HFBoards

Go Back   HFBoards > NHL Eastern Conference > Atlantic Division > Montreal Canadiens
Mobile Hockey's Future Become a Sponsor Site Rules Support Forum vBookie Page 2
Notices

Sabremetrics/Microstats in Hockey

Reply
 
Thread Tools
Old
03-30-2012, 11:38 AM
  #1
Lafleurs Guy
Registered User
 
Lafleurs Guy's Avatar
 
Join Date: Jul 2007
Posts: 21,620
vCash: 500
Sabremetrics/Microstats in Hockey

This topic has come up in a few different threads over the year and seems to take other threads off topic so I figured I'd make a dedicated one here. In light of the fact that we are in the search for a new GM, just figured I'd see what people think about the Billy Beane way of building a hockey club. Do you think it's feasible? Are things like size and toughness irrelevant when evaluating hockey talent or are they attributes that should be valued and coveted by hockey minds? What path should the Canadiens take here? How much weight should be given to this as we go forward?

What do you think?

Edit: Talks to Goalposts has been kind enough to provide some simple explanations on some of the stats below. Have a look at his Glossary here:
Quote:
Originally Posted by Talks to Goalposts View Post
Shot attempts: Shots + Shots attempted blocked + Shot attempted missed.

Corsi: Shot attempts for versus shot attempts against for a particular game situation (i.e. 5 on 5, 5 on 4, 4 on 4 etc). Usually expressed on a per 60 minutes of playing time basis or as a percentage with 50% being even. Originally devised as a measure of the amount of work a goaltender would have to per game. It was discovered that it correlated extremely strongly with amount of time spent in each zone when the NHL tracked that information and that on a team basis it was the strongest predictor of future wins for a team (better than goal differential which was better than current record).

Because the time of possesion and zone time stats aren't tracked its seen as a proxy for puck possession and territorial control. One of its chief advantages is that it tracks so many events so its easy to get a large sample size to work with. The other is that it is one of the most stable aspects of the game (which is probably related to the large sample size). Corsi for a team/player is very good at predicting future Corsi while things like shooting/save% don't predict well. As such 5 on 5 corsi is a very good predictor of future 5 on 5 goals +/-.

Fenwick: Corsi's brother with blocked shots ignored. This gives somewhat better information at the cost of fewer events. Removing blocked shots probably allows it to better ***** defensive/offensive zone play because its has best correlation to "scoring chance" rates.

Scoring Chances: There is some quibiling over what the exact definition of a scoring chance should be but the basic definition is an unblocked shot attempt from within the "homeplate area" consisting of the defensive zone contained by a line from the goalpost to the faceoff dot with another to the ringette line using the ringette line as the top of the area. Some allow for well screened point shots as chances and some do not. Non-official stat tracked by people other than the NHL. Scoring chance data are usually seen as the best for micro-stat analysis (and most often what NHL coaches themselves seem track from information publicly availible) but it isn't publicly available for all teams currently. Oliver of http://enattendantlesnordiques.blogspot.ca/ tracks chances for Montreal.

Scoring chance differentials typically track very well with Fenwick/Corsi and track well with future situation goal differentials.

Score tied Corsi/Fenwick: One of the discoveries on the behavior of NHL teams is the effect of the score on how shot differentials. Teams with the league tend to be outshot, teams behind tend to outshoot. This is the answer the the paradox that teams that outshoot in general tend to win the most games but who had more or less shots in a particular game doesn't line up well with who won. Score tied Corsi/Fenwick is generally the metric most used to describe a team's 5 on 5 ability. More recently people have started to instead use a score situation adjusted shot rate which is probably a slight improvement.

Team shooting percentage: Basically team goals divided by team shots. A team being higher or lower than league average in this is an excellent sign that their are going to be better/worse in the future because there is very little evidence that a team can control their 5 on 5 shooting percentage in the long run.

On a team level its very rare to beat out your score tied 5 on 5 Corsi or do worse than it. Those that have mostly accomplish this by superior goaltending (save percentage) although Vancouver has done better offensively than their shot rates by virtue of their powerplay conversion rate. Similarly Columbus has done worse than average on the powerplay resulting in worse goals for their shots. 5 on 5 though, it seems that in the modern NHL every team comes back to the mean.

The caveat modern NHL is crucial because teams like the 80's Oilers produced their offense on percentages rather than shot rates. But there doesn't seem to be any team that can manage it these days.

Personal shooting percentage: Percentage of shots taken by a player that become goals. Extremely volatile but converges on a player's career ability. This is extremely volatile but on the long term will converge with a player's career talent. From how it progresses over a career in the NHL and the degree it translates from the AHL to the NHL (almost entirely), in my opinion its less of a talent as it is a tendency which reflects a style of play.

