Analysis Was Port's Analytics sending an indirect message?

ozph1870

Brownlow Medallist
Mar 3, 2014
12,121
20,265
AFL Club
Port Adelaide
Other Teams
dallas cowboys ferrari f1
I'm of the opinion this analytics goes to far in regards to our style of play. To much is taken from it as the trends of the game, and we play to suit that, instead of playing to suit the strengths of our team. It infuriates me, I'm not a big fan of analytics in general. I found it telling what Wingard said about it last week, and how we use it. I feel we are to far down the path with analytics. There is room for it in the game, but it can't dictate how we play, which I feel it does.

You don't need analytics to tell you as a footballer, kicking to players 45 metres out, or deep in the pockets results in harder shots on goal, you know that as a footballer. You don't need it telling you where to kick the ball into the forward 50 for better % shots on goal, you know that, or should at any level of football let alone AFL footy.

We should be playing what's in front of us as Schofield says, rather then confusing players at times which I feel analytics does during play.
100% correct
 
Aug 13, 2007
2,644
2,682
NSW
AFL Club
Port Adelaide
Other Teams
36ers
I think analytics should be used as a teaching aid to get players to buy into things like team defence.

I agree that I don't think it works for trying to score in AFL.
Both forward lines are efficient when we have tall forward.

Take out Frampton and Marshall and the Maggies are inefficient.



On SM-G960F using
BigFooty.com mobile app

I went back and checked our games at both levels and it's definitely better when they are playing. Well when Marshall played well at least.
 
Analytics rules. It's how its interpreted and applied that is the issue.

Good analytics in the hands of buffoons can confirm personal biases without considering the why and how, just that it is.

A good analyst will explain the how and the why and whether there are better alternatives even if a certain model says what they are doing is apparently good as it is.

Analytics is a decision support tool, not a decision maker. I highly doubt we make decisions based purely on analytical models, but they would support or refute certain positions.
 

Kiss from a Rozee

Goddess
Bring Back the Bars
Nov 23, 2018
2,755
4,503
AFL Club
Port Adelaide
I don’t think we have the necessary structure to use analytics. We need to hire 3 full time Data Interpreters and an Information Synthesis Manager in order to adequately allow the transition of Data into Information which can then be Utilised by our coaching staff to achieve Positive Outcomes.

I recommend HugeJohnson for the role.
 
Yeah I'm absolutely sure that analytics is the reason why we went in to that 'always kick in to the pocket and get it out of play then set up for the throw in' strategy a couple of years back.

It was bad then and I'd say we are making equally bad yet less obvious data-based decisions now
Back when Chocco was still coach and we went back to Alberton for our post-match stuff I remember him telling the crowd that kicking long to a marking contest for our forwards was dumb football (not those exact words). I'm guessing we were early in the horrible chippy, chippy days; and the natives were becoming restless. In our heads we fans seemed to remember the exciting long kick working often enough that it outweighed the ability of defenders to punch it away and rebound. But even then data analysis was apparently saying it was dumb football (not those exact words).
So who is right? Sure in the last quarter v Richmond last week we apparently bombed it into our forward 50 for 8 failed attempts to get our hands on it, but who can really say?
 

chickentendies

Cancelled
BeanCoiNFT Investor
Mar 26, 2017
1,930
7,366
AFL Club
Port Adelaide
Back when Chocco was still coach and we went back to Alberton for our post-match stuff I remember him telling the crowd that kicking long to a marking contest for our forwards was dumb football (not those exact words). I'm guessing we were early in the horrible chippy, chippy days; and the natives were becoming restless. In our heads we fans seemed to remember the exciting long kick working often enough that it outweighed the ability of defenders to punch it away and rebound. But even then data analysis was apparently saying it was dumb football (not those exact words).
So who is right? Sure in the last quarter v Richmond last week we apparently bombed it into our forward 50 for 8 failed attempts to get our hands on it, but who can really say?

Which is probably why post 2007 GF, you heard Choco regularly say;
"We don't get it inside 50 often enough, but when we do, we are very efficient."

Hinkley doesn't say this, but you can sum up his tenure as;
"We get it inside 50 often enough, and when we do, we are very inefficient."
 

