Top 50 players 2017
Users browsing this topic:0 Registered, 0 Hidden and 0 Guests
Registered Users: None |
 |
| View previous topic :: View next topic |
| Author |
Message |
Pies4shaw 
pies4shaw
Joined: 08 Oct 2007
|
| Post subject: |  |
|
| K wrote: | | Pies4shaw wrote: | | Ultimately, the tragedy is that what gets measured seems to have a rather remote connection, at best, with what wins or loses games. [1a] I am very familiar with the stats used in NFL [2]- many of those are quite unrelated to the skills thought to be necessary to play well in a particular position but, nevertheless, closely-connected to assessing whether the things a player did or didn’t do affected the likely result of the game positively or negatively [1b]. I accept it is easier to do that in games that aren’t continuous. [3] Nevertheless, I understand why a quarterback is rated as “elite” and the same with the various skill positions etc - it’s because the stats say that you are very much more likely to win with player x on your team in that position than the various alternatives, all other things being equal. [1c] Here, the stats don’t seem to assess a player’s likely contribution to a win or a loss. [1d] Thus, I know nothing more about a player’s exposed/assessed ability to impact a team’s win prospects after reading these stats than I did before. I don’t object to this sort of analysis, I’d just prefer it to be assessing something relevant to whether I would like a particular player on my team, rather than another team. |
P4S, Please clarify what you mean. I'm uncertain about your meaning. (My numbering of statements above comes after the statements of interest.)
[2] is good for us and this discussion. It means we have another sport to use as a reference example.
I'm not sure what you're meaning by [1a]--[1d], though. [1a] sounds like you're saying the measured quantities are not correlated well with winning or losing. But the second part of [1b] sounds like you're saying they actually are, with the first part of [1b] sounding like you're just saying these go against received wisdom (which could be wrong). Likewise, [1c] and [1d] seem to be in some tension with one another. Is [1c] "bad" in your opinion, because you think they should show the contributions of player x relative to player y who plays on the same team in a different position, rather than player x in his team relative to player z in the same position in a different team? (I'm assuming in stating that question that there is exactly one player in each position, not multiple players in the same position.)
On [3]: it's not continuity that's the problem, though lots of stoppages no doubt make the data collectors' (not statisticians'!) job easier. The difficulty comes from strong interactions between many objects (players) ---- a many-body problem, if you like. That's why baseball is one of the easiest non-trivial sports to analyse: at any given time, the interactions are very weak for all but a couple of players (very, very weak in the case of the players just sitting in the dugout). |
It isn't difficult - CD's stats (and football stats, generally) do not measure the contribution a player makes positively or negatively to a game on a particular day. They measure activity, not game impact. In the NFL, the analyses do measure (amongst many other things) game impact. I have no idea at all about baseball statistics (I could not care less about the sport, let alone about analysing it) and didn't comment on them. Continuity is, though, the obvious point of difference between analysing the outcome of an NFL play and a passage of play in footy. That is because the NFL statistics are able, amongst other things, to evaluate - and do evaluate - the expected score impact of a player having done or not done a thing at a particular time and it's obviously easier to do that in a sport where play halts immediately after the thing happens and there is a specific agreed outcome from the play (eg,after a 2nd and 10, the ball is placed on the defensive 33 yard line for a 3rd and 27 - Rodgers coverts that 3rd down with a 55 yard pass to Jordy Nelson, we know Crosby is now likely to get a field goal, at least and the analysts all know how special he has to be to make that throw and what assessed points impact it has on the game). From where I sit, the problem with the CD analysis is just obvious - it calls players "elite" because they do more or less of some things than other players typically do, rather than because they have a statistically-identifiable special capacity to influence a game. The analysis of available data will, no doubt, reach that point with increased sophistication over time but, for now, identifying players as "elite" based on them engaging in more or less of a particular kind of on-field activity than some (or most or, for that matter, all) other players is a kind of collective mania. Except in relation to things that we know definitely do influence the outcome of games. So, eg, I'm confident that Peter McKenna was a fabulous full-forward partly because his goals-per-game average was so ridiculously high. On the other hand, having seen him play so many times and recalling full-well what he did, how he did it and his (still) unparalleled accuracy of kicking over distance and from "impossible" angles (when adjusted for the difficulty of the shots he took compared to those that, say, Lockett took, the stats - God forbid that such useful information might be available - would likely establish that he was the most consistently accurate shot for goal in the last 50 years), his goal-per-game stat serves only to confirm what was obvious watching him play.