Team on ice shooting percentage: The percentage of shots by his team while a player is on that become goals. Much like personal shooting percentage is heavily volatile but converges on a mean, in this case usually league average. Almost all players tend to be in a very narrow band of 7-8.5% team on-ice shooting with changes year to year being fairly random. Their seems to be a very small cadre of players that do better to average year to year. Usually these are player that play consistently on very good lines such as Sedin-Sedin and Stamkos-St. Louis. There are so few of them its debatable if the ability to be significantly better than average in this regard is real or just the mark of their being hundreds of players and a few are going to be better than average over multiple years by shear population dynamics.
More here: I have bolded where I could as per above.
Quote:
Originally Posted by Talks to Goalposts View Post
Difficulty of minutes measurements:

In my opinion the more important way of analyzing the modern NHL game isn't in whether you accept shot based metrics or not (goal based ones generally get you close enough on that front anyway, at least if you accept goaltending as a factor which is uncontroversial) its acknowledging the major role that what type of minutes a player plays is a major factor in both how he looks to the eye and what kind of results they put up. NHL most NHL coaches manage their bench for particular territorial or line matchup roles and if there are a few that don't they'll get matched by the opposition resulting in largely the same effect.

Doing so may not have the direct intuitive effect you'd expect, i.e. starting in the offensive zone more might not directly raise a player's offensive stats or playing tougher minutes might not result in more goals against since such things will also depend on how a player responds to different situations but the effect will be strong on a player's differentials.

Two flavours of quality of minute situations are in current use. Quality of Competion (strength of opponent) and zone starts (location of starting shifts). An obvious third catagory would be strength of teammates but no one has really found a reliable measurement for that, its extremely hard to disentangle the strength of a player's linemates versus his own contribution. For that kind of thing I'd suggest looking at most common linemates by ice time and forming your own conclusion although you can do a more labourious "With You Without You" analysis (WOWY).

Quality of competition is measured two ways currently in the mainstream. Opposition +/- (QUALCOMP) and opposition relative Corsi (relCorsi QoC). The methodology of both is a bit weak so the straight numbers aren't always reliable comparing between teams, largely because each team has a different schedule but they work well within a team. Generally relCorsi QoC is the more reliable one when the two disagree for basically the same reasons Corsi is more reliable than +/-. For following a particular team though I recomend also looking for yourself what matchup are used game per game. For Montreal, Oliver's scoring chance reports for each game will also have matchup information while Habs Eyes on the Prize game threads will have a like to timeoneice.com script for a breakdown on minutes played per opponent each game. From tracking matchups the past couple years I can tell you that relCorsi QoC corresponds very strongly with who gets what kind of opponent but your welcome to test it out on your own. I didn't believe they worked until I did that myself.

Measuring where a player starts his shifts is generally done by way of Ozone % which is simply the porportion of non-neutral zone faceoffs (neutral zone faceoffs are considered neutral events for this) taken in the offensive zone. Unlike quality of competition which basically every team will have a spread on, whether to control your player's starting position is a decision made by a particular coach so not all teams will have a meaningful difference in zone starts. Vancouver in particular embraces this method, with Malhotra basically taking only defensive draws while Sedin takes tons of offensive zone ones. Chicago is a slightly less extreme adherent with the much under-rated David Bolland taking the heavy defensive duties while guys like Kane and Toews generally get a favourable starting point.

Montreal under Martin seemed to favour this strategy to start the season (a departure from is normal modus opperendi of pursuing straight power versus power matchups). With Desharnais getting in the range of a 60% Ozone while Plekanec (and usually Gionta) was around 42-45%. This was pretty wise considering that early season Desharnais produced lots of offense when it the offensive zone but had a weak possession/territorial game and the bad tendency to get pinned in his own zone when starting their facing a good line. One time when this broke down due to injury in a game against Pittsburgh, Desharnais basically got pinned in his own zone the entire game and ended up with something like a -20 even strength shot differential for the game. This kind of set up became less pronounced as the season continued but fortunately Desharnais developed the ability to face better competition and was paired with a very good two-way winger in Cole to compensate.

Subban also tends to be used a lot in the defensive zone, pretty much since he took over the first pairing last season when he moved up to play with Gill. That he generates team leading results in 5 on 5 shots and goal differential with unfavourable zone starts and very high quality of competition measurements is another reason for considering him to be an exceptional player.




For example of how this kind of analysis can be helpful, look at how a few of us were arguing at the beginning of the season that Cole was an awesome sigining while Leino was terrible. It was based largely on the principle that while they had similar surface stats, Cole was getting it done in a difficult situation while Leino had probably the easiest job in the league. Buffalo couldn't put Leino in a similar situation as he had in Philadelphia and he sunk. Cole ended up playing in an easier situation in Montreal and flourished.

Meanwhile the major drop in both Cammalleri and Plekanec's offensive totals since 2009-10 co-incides with when they got moved to playing a heavy shutdown role to start last season. Previously both tended to play about 2nd line competition, albeit with a major defensive zone faceoff responsibility, and both were on pace for a very good ~45 even strength points the previous season in that role. For reference, 40+ ES points is pretty much only achieved these days by top scoring line players. 45 points was good for 33rd in the league that season, 40 points 55th.


Last edited by Lafleurs Guy: 04-05-2012 at 04:45 PM.
Lafleurs Guy is online now   Reply With Quote
Old
03-30-2012, 12:54 PM
  #2
Et le But
Moderator
 
Et le But's Avatar
 
Join Date: Nov 2010
Location: New York
Country: Argentina
Posts: 17,779
vCash: 500
Advanced stats in hockey are somewhere in between baseball (where you can basically build a team with spreadsheets due to the nature of the sport) and soccer (where there's far too many scenarios to properly quantify all but the basics) in usefulness.