Harlott

Premiership Player
Jul 5, 2010
3,381
6,708
Adelaide
AFL Club
Port Adelaide
Back when Chocco was still coach and we went back to Alberton for our post-match stuff I remember him telling the crowd that kicking long to a marking contest for our forwards was dumb football (not those exact words). I'm guessing we were early in the horrible chippy, chippy days; and the natives were becoming restless. In our heads we fans seemed to remember the exciting long kick working often enough that it outweighed the ability of defenders to punch it away and rebound. But even then data analysis was apparently saying it was dumb football (not those exact words).
So who is right? Sure in the last quarter v Richmond last week we apparently bombed it into our forward 50 for 8 failed attempts to get our hands on it, but who can really say?
It's dumb football sometimes and smart other times. Depends who's there, how many, how willing they are to contest, ability to mark etc. You can train all you like for scenarios but in the end football is this purely improvised contest where the smart successful coaches I think have the ability to use just the right amount of analytical information to complement the footballer's natural instincts and abilities.
 

agmsy

All Australian
Sep 28, 2014
868
2,753
AFL Club
Port Adelaide
A good analyst will explain the how and the why and whether there are better alternatives even if a certain model says what they are doing is apparently good as it is.

The best analysts are comfortable asking domain experts when they need insight and clarification to understand and improve models and data. The best domain experts are confident enough in their knowledge of the field not to be intimidated by sophisticated models and data.

I wonder if we have either at Port? Given how much of what the club does is flawed, it's probably safer to assume this particular tail is wagging the dog, too.
 
May 26, 2017
20,813
42,868
Uruguayana, RS (BRA)
AFL Club
Port Adelaide
Other Teams
Grêmio, DC United, Pistons
Too much wanky data and not enough simple football analysis.. The greatest coaches of all time weren't relying on shitty data, they had a general feel for the game and worked it out for themselves... Way to many pointless frauds hanging around footy clubs these days.
There are two equally wrong approaches to stats: disregarding them as useless; and believing they are flawless.

They are tools that, properly understood and handled, may be very useful. Nothing more, but nothing less either.
 
Last edited:
The book Footballistics (great read) confirms that we hired him as he basically was writing about something the club already believed in re; scoring shots.
Thanks for posting that. It reminds me that August last year in Inside Sport they published about 2/3rds of chapter 2 from the book including the appointment of Robert younger and his explanation of his ideas.

You can read that Inside Sports article at this press reader link. A few diagrams and charts are missing from Chapter 2 of Footballistics that Robert Younger wrote/contributed. GremioPower have a read. Let me know if you are geoblocked.

https://www.pressreader.com/australia/inside-sport/20180801/281530816778936
 
Last edited:

StrappingTape

Norm Smith Medallist
Feb 29, 2012
6,170
7,226
AFL Club
Port Adelaide
I'm all over stats and analytics being used in the nba but it just doesn't apply anywhere as much in AFL. You don't need an analytics genius to tell you you will kick a better percentage of goal attempts if you are taking shots with a lot lot lower degree of difficulty. In basketball there's only five on five going one direction on a small field with a small goal thus the data and research we can gain on improving plus game plan matchups is ridiculously good.

At AFL level sure use some stats but I think use it for individual personal decisions within our game plan and tactical gameplans against other teams but not for building our playing system/style. As we have witnessed ourselves you have to be a ******* idiot to spend a season or two kicking to pockets because stats and then have a guru come in to tell you oh hey hang on you will score more from like kicking at goals with the best angle..
 

agmsy

All Australian
Sep 28, 2014
868
2,753
AFL Club
Port Adelaide
You don't need an analytics genius to tell you you will kick a better percentage of goal attempts if you are taking shots with a lot lot lower degree of difficulty.

You identify, from a footy fundamentals and first principles stand-point, that you want to generate the easiest possible shots for your forwards. The question is, "How do we generate the easiest possible shots?", and the analyst's aim is to help answer this question with models and data.

In basketball there's only five on five going one direction on a small field with a small goal thus the data and research we can gain on improving plus game plan matchups is ridiculously good.

When modelling a system, the quality of your model depends on two factors: complexity and data. It's easier to model simple systems, and it's easier to model systems with abundant data. It's difficult to model complex systems, and it's difficult to model systems when data is scarce.

In the 2017-18 NBA season, the league averaged 106.1 ppg. 106.1 points per team, accumulated 1, 2, or 3 points at a time. That's in a league that hits a bit under half of all shots from the field, just over a third of all shots from outside the arc, and about three-quarters of free-throws. That makes for a huge amount of data. Simple system, abundant data, good models.

In the 2018 AFL season, the league averaged 83.5 ppg. 83.5 points per team, accumulated 1 or 6 points at a time. In a league where just over half of all shots at goal resulted in a goal. That's really not a lot of data. Complex system, scarce data, weak models.