Speaking positively, the kind of statistic that I find useful is the sort of thing they occasionally show when a particular player is lining up for goal (from this angle, over this distance, this player has a 20/40/90% or whatever prospect of nailing it). We do identify involvements in scoring chains but little, if any, of the available analysis identifies the statistical importance of a particular thing a particular player did (or didn't do) in a scoring passage. By way of trivial example - I always felt that Andy Krak's goal-scoring contributions to the forward-line were particularly valuable because it was reasonably plain that he was creating goals and scoring opportunities that other players on our team would not have created. Without wanting to criticise any player particularly, there are a few we could think of who might have a three-goal game merely because they happened to finish up with a simple goal after a run of play broken open by someone else altogether, whereas if Krak kicked 3, the chances were a couple of them were made "from nothing". What I'm very crudely assessing when I say that about Krak is my (statistically uninformed - because those stats aren't given) view of his "game impact". I don't want to open another round with Qld about Sidey but, eg, the reason I value his contributions higher than plenty of others is because of my view about his game impact - unfortunately, though, it can only be "my view" because we don't keep stats that measure his "game impact", we keep stats that measure surrogates. It is plain enough that a Sidebottom game with 15 effective kicks to targets is likely to be better than a Sidebottom game with 5 effective kicks to targets. But that's more or less where the usefulness of such stats ends - it doesn't really translate between players and provides a very dubious basis for comparison because different players do what they do very differently and with very different game impact - and we measure only what they do, not their game impact.
Have a look at the following career averages and tell me which player you want in the centre:
Kicks Per Game 13.5 versus 13.2
Handballs Per Game 4.5 versus 11.5
Disposals Per Game 18.0 versus 24.7
Marks Per Game 2.7 versus 4.2
Tackles Per Game 0.6 versus 4.5
You want the guy on the right because he engages in more possession, disposal, marking and tackling activities than the other guy? If I tell you the guy on the left is Daicos, you know that (whoever the other guy is - it isn't Pendlebury or Treloar) Daicos is the one you want. That's because you actually want to make a selection based on game impact, not activity - but the stats measure activity and call people who do more of it "elite".
The problem doesn't arise in the same way with cricket statistics. We know that a guy with a solid career-average of 65 batting and 23 bowling at Test level is going to get knighted. That's because the game impact is what's measured by the stats, however crudely. The job of batsmen is to make lots of runs and the job of bowlers is to get them out cheaply, so the averages measure (more or less - and subject to some situational questions about who they achieved their averages against etc) the game impact.
By contrast, the point of footy is not to get a kick or a handball or lay a tackle - rather, it is to do those things in a way that effectively improves one's team's prospects of winning. That's the very thing the CD stats, as presently collated, are incapable of measuring. |
|
|
|
 |
Pies4shaw
pies4shaw
Joined: 08 Oct 2007
|
| Post subject: | |
|
| Ooops. DP |
|
|
|
|
MightyMagpie
Joined: 04 Jun 2013
Location: WA
|
| Post subject: | |
|
P4S Thanks for the comments. I think you correctly highlight what the aim should be: "measure the contribution a player makes positively or negatively to [winning] a game on a particular day" and that is exactly what CD is trying to get to. Are they there yet? No, but they are making progress.
I absolutely agree the ultimate measure is the contribution towards winning a game - that is the progression that sabermetrics has made and continues to make in the last decade or so with, for example, the simplistic OBP being a better measure than batting average moving towards various run contribution and subsequently win contribution measures. AFL is well behind as I said, but CD is moving towards these sorts of measures and the detailed GPS and LPS data now collected will provide considerable scope for this.
I think your cricket example may demonstrate the opposite of what you intend. Cricket stats are probably the most misleading of any I have come across ... take for example the guy who is dropped on a duck, but goes on to score 300. In baseball-type terms those would be unearned runs that should not be credited to the batsman. Also cricket treats the same a gritty first innings century as a worthless century towards the end of an impending drawn match. Similarly, a bowler gets no credit for a jaffa that the gumby at first slip drops.
Your critique focusses on the simple tally of kicks, marks and handballs, yet this is exactly what CD are trying to move beyond. The AFL is in the infancy of analytics and as I said it is imperfect, but clubs are much better placed with the information than not. I think we will soon start to see some value add type measures in AFL, eg contribution towards a goal (and ultimately contribution towards a win) and I have no doubt CD have this in mind with the GPS and LPS data they are logging.
_________________
All We Can Be |
|
|
|
|
K
Joined: 09 Sep 2011
|
| Post subject: | |
|
| Pies4shaw wrote: | ...