I've only just started trying to learn about these stats as I'm far from a numbers person, but I have to say I'm impressed by how much the line up with what I'm seeing. I would love to hear that there's management and scouts working for the Habs organization that use advanced statistics.

That being said they will never replace the more traditional forms of analysis. They are, and should be a compliment. Things like intangibles including the holy "toughness" factor can't be measured, one reason I've been skeptical about stat-centric sports evaluations is they ignore the very real psychological factor. But at the same time it's easy to get carried away by an unrealistic good run (see the Leafs this year) or a bad run (see the Habs this year before Martin was sacked), so stats are a great way to keep things grounded.

Had Gauthier looked at the microstats earlier in this season, he might have avoided the Kaberle trade and waited until the end of the season to change coaches.

Et le But is offline   Reply With Quote
Old
03-30-2012, 12:59 PM
  #3
PricePkPatch
Registered User
 
PricePkPatch's Avatar
 
Join Date: Mar 2008
Location: Montreal
Country: Canada
Posts: 16,547
vCash: 500
There sure some highly subjective stats that I'd like to see on paper:

- # of successful zone entry while not on a turnover
- # of zone loss prevented, both on the PP and ES
- # or % of shot rebound captured to carry, pass or reshoot
- % of shots that resulted in a manageable rebound for your teammate

All these stats would be more about preserving the flow of the game, and keeping the initiative in the play. Something we sorely sucked at this season

PricePkPatch is offline   Reply With Quote
Old
03-30-2012, 01:06 PM
  #4
Slew Foots
Haters gonna hate
 
Slew Foots's Avatar
 
Join Date: Sep 2006
Country: Canada
Posts: 770
vCash: 500
I agree that one cannot rely on such tools as much in hockey as one could in baseball. Baseball is essentially a series of one-on-one battles, making reliance on mathematical modeling as an abstraction of reality much more accurate than in sports like hockey and soccer, where there is a greater amount of external factors and random variability.

That being said, I believe sabremetrics/microstats can apply to hockey very well, as long as organizations are cognizant of the limitations of such mathematical modeling. Models in themselves are good, but it is the application of those models by people that can be dangerous. The Financial industry is an example of where people have overly relied on these models to make their decisions, with dire consequences such as the recent financial crisis. With respect to hockey, organizations cannot use these models as a single source of truth, but rather as a tool or one input into a holistic decision-making process. Just like experience (e.g. physicality and intimidation being important in the game) and gut feeling should also be an input among many in the decision-making process, but not the sole input.

Slew Foots is offline   Reply With Quote
Old
03-30-2012, 01:12 PM
  #5
Fozz
Registered User
 
Fozz's Avatar
 
Join Date: Aug 2002
Posts: 6,401
vCash: 500
No, stats models do not work well enough in hockey. Baseball is an individual sport,as almost every single action made is by one player at a time and is unaffected by the other players around him. Therefore, it's possible to form an opinion on the value of a player by stats alone.

It's the complete opposite in hockey, where almost every single action is influenced by what other players do and where systems have a great impact on the way the player plays. Too many relevant and important actions in hockey are not quantifiable and are simply impossible to be supported by stats. When I hear people stating that a player's ability to read the play position himself well is captured by his total of takeaways, I just want to shoot myself.

Fozz is offline   Reply With Quote
Old
03-30-2012, 01:22 PM
  #6
MathMan
Registered User
 
Join Date: Jan 2006
Posts: 17,025
vCash: 500
Given that they are employed unevenly across the league, I think analytics represent a fantastic competitive advantadge that could be achieved for what a NHL club like the Habs would consider pocket change, and I think the next GM would be an idiot not to invest in it.

And yes, hockey analytics do give us extremely valuable information. Notably, analytics are especially useful in helping us distinguish durable performance from streaks; a vital skill for GMs, in a field where overpaying for short-term past performance is a commonly-made mistake.

They are very different than in baseball precisely because the game is flowing and context is so important, but that does not make hockey unmeasurable. Really, the biggest difference between those who say hockey can't be analyzed and whose who used to say that baseball can't be analyzed thirty years ago is that the latter didn't have a sport X to point to to say "baseball is not like X".

There's a lot of research on analytics over the net, and I can't recommend enough that you look over and see. Some places I like:
arcticicehockey.com (formerly behindthenet.com; Gabriel Deshardins is a venerable, terrifying authority on the subject and has done consulting work for teams)
objectivenhl.blogspot.com (excellent but sadly hasn't been updated in a while)
enattendantlesnordiques.blogspot.com (Habs-centric, in French, and the Habs' chances counter; I can't recommend this blog enough).
habseyesontheprize.com (Habs-centric and the source of some excellent analysis)
coppernblue.com (Oilers blog; many of the analytics information out there originated in the Oilers' robust blogosphere)

behindthenet.ca (mostly charts, but it gives you an idea of the amount of information collected)

There's other sites out there but these are a pretty good cross-section with a Habs slant.


Last edited by MathMan: 03-30-2012 at 01:28 PM.
MathMan is offline   Reply With Quote
Old
03-30-2012, 01:24 PM
  #7
PricePkPatch
Registered User
 
PricePkPatch's Avatar
 
Join Date: Mar 2008
Location: Montreal
Country: Canada
Posts: 16,547
vCash: 500
If a player consistently manage to get the zone when the opposing team had the time to set themselves up defensively, I think it says a lot of the player's speed, play smarts or skill.

