will think it means.


#449

Over the last 2 years, especially 2017, Hurley has not played as a 1 on one defender. So I hardly see that the premise of including him in that rating is appropriate. Its like putting Johannisen or BJ in that group. Just because Hurls is tallish, doesn’t mean he always plays tall one on one. For us, a more appropriate candidate is Hooker, or Ambrose


#450

Whatever it means I’m hoping Francis takes over the zoning role and Hurley gets back to focusing on playing in front and beating his man, like he did in the buddy game.


#451

What does it mean? Does it mean that every time Hurley is in a one on one in the defensive 50, 20% of the time his opponent has a shot on goal? If it is, I’d say beating you opponent 4/5 times is not bad. But I’d also say it doesn’t mean that because I’ll be buggered if Lynden Dunn wins 95 out of every 100 contests.

Last year Hurley only really had one match where he played on a key forward for the whole game and he pantsed him. He also beat Dustin Martin in contests.


#452

Pretty sure that’s what it means. A mate who is a mad pies fan (I know, I know) raved about how good Lynden Dunn was last year - I rolled my eyes but maybe he was right.


#453

Whatever it means will be irrelevant in 2019.

Just have a feeling that Ruttens work will really help Hurls to return to that beast one on one defender we know he can be, while still doing all those other things we need him to do.


#454

It’s the same old CD flaw.
Who knows how they define one on one contests?

I’d be amazed if they defined it as “playing on your designated opponent and the ball comes directly to a 50/50 contest with no support from other players and your direct opponent wins the contest”

As distinct from finding yourself in a one on one contest accidentally but being late to arrive - eg running to cover the loose ruckman or forward pocket.
Or, being suspicious about CD’s wording “leads to a shot on goal “ rather than the clearer “direct opponent has a shot on goal”, which skews the stats towards those defences that cover better for each other if the original contest is halved.

Giving the absolute number of these one on one contests might add clarity.
Or it might not.
Dunn is a good example last year- my memory is of him doing a lot of leaving his man to spoil shorter flankers and wingmen, and punch the ball out of bounds. Are these counted as contests?


#455

Yeah I agree… Richmond (adding Lynch which makes their forward line utterly menacing) and Melbourne (adding May and KK which makes their weak backline suddenly a strength) are the teams to beat this year.

Think GWS will be better if they can keep some of their better talent on the park and also don’t imagine Collingwood plus Beams bring too much worse than last year.

West Coast won a grand final with their two best players in Gaff and Naitanui sitting on the pine. They are still a great team.

Going to be a lot of teams in it to win in next year


#456

I like to see Hurley’s stats per game or first ten rounds to the rest of the season. I reackon he was much better second half and diabolical in the first which would really bring his average down.


#457

In 2017 West Coast finished 8th, C’wood 13th, grand finalists in '18.

Adel didn’t play finals, GWC nowhere near it. Tiges, easily the best team, lost badly to C’wood in final.

As the ‘Forest from the Trees’ title suggests, there are so many factors contributing to where teams finish on the ladder… not least is the finals system where one off day and you’re cactus, regardless of the 22 rounds before (Tiges).

Adelaide, off the back of one disastrous team building exercise and some injuries, went to sh*t in 2018 after looking like they would go one better in 18. AFL teams play on a knife edge in terms of fitness and psychology - doesn’t take much for them to climb quickly or drop down hard depending on coaching, player welfare, systems and luck.

Stats are useful for the opposition in terms of identifying how a team plays and nullifying them. Wouldn’t give them as much credence in being able to accurately predict how a team will perform over a whole season. Our Neeld / post-Neeld rounds showed how valuable coaching, team harmony and player buy-in to a system, is - none of that shows up in Champion Data. From another sport, have a look at Man U this season, Mourinho / post-Mourinho… Player stats overnight went from ‘abysmal, get rid of this guy’ to ‘this guy is the best in his position in the world’, and the only thing that changed was the coach.


#458

I do wonder how they measure quality of delivery to the one on one contest.

To go back to that Ariel Steinberg playing on Ben Brown night as the example. In the first half with our midfield supplying pressure Steinberg was able to do well enough in these contests. When the ball started coming in under no pressure he was immediately no chance. My guessing would be that this sort of theory holds true across seasons and careers. Is the causality with Rance and Dunn that their teams are good because they have job defenders who can win one on ones, or are the defenders good because they are in good teams? Or is it a bit of each.

I reckon coaches over the last decade have made the call that it is the team, and when they’ve had defenders who can hurt going the other way have preferred to use a lesser player to try and shut out the opposition, and “free up” their weapon, when if they’re the best one on one defender in the game.


#459

When a fullback in a good team is winning the best and fairest, then he really is a good player.

A big shout out to my favourite EFC player.


#460

A quick addition - I don’t think it’s counted as a one on one contest if the forward gets a 3 metre break on the lead and marks unopposed. Despite in reality the back being beaten.
Perhaps Hurley can almost keep up and it counts as a contest.
Whereas Dunn is just done.