It isn't difficult - CD's stats (and football stats, generally) do not measure the contribution a player makes positively or negatively to a game on a particular day. They measure activity, not game impact. In the NFL, the analyses do measure (amongst many other things) game impact. I have no idea at all about baseball statistics (I could not care less about the sport, let alone about analysing it) and didn't comment on them. Continuity is, though, the obvious point of difference between analysing the outcome of an NFL play and a passage of play in footy. That is because the NFL statistics are able, amongst other things, to evaluate - and do evaluate - the expected score impact of a player having done or not done a thing at a particular time and it's obviously easier to do that in a sport where play halts immediately after the thing happens and there is a specific agreed outcome from the play... From where I sit, the problem with the CD analysis is just obvious - it calls players "elite" because they do more or less of some things than other players typically do, rather than because they have a statistically-identifiable special capacity to influence a game. ...
Speaking positively, the kind of statistic that I find useful is the sort of thing they occasionally show when a particular player is lining up for goal (from this angle, over this distance, this player has a 20/40/90% or whatever prospect of nailing it). We do identify involvements in scoring chains but little, if any, of the available analysis identifies the statistical importance of a particular thing a particular player did (or didn't do) in a scoring passage.
...
The problem doesn't arise in the same way with cricket statistics. We know that a guy with a solid career-average of 65 batting and 23 bowling at Test level is going to get knighted. That's because the game impact is what's measured by the stats, however crudely. The job of batsmen is to make lots of runs and the job of bowlers is to get them out cheaply, so the averages measure (more or less - and subject to some situational questions about who they achieved their averages against etc) the game impact.
By contrast, the point of footy is not to get a kick or a handball or lay a tackle - rather, it is to do those things in a way that effectively improves one's team's prospects of winning. That's the very thing the CD stats, as presently collated, are incapable of measuring. |
First, what you say regarding "the kind of statistic I [i.e. P4S] find useful" hints at the same distinction I've been making: it's the minimally processed full individual statistics (raw data, I called it) that I want, not just a single number they made up from all the statistics. [Well, I'd have to clarify "minimally processed, "raw", etc. but that's a topic for a separate post.]
But you do seem to be too hung up on this term "elite". Forget it. It's not really an intrinsic part of the process. Just call these category I, category II, etc., or something else really bland. In fact, ignore the categories altogether.
I am no fan of the company at all (but part of that is their short-sighted practices that harm the game, rather than the question whether they are technically competent or not), but you seem to be mischaracterizing what they are doing to come up with the ratings. Your example is looking at the stats that were all that was available maybe half a century ago. That's misleading when they're clearly using much more info than that [as MM alludes to above]. Their Equity Rating, as they call it, seems to be exactly an attempt (successful or otherwise) to capture game impact rather than activity [again, as MM alludes to above]. Some of the discussion a couple of pages ago was about this. Nicksters used Mitchell as the whipping boy in their examples of alleged lack of impact of considerable activity. Less pleasingly to us, CD used Swan vs. Rioli as their hypothetical example.
Ultimately, the only thing that matters in the sports we're discussing here is the scoreboard. The reason why baseball is much easier to analyze is that when a slugger hits a home run (for example) this is basically an individual act, against an individual opponent, the opposition pitcher, with essentially no influence from the other players on and off the field. Both NFL and AFL are thus much harder than this, because many, perhaps all, of the players on the field will influence the scoring of a touchdown or goal. This explains why it's believed that data analytics in US sports has had the most impact and success in baseball, followed by basketball, with American football lagging far behind. It's simply much, much more difficult in the latter case. |
|
|
|
|
qldmagpie67
Joined: 18 Dec 2008
|
| Post subject: | |
|
P4S the original post was done to highlight the opinion champion data held that at the commencement of 2017 we had by there rankings system the best midfield in the competition.
I highlighted all our players in the top 100 and if there ranking & overall ratings had risen or fallen.
If you go through the posts I actually mentioned as how I saw the failure in the system in the changing roles of players.
Sidey was a case in point
2015/16/17 his ranking and ratings points fell year on year (or there 40 game cycle) but for what reason ?
For mine it's because his role within the team changed.
He appears to play more a defensive mid role compared to an attacking mid role.
Stats seem to suggest that his rebound 50's over that period are higher on average per game than the years prior and conversely his inside 50 averages were down.
To me this means his role changed for the betterment of the team (well in theory results of the team didn't improve but that's not 1 players fault)
So P4S we actually agree that in that case the stats are meaningless there isn't an adjustment for these changes.