Which is why I want to calculate the end results rather than mere factual play. I know the circumstances will change at every play, but a player who is more skilled at taking the zone will still end up with higher average than his teammate.

Another example would be how long other team's player are allowed in front of the crease when we play in our zone. It doesn't matter if he gets rids of them by force, smarts, speed or skills. What matters is how effective he is at keeping them away.

And even if a player can't make another player to get out of the crease, can he at least nullify his presence? How often does he disrupt their pass receipt, their rebound-takeover?

These are situational stats but still show how good players are at certain roles. You could potentially make a team of players who have overall amazing puck retention skills and gameflow control, but have subpar scoring talent. Im curious to see what it'd give.

PricePkPatch is offline   Reply With Quote
Old
03-30-2012, 01:49 PM
  #8
Cyclones Rock
Registered User
 
Join Date: Jun 2008
Posts: 1,625
vCash: 500
Another tool in the arsenal for a GM. I'm sure that most GMs and socuts in the NHL incorporate certain of these micro stats and dismiss others. A firm grasp over what exactly goes into manufacturing each stat would be essential for proper use of any microstat.

Certain posters on this site (not all) will point to some micro stat and claim that it makes one player superior to another. Some number generated by some statistician magically makes player X better than Y-despite player Y being superior by traditional statistics. "Why X has a 6.3 on his XYZ and Y only has a 5.2 on his XYZ so X is obviously better than Y!!", such a poster will type. Usually, when asked, such a poster won't be able to explain the XYZ metric at any level, let alone its possible shortcomings. If a GM uses microstats in a similar manner, watch out

I find it silly that many microstatters (is that a new term?) dismiss a player's points scored with confidence bordering upon haughtiness. A dramatic example which I've seen presented repeatedly this board, has been by those whom have argued that Gomez has been a more effective player based on a Corsi (a micro stat) than many top point producers on the Habs. That argument does seem to have vanished of late. The point is, that a GM who uses micro stats is destined to failure if he elevates them to a point where observation and traditional measures are reduced to the point of irrelevance.

In general, I think the value of micro stats in hockey is far less than baseball. Many individual actions in baseball can be analyzed such that the performance of other players don't impact those individual actions. Baseball is unique in that respect. No other team sport allows for as much isolation on various individual performance characteristics. In hockey, up to 11 other individuals can impact upon one player's performance, thus making the measurement and evaluation of many statistics all the more difficult. So, at the end of the day, I can't see a team being built along a Billy Beane micro model. That, once again, doesn't mean that microstats are useless; just limited in their value. Also, most players with good micro stats probably have good traditional statistics. So, it's possible that the proper use of micro stats from a GMs perspective would be to confirm traditional stats, and if confirmation isn't there on an individual players, re-evaluate each in the context of the non-confirming microstat.


Last edited by Cyclones Rock: 03-30-2012 at 02:04 PM.
Cyclones Rock is offline   Reply With Quote
Old
03-30-2012, 01:52 PM
  #9
Kjell Dahlin
Registered User
 
Kjell Dahlin's Avatar
 
Join Date: Jan 2010
Location: Québec, Québec
Posts: 1,998
vCash: 500
I was happily surprised by Moneyball ; http://www.imdb.com/title/tt1210166/ ; anyone interested by the managerial aspect of pro sports (or in general) and the use of microstats, should watch it.

And yes I think it is feasible from a hockey perspective. Of course, and it is the case with all the data modeling, ground checks (luckily for us, Timmins is one of our guys!) should be realised (there are plenty of people who over trusted their environmental models in my line of work) but I think it can be applied to hockey.

Mathman provided an impressive list (http://hfboards.hockeysfuture.com/sh...65&postcount=6 - thanks!) but it is possible to create a very basic and simplistic (VERY simplistic!) model with two parameters imo: (Giveaways and takeaways differential) + (+/- of a player adjusted with the team +/- and the player's TOI).

When I really don't know a player, that's the first thing I look, and generally, it gives me an ok idea about the value of the player.



Quote:
Originally Posted by Et le But View Post
Advanced stats in hockey are somewhere in between baseball (where you can basically build a team with spreadsheets due to the nature of the sport) and soccer (where there's far too many scenarios to properly quantify all but the basics) in usefulness.

I've only just started trying to learn about these stats as I'm far from a numbers person, but I have to say I'm impressed by how much the line up with what I'm seeing. I would love to hear that there's management and scouts working for the Habs organization that use advanced statistics.

That being said they will never replace the more traditional forms of analysis. They are, and should be a compliment. Things like intangibles including the holy "toughness" factor can't be measured, one reason I've been skeptical about stat-centric sports evaluations is they ignore the very real psychological factor. But at the same time it's easy to get carried away by an unrealistic good run (see the Leafs this year) or a bad run (see the Habs this year before Martin was sacked), so stats are a great way to keep things grounded.

Had Gauthier looked at the microstats earlier in this season, he might have avoided the Kaberle trade and waited until the end of the season to change coaches.
"... I would love to hear that there's management and scouts working for the Habs organization that use advanced statistics..." <<< same here.