At AFL level sure use some stats but I think use it for individual personal decisions within our game plan and tactical gameplans against other teams but not for building our playing system/style. As we have witnessed ourselves you have to be a ******* idiot to spend a season or two kicking to pockets because stats and then have a guru come in to tell you oh hey hang on you will score more from like kicking at goals with the best angle.

Maybe, but data gets even scarcer at an individual level, and that causes even more problems. I actually think gameplan tweaks and changes are the best contemporary targets for footy analytics, because you can base them on the largest, aggregated, data sources.

The "just get it forward and then worry about squaring it up" plan was always doomed to fail because it ignored how the game tightens up as you move the ball forward. If you give your opponent time to get numbers back, man-up, and fill space, they will. That doesn't change the probability of scoring from your desired zones, but it changes the conditional probability of scoring from your desired zone given that you now have to get the ball there in congestion. Perhaps our data overstated this conditional probability because our fitness advantage in 2013-2014 overstated our group's precision by hand and foot?

Our current "run-and-gun" style works because we can deliver the ball to forward targets before the opposition can clog their defence. If our patterns of efficiently and effectively delivering the ball inside 50 are data-driven, that's a good thing. If more data and analysis helps us refine or improve our run-and-gun style, to maintain or increase an edge, that's a good thing to.

ANYWAY, to tie all of this back to the original tweet that Younger liked, it's not immediately apparent that more money would help in this instance, but he is in a much better (albeit partial) position to judge than most of us are.
 

StrappingTape

Norm Smith Medallist
Feb 29, 2012
6,170
7,226
AFL Club
Port Adelaide
You identify, from a footy fundamentals and first principles stand-point, that you want to generate the easiest possible shots for your forwards. The question is, "How do we generate the easiest possible shots?", and the analyst's aim is to help answer this question with models and data.

Yeah.. you don't need data to answer that one. You work out do we want to get them running over the back a lot or from front on and then start devising a game plan from that. We can use some modelling to look at how many successful goals were kicked either of our preferred methods etc but yeah no square peg round hole on this one.

When modelling a system, the quality of your model depends on two factors: complexity and data. It's easier to model simple systems, and it's easier to model systems with abundant data. It's difficult to model complex systems, and it's difficult to model systems when data is scarce.

In the 2017-18 NBA season, the league averaged 106.1 ppg. 106.1 points per team, accumulated 1, 2, or 3 points at a time. That's in a league that hits a bit under half of all shots from the field, just over a third of all shots from outside the arc, and about three-quarters of free-throws. That makes for a huge amount of data. Simple system, abundant data, good models.

In the 2018 AFL season, the league averaged 83.5 ppg. 83.5 points per team, accumulated 1 or 6 points at a time. In a league where just over half of all shots at goal resulted in a goal. That's really not a lot of data. Complex system, scarce data, weak models.

I'd argue that NBA is complex, extremely complex.

Maybe, but data gets even scarcer at an individual level, and that causes even more problems. I actually think gameplan tweaks and changes are the best contemporary targets for footy analytics, because you can base them on the largest, aggregated, data sources.

The "just get it forward and then worry about squaring it up" plan was always doomed to fail because it ignored how the game tightens up as you move the ball forward. If you give your opponent time to get numbers back, man-up, and fill space, they will. That doesn't change the probability of scoring from your desired zones, but it changes the conditional probability of scoring from your desired zone given that you now have to get the ball there in congestion. Perhaps our data overstated this conditional probability because our fitness advantage in 2013-2014 overstated our group's precision by hand and foot?

Again though you don't need an analyst to tell you if you're slow getting the ball to where you want to get it by the time you get it there the opposition will have filled the space of your preferred method. What you could though is have an analyst using stats on players so ours are footy iq'd up that in this position when they look up and they see 3 matchups in the preferred area that is our first option to go to in attack that one of the defenders 80% of the time plays infront so if you kick it to that match up but put it over the back and not out infront then we will have the best chance of a mark or a shot at goal within our game plan and how we want to play.

What you are doing here is again trying to fit analytics to explain something in AFL where it doesn't work. Infact you've described what coaches have been struggling with since day dot, the opposition is trying to screw the best avenue to goal.

Our current "run-and-gun" style works because we can deliver the ball to forward targets before the opposition can clog their defence. If our patterns of efficiently and effectively delivering the ball inside 50 are data-driven, that's a good thing. If more data and analysis helps us refine or improve our run-and-gun style, to maintain or increase an edge, that's a good thing to.

It's worked well once for one week. A change to forward 50 delivery that can be explained by the complete * up the week before where again you don't need data in AFL to tell you kicking the ball to a midget forward line in the air is not gonna work and you'll get out marked. Infact I'm sure the data would well should have told us exactly that and if modelled surely the scenario's would have significantly outweighed playing Sam Gray at CHF most of the night.. which would indicate then we suck ass at data.