Your point about players accumulating touches is correct. I said previously would your team prefer to have Tom Mitchell getting 50 touches or Sidey getting 25 and I said Sidey every time with 25 because 15 of those are going to be meaningful where as Mitchell's percentage of meaningful touches would be considerably lower. That game last year when he had 50 against us I reckon about 12-15 may have been damaging the rest were soft handball receives and a lot of 1-2 give and goes.
When I look at the stats I often wonder how many of them actually mean something to team success and how many are individual success |
|
|
|
|
MatthewBoydFanClub
Joined: 12 Feb 2007
Location: Elwood
|
| Post subject: | |
|
| Sidey has been hindered in the natural game he plays, by having to play team roles in recent years. I have no idea what effect that has had on his champion data scores and I don't care. As long as the sides wins it doesn't matter. When Pendles was injured last year and Treloar was hampered by a groin strain, Sidey was moved into a more permanent midfield role later in the year and ends up winning the Copeland trophy which is his truest indication of his value to the team. |
|
|
|
|
K
Joined: 09 Sep 2011
|
| Post subject: | |
|
| MightyMagpie wrote: | ...
The early modern history of sabermetrics (Bill James, Sandy Alderson et al) is quite interesting... |
MM, do you get baseball data too? e.g. have you purchased the Bill James almanac or whatever it's called, or do you subscribe to the site?
Do you know good sources of baseball data, free online or not free and not online? |
|
|
|
|
MightyMagpie
Joined: 04 Jun 2013
Location: WA
|
| Post subject: | |
|
| K wrote: | | MightyMagpie wrote: | ...
The early modern history of sabermetrics (Bill James, Sandy Alderson et al) is quite interesting... |
MM, do you get baseball data too? e.g. have you purchased the Bill James almanac or whatever it's called, or do you subscribe to the site?
Do you know good sources of baseball data, free online or not free and not online? |
I have a fair bit. The Bill James stuff is pretty old though.
This is probably a good starting point: http://sabr.org/sabermetrics/data
_________________
All We Can Be |
|
|
|
|
K
Joined: 09 Sep 2011
|
| Post subject: | |
|
^^^
Thanks for the info, MM.
| Pies4shaw wrote: | | ... I am very familiar with the stats used in NFL ... |
And let me essentially repeat the question to you, P4S: from where do you obtain your familiarity with the stats used in NFL?
Do you know good sources of NFL data, free online or not free and not online? |
|
|
|
|
K
Joined: 09 Sep 2011
|
| Post subject: | |
|
| MightyMagpie wrote: |
... but for those bagging Champion Data, I would just say this ... AFL is about 20 years behind baseball with analytics (think baseball in late 1990s/early 2000s) and, whilst not perfect by any means, it improves each year and improves the understanding of the game for those taking it onboard. The early modern history of sabermetrics (Bill James, Sandy Alderson et al) is quite interesting and Bill James was regarded by baseball organisation stalwarts as a bit of a crackpot in the early days. |
| MightyMagpie wrote: | ...
I absolutely agree the ultimate measure is the contribution towards winning a game - that is the progression that sabermetrics has made and continues to make in the last decade or so with, for example, the simplistic OBP being a better measure than batting average moving towards various run contribution and subsequently win contribution measures. AFL is well behind as I said, but ... |
MM, I think it's definitely interesting to compare what's been done in other sports. But, for the reasons I explained above, baseball is much, much easier to analyse. It's of course easy to use hindsight, but the on-base percentage example always rolled out seems such an obvious thing... And also let me repeat what I said on an earlier page...
| K wrote: | ...
And there's been a huge mythology built up around baseball, largely the result of a particular guy who apparently has a large gift for story-telling but not an equally large desire to let facts get in the way of a good story. Cue movies with Hollywood A-listers, etc. |
It's not easy to strip away the myths and determine exactly how successful sabermetrics has been. Theo Epstein's success may be the best place to start, and surely he did not have to worry about saving money in the pursuit of success. |
|
|
|
|
MightyMagpie
Joined: 04 Jun 2013
Location: WA
|
| Post subject: | |
|
So K, if baseball is so easy to analyse then please briefly explain the contribution fielding makes towards wins ...
If you look at the work that has been done in that area alone in the last decade you will begin to understand that it is anything but easy. Separating out pitching and fielding contributions is quite analogous to the interdependence issues we face with AFL.