Edit:

I agree with Cyclones Rock - Especially: "... On this board, there have been those who have argued that Gomez has been a more effective player based on a Corsi than many top point producers on the Habs. That argument does seem to have vanished of late The point is, that a GM who uses micro stats is destined to failure if he elevates them to a point where observation and traditional measures are reduced to the point of irrelevance..."


Last edited by Kjell Dahlin: 03-30-2012 at 02:00 PM.
Kjell Dahlin is offline   Reply With Quote
Old
03-30-2012, 02:01 PM
  #10
Mathletic
Registered User
 
Mathletic's Avatar
 
Join Date: Feb 2002
Location: St-Augustin, Québec
Country: Canada
Posts: 11,961
vCash: 500
btw, basketball has pretty much taken over for baseball in statistical analysis. At the recent MIT conference on sports analytics most papers were on basketball and there's tons of other good papers published every month. So, when you work at it, it's quite evident you can do statistical analysis in flowing sports team. That said, hockey is different but I'm finding it "easier" to analyse even than basketball.

Lots of good posts here but like MathMan said, statistical analysis being done now by NHL teams is so basic. Goals, Assists, +/- and other basic stats that don't explain much. A team could gain a huge competitve advantage if they gave analytics a shot.

Analytics at the moment will at least help you optimize your lineup, matchups, project future players performances, draft and so on.

Also, in any field where luck plays a large role. Hockey being one, since either game results, drafting and so on involve luck. Decision-making analytics tell you that you should have as many independant sources as possible. So a stats analyst would only help and be involved in the process.

Mathletic is offline   Reply With Quote
Old
03-30-2012, 02:26 PM
  #11
Lafleurs Guy
Registered User
 
Lafleurs Guy's Avatar
 
Join Date: Jul 2007
Posts: 21,620
vCash: 500
Quote:
Originally Posted by Mathletic View Post
btw, basketball has pretty much taken over for baseball in statistical analysis. At the recent MIT conference on sports analytics most papers were on basketball and there's tons of other good papers published every month. So, when you work at it, it's quite evident you can do statistical analysis in flowing sports team. That said, hockey is different but I'm finding it "easier" to analyse even than basketball.

Lots of good posts here but like MathMan said, statistical analysis being done now by NHL teams is so basic. Goals, Assists, +/- and other basic stats that don't explain much. A team could gain a huge competitve advantage if they gave analytics a shot.

Analytics at the moment will at least help you optimize your lineup, matchups, project future players performances, draft and so on.

Also, in any field where luck plays a large role. Hockey being one, since either game results, drafting and so on involve luck. Decision-making analytics tell you that you should have as many independant sources as possible. So a stats analyst would only help and be involved in the process.
I don't think anyone doubts that analytics can help or be enlightening in some way. I guess the question is... how much weight should be given to them?

In baseball (which I think most would agree lends itself very well to sabremetrics) the Billy Beane example is very interesting because he decided to forgoe 'intangibles' altogether and it was an unmitigated disaster. He didn't look at desire to win, behaviour or soft skills at all. His team absolutely sucked. Then he ripped it apart, got rid of the head cases and in the very same season ran up a 20 game win streak. It's obvious that it can serve as a valuable tool for baseball and we've seen teams using it to great success.

Still haven't seen a method in hockey that captures a smashing hit into the boards though. Even basketball (another fluid game) does not have this factor. How do you distinguish between a hit by Chara vs. one by Deharnais? That's the example I keep coming back to because I think it bears bringing up.

When I hear people dismiss size and grit altogether... I become extremely skeptical of whatever comes out of their mouth next. Size and grit ARE attributes that GMs simply can't ignore. And I'm not sure how you capture this via analytics.

Lafleurs Guy is online now   Reply With Quote
Old
03-30-2012, 02:34 PM
  #12
Roke
Registered User
 
Roke's Avatar
 
Join Date: Jul 2003
Location: Winnipeg
Country: Canada
Posts: 1,889
vCash: 500
Quote:
Originally Posted by Kjell Dahlin View Post
I was happily surprised by Moneyball ; http://www.imdb.com/title/tt1210166/ ; anyone interested by the managerial aspect of pro sports (or in general) and the use of microstats, should watch it.
I haven't seen the movie, but the book is a very good read. I actually got less out of the Billy Beane/Oakland A's parts of the book than the bits on Bill James. This quote from one of Bill James' abstracts still stands out for me:

Quote:
So if we can't tell who the good fielders are accurately from the record books, and we can't tell accurately from watching, how can we tell? By counting things
Those are the two most powerful lines in the book for me.

Quote:
btw, basketball has pretty much taken over for baseball in statistical analysis. At the recent MIT conference on sports analytics most papers were on basketball and there's tons of other good papers published every month. So, when you work at it, it's quite evident you can do statistical analysis in flowing sports team. That said, hockey is different but I'm finding it "easier" to analyse even than basketball.
I don't follow basketball but I understand the analysis is fairly in depth (and the MIT conference was created, at least in part, by the Houston Rockets GM). MathMan mentioned Gabe Desjardins earlier and some of the early hockey analysis he did was informed by the basketball stuff a few years ago. Desjardin's also done consulting work on football arbitration and with hockey clubs and he's venturing deeper into soccer analysis. Basketball's advantage over hockey is that you have pretty distinct possessions.