ANYWAY, to tie all of this back to the original tweet that Younger liked, it's not immediately apparent that more money would help in this instance, but he is in a much better (albeit partial) position to judge than most of us are.

More money would help if we need more resources to beef up data on individuals if it is where we lack for both our own list and every other player in the competition. I want our players to have it driven into them NBA style what each opposition player does, I want them to know what they are going to do as they do it so that when we are playing to our game plan and we do run into decisions on the ground we have an upper hand thanks to our data. As well as putting our right players in the right positions of when we give them the ball\position them in certain situations.

Not to mention money balling in potential targets to fill gaps.
 

agmsy

All Australian
Sep 28, 2014
868
2,753
AFL Club
Port Adelaide
Yeah.. you don't need data to answer that one.
What you could though is have an analyst using stats on players so ours are footy iq'd up

These statements are directly contradictory. I am specifically advocating an analytics program that increases our players' knowledge and understanding re: situational football. I'm also arguing in favour of a program where data-based insights motivate broad aspects of the game plan, rather than restricting yourself to "over the back" or "from the front" and finding the best approach within that framework.

I'd argue that NBA is complex, extremely complex.

Sure, but it was meant as a comparative statement, not absolute. NBA is complex, but it is less complex than AFL. Less moving parts. Less ability to "reload" on any given possession or play, but more ability to "reload" once in scoring range. The power to halt play and run highly choreographed and situational set plays.

What you are doing here is again trying to fit analytics to explain something in AFL where it doesn't work. Infact you've described what coaches have been struggling with since day dot, the opposition is trying to screw the best avenue to goal.

You might consider using this to keep the birds away from your crops.

I'm not saying that data can solve all problems. I'm not saying that all formal models are better than the implicit and heuristic models people develop to explain phenomena. I completely agree that building a data-driven model that shows only that slow ball movement doesn't work, or that our fast ball movement against the Eagles did work, is a waste of time and resources, because you can figure this out by watching the game.

However, I'd add that:
  • last week is not the only week we've played a "run-and-gun" style. There's been a noticeable effort to use it all season, and it's, in many ways, similar to how we played in 2014
  • a data-driven model could easily have identified that slow ball movement is dangerous, but that, based on the model's inputs, we were sufficiently skilled to derive a net benefit from trying to follow the "get it forward, then square up" strategy anyway (garbage in, garbage out). In quaint mathematical terms, a model could well say P(Goal|shot from position X) < P(Goal|shot from position Y and the ball is passed to Y from position X) if P(Receiving the ball in position Y with sufficient ability to score|ball is passed to Y from position X) is large enough. If garbage data overstates that last probability, how do you figure it out except by sampling from the new (flawed) regime, and how much data do you need to sample under that regime to be confident that the problems are rooted in a faulty assumption and not, say, players learning a new system, or statistical variance doing its thing?
  • if you've got a problem of significant interest that has yet to be solved using "traditional" methods and the heavy cognitive biases of human-memory-based data analysis, a well-motivated and well-managed data analytics program may help. You may still find that the problem is too complex, or your data is rubbish, or you can't collect enough data.
  • a well-motivated and well-managed data analytics program works hand-in-glove with the rest of the footy department - analysts and domain experts must be in constant and clear communication to make this work.
you don't need data in AFL to tell you kicking the ball to a midget forward line in the air is not gonna work

No argument here, but I doubt this "strategy" was borne of either traditional footy smarts on the part of our coaches or a model produced by a competent analytics team.

More money would help if we need more resources to beef up data on individuals

As part of the football department, increasing analytics spending means either robbing Peter to pay Paul or raising enough money to cover what we miss out on from (or have to pay into) the equalisation fund - a tall order for a club that was without a joint major sponsor as recently as last season.

There's also a ceiling on the amount of data that can be collected, and, in an environment as dynamic as AFL, a good chance that player-level data may never hold the predictive power some hope (and others insist) it does.

Not to mention money balling in potential targets to fill gaps.

I think we're a long way off having the kind of liquid player acquisition market moneyball needs, and likely much farther away from being able to describe AFL in anything like the detail that sabermetrics describes baseball, unless you're prepared to accept estimates with very, very large variance.

Back in 2014, at least one media pundit (Healy on On the Couch, I think) described us as being predictable to ourselves but unpredictable to our opponents. This is exactly the sort of edge that a good analytics program can provide and maintain.
 
Back