_________________
All We Can Be |
|
|
|
|
K
Joined: 09 Sep 2011
|
| Post subject: | |
|
| MightyMagpie wrote: | So K, if baseball is so easy to analyse then please briefly explain the contribution fielding makes towards wins ...
|
Again, there's much worthy of discussion here, but a brief continuaton first:
Nothing in life is easy. Just much, much easier. That's strictly a relative statement. I'm kind of thinking of anything substantially easier as 'trivial', but 'trivial' wouldn't really mean 'so easy' either.
The OBP really does seem an obvious thing to at least consider, but I guess that's just "low-lying fruit". Why it took decades to take it seriously, I don't know. |
|
|
|
|
Pies2016
Joined: 12 Sep 2014
|
| Post subject: | |
|
Interesting thread.
FWIW, I was talking to John Deeble a while ago ( Australia's mr baseball and former overseas talent scout for Boston Red Sox ) who was saying that a company had developed a software package for AFL clubs based on sabernetrics. The package was AFL modified and designed to form an outcome as to the game winning differential that an established player would make if he was traded into your club. I was fascinated by it ( also as a former baseball player ) but Deeble wouldn't go into much detail, presumably because of a vested interest.
No one I know in footy knew much about its application but over time I heard that three clubs in particular had paid to use the program.
Supposedly ( and all this is only hearsay ) north Melbourne in the recruitment of Dal Santo and he came with a rating of just under + 2 wins per season. Also GCS with Ablett at a rating of just over + 3 wins per season.
I also heard the Hawks used it but it actually deterred them from pursuing a player they were interested in.
Any club at an elite level would be silly not to at least remain open minded and consider this programs effectiveness. Having said that, the proof is in pudding and this recruitment tool like most, will live and die by its effectiveness and accuracy. |
|
|
|
|
K
Joined: 09 Sep 2011
|
| Post subject: | |
|
| Pies2016 wrote: | ...
FWIW, I was talking to John Deeble a while ago ( Australia's mr baseball and former overseas talent scout for Boston Red Sox ) ...
No one I know in footy knew much about its application but over time I heard that three clubs in particular had paid to use the program.
Supposedly ( and all this is only hearsay ) north Melbourne in the recruitment of Dal Santo and he came with a rating of just under + 2 wins per season. Also GCS with Ablett at a rating of just over + 3 wins per season.
I also heard the Hawks used it but it actually deterred them from pursuing a player they were interested in.
... |
Yes, P16, I was actually going to mention Deeble in a future post (given we're starting to talk about comparisons with other sports, incl. MLB). Do you know him, or did you just bump into him at some function?
I know about it only because there was a newspaper article about it. In the article, they were very cagey about how many (if any!) clubs were using their software. The opinion in the AFL data community seemed to be that it was probably snake oil. [Clarification: this opinion was not expressed in the newspaper article; it was expressed elsewhere.]
There is much to discuss just about this one point... |
|
|
|
|
Pies2016
Joined: 12 Sep 2014
|
| Post subject: | |
|
Re K ^^^
Deeble isn't a friend as such but we share some common links, so we tend to gravitate towards each other at various baseball functions and reunions. It's not that often these days because he still spends so much of his time in America with his baseball connections.
I think the biggest difference between baseball and AFL statistically, is that one has a far more greater reliance on teammates than the other.
I still like a good statistic as long as everyone 1 ) completely understands the parameters 2) has a big enough sample and is 3) totally focussed and relevant to the topic.
In recent years CD have correctly identified that Sidebottom having a kicking efficiency of 70% ( example only ) means absolutely nothing.
His differential compared to other players with specific nominated parameters is much more relevant.
By example, firstly we take everyone in the league who is a bona fide outside mid ( caters for fitness capabilities and higher uncontested vs contested ball ratio ) Then we add a game spread between 100 and 200 ( caters for experience elements ) You would also pick age parameters ( possibly 23 - 28 grouping )and then you can compare his efficiency agsinst all others in that overall grouping.
This will show him as a differential to all others in that grouping. Usually the top 10% of that group would be considered as elite and then from there you classify the remaining players as they appear in order of merit. Elite, good, above average, average, poor etc.
it's not everything but it hones in more on a players true value particularly when compared against his peers.
Not sure we will ever be able to put a figure on the intangible value of leadership qualities, smarts and half percenters ( even less than a 1%er )
I'm sure someone will try though. |
|
|
|
|
|
|
|
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You cannot attach files in this forum
You can download files in this forum
|
|
The Rules
FAQ
Usergroups
Calendar
Search
Log in
Register