The advantage baseball has over the other sports (especially soccer) is that most of the analytical developments happened outside the front offices so it's not proprietary and available to everyone.

Hockey Analytics are no replacement for scouting (as the Toronto Blue Jays discovered under Riccardi... though there were other decisions made by him that hamstrung the team) but at least I think they're an important complimentary tool. At the least you can get context for what you're seeing on the ice and on the basic stat sheet such as quality of opposition or where a player starts their shifts.

If a team was to say, hire a bunch of University students to watch games and count "things" (I'd go with scoring chances in every NHL game) you could have a pretty good advantage for evaluating players outside your organization and only you would have the data.

We don't really know what teams track but San Jose's assistant GM Joe Will uses some proprietary software of some sort, at the Sloan Conference Peter Chiarelli mentioned at the Sloan Conference that the Bruins use some form of analytics, and I've heard 3rd-hand the Islanders consulted with some stat-minded types when they signed Matt Moulson out of the AHL (well, the Kings organization) and when they took a look at the top prospects in the 2009 draft.

There's no salary cap on front office operations. For less than what Georges Laraque cost the Habs they could probably put together an elite stats department to compliment the rest of the organization. As with scouting they'd get things wrong, but having another perspective within the organization probably wouldn't be a bad thing


Last edited by Roke: 03-30-2012 at 02:41 PM.
Roke is offline   Reply With Quote
Old
03-30-2012, 02:38 PM
  #13
Kjell Dahlin
Registered User
 
Kjell Dahlin's Avatar
 
Join Date: Jan 2010
Location: Québec, Québec
Posts: 1,998
vCash: 500
Quote:
Originally Posted by Lafleurs Guy View Post
I don't think anyone doubts that analytics can help or be enlightening in some way. I guess the question is... how much weight should be given to them?

In baseball (which I think most would agree lends itself very well to sabremetrics) the Billy Beane example is very interesting because he decided to forgoe 'intangibles' altogether and it was an unmitigated disaster. He didn't look at desire to win, behaviour or soft skills at all. His team absolutely sucked. Then he ripped it apart, got rid of the head cases and in the very same season ran up a 20 game win streak. It's obvious that it can serve as a valuable tool for baseball and we've seen teams using it to great success.

Still haven't seen a method in hockey that captures a smashing hit into the boards though. Even basketball (another fluid game) does not have this factor. How do you distinguish between a hit by Chara vs. one by Deharnais? That's the example I keep coming back to because I think it bears bringing up.

When I hear people dismiss size and grit altogether... I become extremely skeptical of whatever comes out of their mouth next. Size and grit ARE attributes that GMs simply can't ignore. And I'm not sure how you capture this via analytics.
"... Size and grit ARE attributes that GMs simply can't ignore. And I'm not sure how you capture this via analytics."

Size... well... pounds or kilos should do the trick!

As for grit: fighting majors, roughing minors, hits and (Giveaways- Takeaways differential) are a good start imo.

Kjell Dahlin is offline   Reply With Quote
Old
03-30-2012, 02:43 PM
  #14
The Gal Pals
Breaking Hab
 
The Gal Pals's Avatar
 
Join Date: Oct 2006
Country: Canada
Posts: 1,829
vCash: 500
Quote:
Originally Posted by Fozz View Post
No, stats models do not work well enough in hockey. Baseball is an individual sport,as almost every single action made is by one player at a time and is unaffected by the other players around him. Therefore, it's possible to form an opinion on the value of a player by stats alone.

It's the complete opposite in hockey, where almost every single action is influenced by what other players do and where systems have a great impact on the way the player plays. Too many relevant and important actions in hockey are not quantifiable and are simply impossible to be supported by stats. When I hear people stating that a player's ability to read the play position himself well is captured by his total of takeaways, I just want to shoot myself.
Pretty much this. One player may play well with one linemate and be completely lost with another. There's too many intangibles. The Billy Beane model might work for goalies and for players in shootouts but other than that it's useless.

The Gal Pals is offline   Reply With Quote
Old
03-30-2012, 02:51 PM
  #15
Mathletic
Registered User
 
Mathletic's Avatar
 
Join Date: Feb 2002
Location: St-Augustin, Québec
Country: Canada
Posts: 11,961
vCash: 500
Quote:
Originally Posted by Lafleurs Guy View Post
I don't think anyone doubts that analytics can help or be enlightening in some way. I guess the question is... how much weight should be given to them?

In baseball (which I think most would agree lends itself very well to sabremetrics) the Billy Beane example is very interesting because he decided to forgoe 'intangibles' altogether and it was an unmitigated disaster. He didn't look at desire to win, behaviour or soft skills at all. His team absolutely sucked. Then he ripped it apart, got rid of the head cases and in the very same season ran up a 20 game win streak. It's obvious that it can serve as a valuable tool for baseball and we've seen teams using it to great success.

Still haven't seen a method in hockey that captures a smashing hit into the boards though. Even basketball (another fluid game) does not have this factor. How do you distinguish between a hit by Chara vs. one by Deharnais? That's the example I keep coming back to because I think it bears bringing up.

When I hear people dismiss size and grit altogether... I become extremely skeptical of whatever comes out of their mouth next. Size and grit ARE attributes that GMs simply can't ignore. And I'm not sure how you capture this via analytics.
There have been other advances in quantification of "intangibles". I remember a conference at the MIT Sloan Conference on a group of researchers who have analyzed speech of different players in order to predict future behaviour as well as another team measuring and assessing various leadership skills. For instance, players who touch teammates more than others like Garnett IIR ... I know it doesn't sound good but anyways ... I don't remember the whole thing but anyways, it's being analyzed.

Then again, you have to approach stats analysis in an open way. In general analysts will tell you that hits are vastly overrated unless you create turnovers out of it. Teams that hit a lot, don't have the puck a whole lot. Though 1 smahing hit may make an impact in injuring a player, in general it doesn't do a whole lot to help you win. It only serves the macho in us who wants "displays of power" if I may put it this way.

Mathletic is offline   Reply With Quote
Old
03-30-2012, 03:15 PM
  #16
Roulin
Registered User
 
Roulin's Avatar
 
Join Date: Mar 2007
Location: Montreal
Posts: 4,242
vCash: 500
Quote:
Originally Posted by airic000 View Post
One player may play well with one linemate and be completely lost with another. There's too many intangibles.
This is actually very much tangible/measurable. Results with specific teammates are available at the timeonice site. More general quality of teammate stats are available at behindthenet.

Roulin is offline   Reply With Quote
Old
03-30-2012, 03:17 PM
  #17
MathMan
Registered User
 
Join Date: Jan 2006
Posts: 17,025
vCash: 500
Quote:
Originally Posted by Kjell Dahlin View Post
but it is possible to create a very basic and simplistic (VERY simplistic!) model with two parameters imo: (Giveaways and takeaways differential) + (+/- of a player adjusted with the team +/- and the player's TOI).
If you don't mind me criticizing, all the stats used in your metric have problems.

Giveaways and Takeaways are not counted consistently across arenas; some places give as much as five times more than others. You can adjust for that, but the other problem is that giveaways are often accrued by players who handle the puck a lot and often try to pass it around -- the giveaway top-30 often ends up looking like a billboard of the top playmakers in the game. (In a similar fashion to Gorges' perceptive comment on how a lot of blocked shots means you're in your zone a lot.)

+/- has issues, foremost among them though the fact that shorthanded goals and empty netters count -- this penalizes players with offensive talent who are on the PP and in high-leverage scoring situations a lot, while favoring defensive players who play the PK and high-leverage defensive situations. If there's one thing micros have taught us, it's that context matters.

MathMan is offline   Reply With Quote
Old
03-30-2012, 03:19 PM
  #18
Goldthorpe
Meditating Guru
 
Join Date: Jan 2003
Location: Montreal
Posts: 4,367
vCash: 500
Quote:
Originally Posted by Lafleurs Guy View Post
Still haven't seen a method in hockey that captures a smashing hit into the boards though. Even basketball (another fluid game) does not have this factor. How do you distinguish between a hit by Chara vs. one by Deharnais? That's the example I keep coming back to because I think it bears bringing up.

When I hear people dismiss size and grit altogether... I become extremely skeptical of whatever comes out of their mouth next. Size and grit ARE attributes that GMs simply can't ignore. And I'm not sure how you capture this via analytics.
A big difference between the characteristics you mention (size, grits) and microstats, is that microstats measure results. If I show that player X is very good at generating shots on his line over his career with different linemates, it doesn't say anything about how player X does it. Maybe he is big and tough and muscle up his opponents, or maybe he's small and feisty and just skates past them. Different type of skills, but potentially similar microstats, because the results (in term of what is observed) is only what count. This is the best benefit of microstats, as being objective helps removing the typical and easy bias one creates when "just watching a game". Plenty of players are big and gritty but also suck ass, and you mean miss that fact if you don't look at the actual results.


Edited:

And to answer the OP question: microstats is just a different strategy to evaluating players and team. It won't become the ONLY strategy, but a lot of teams already use them in different manners. It is certainly there to stay and to grow as a field.

Goldthorpe is offline   Reply With Quote
Old
03-30-2012, 03:21 PM
  #19
MathMan
Registered User
 
Join Date: Jan 2006
Posts: 17,025
vCash: 500
Quote:
Originally Posted by Mathletic View Post
In general analysts will tell you that hits are vastly overrated unless you create turnovers out of it. Teams that hit a lot, don't have the puck a whole lot.
Hits, as a stat, are actually negatively correlated to winning, presumably for this very reason. I don't think that being willing to hit actually causes a team to lose more, but you only hit if you don't have the puck... and puck possession is how you win hockey games.

If we could get a decent giveaway stat, I wonder if we might not end up with the entirely counter-intuitive correlation that more giveaways correlate to more winning (because giveaways are a sign of puck possession).

MathMan is offline   Reply With Quote
Old
03-30-2012, 03:24 PM
  #20
MathMan
Registered User
 
Join Date: Jan 2006
Posts: 17,025
vCash: 500
Quote:
Originally Posted by Roulin View Post
This is actually very much tangible/measurable.
This actually brings to mind another point that, I must admit, occasionally bugs me.

If one can think of an objection to microstats within thirty seconds (what about shot quality? what about teammates? what about those guys who always play against the top lines?), that doesn't give leave to dismiss the whole thing. The odds are very good that the objection has already been thought of by analysts, examined, and either had a metric introduced or was dismissed as not making enough difference to matter on aggregate.

MathMan is offline   Reply With Quote
Old
03-30-2012, 03:31 PM
  #21
Et le But
Moderator
 
Et le But's Avatar
 
Join Date: Nov 2010
Location: New York
Country: Argentina
Posts: 17,779
vCash: 500
Quote:
Originally Posted by Mathletic View Post
btw, basketball has pretty much taken over for baseball in statistical analysis. At the recent MIT conference on sports analytics most papers were on basketball and there's tons of other good papers published every month. So, when you work at it, it's quite evident you can do statistical analysis in flowing sports team. That said, hockey is different but I'm finding it "easier" to analyse even than basketball.
I'm still not entirely convinced statistical analysis can work for soccer, concepts like shooting percentage and balls lost/recovered are just far less meaningful than they are in basketball or hockey. Hockey is closer to basketball here, there's far more stoppage and less players meaning more emphasis on individual variables. Perhaps someone will prove me wrong someday but I have yet to see proof that soccer can be quantified.

Et le But is offline   Reply With Quote
Old
03-30-2012, 03:32 PM
  #22
Goldthorpe
Meditating Guru
 
Join Date: Jan 2003
Location: Montreal
Posts: 4,367
vCash: 500
Quote:
Originally Posted by MathMan View Post
This actually brings to mind another point that, I must admit, occasionally bugs me.

If one can think of an objection to microstats within thirty seconds (what about shot quality? what about teammates? what about those guys who always play against the top lines?), that doesn't give leave to dismiss the whole thing. The odds are very good that the objection has already been thought of by analysts, examined, and either had a metric introduced or was dismissed as not making enough difference to matter on aggregate.
I agree that this is very annoying, yet a common occurrence in these discussions. People, it's not because you just thought of a zinger 3 minutes ago that it hasn't already been answered 3 years ago.

Also, the "you can make statistics say anything" argument.

Goldthorpe is offline   Reply With Quote
Old
03-30-2012, 03:33 PM
  #23
Poulet Kostopoulos
Registered User
 
Poulet Kostopoulos's Avatar
 
Join Date: Oct 2009
Posts: 4,843
vCash: 500
Since hockey is a free flowing sport, I'd imagine to analyze and quantify a given situation/play, we'd need to first divide the hockey rink into cells of a certain size. Then, obviously, look at the location, position, condition (e.g. tied up by another player), speed, type, size, reach, style (defensive-oriented, "goon", etc), time on ice of each player and the position of his stick at the time when the subject is about the make the play. We'd also take into account how the subject was positionned (e.g. facing the board with puck on the backhand). The state of the puck (bouncing? deflected?), the time on the clock, the score, the officials on the ice (their position and tendencies) would also be factors.

It's possible but it's much more complicated than baseball.


Last edited by Poulet Kostopoulos: 03-30-2012 at 03:38 PM.
Poulet Kostopoulos is offline   Reply With Quote
Old
03-30-2012, 03:39 PM
  #24
Mathletic
Registered User
 
Mathletic's Avatar
 
Join Date: Feb 2002
Location: St-Augustin, Québec
Country: Canada
Posts: 11,961
vCash: 500
Quote:
Originally Posted by MathMan View Post
Hits, as a stat, are actually negatively correlated to winning, presumably for this very reason. I don't think that being willing to hit actually causes a team to lose more, but you only hit if you don't have the puck... and puck possession is how you win hockey games.

If we could get a decent giveaway stat, I wonder if we might not end up with the entirely counter-intuitive correlation that more giveaways correlate to more winning (because giveaways are a sign of puck possession).
You're right, on average the raw stat of hits for is negatively correlated with winning. Inversely, being hit is actually positively correlated with winning. I'll re-phrase McDonald's joke at the MIT Conference, which basically went, imagine a coach telling his players to get hit and start missing the net more in order to win games ... since totals shots are a better indicator of future goal scoring ... during intermission. It makes no sense.

Like you said, giveaways and takeaways tell nothing about winning and losing because of the way they are measured by the NHL. Some obvious giveaways aren't counted in for the home team whereas the road won't get away with much.

Some people like Johnson I believe is his name at hockeyanalysis started adjusting for those stats. I did the same with my own data and you get much better results and you get to the point where it actually a part of winning and losing games.

I'm of the opinion at the moment that hits are relevant as long as they create turnovers. Some players are great at hit like Landeskog. Other hits that come 2 seconds after the puck is gone, I guess aren't as important.

Also, another thing I found odd at first was that teams that don't hit a lot or don't put pressure tend to create more turnovers. You will find the same in football. QBs who have lots of time to throw tend to throw more interceptions. Main reason for that would be that your opponent starts thinking and knows he has no options and starts making bad decisions.

Anyhow, the whole subject of giveaways and takeaways is very interesting. Like you said, playmakers tend to commit more giveaways. Same is observed in basketball. Point guards who handle the ball will commit more turnovers than your centerman who can't handle the ball at all in the first place. So, time of possession obviously plays a role into this.


Quote:
Originally Posted by MathMan View Post
This actually brings to mind another point that, I must admit, occasionally bugs me.

If one can think of an objection to microstats within thirty seconds (what about shot quality? what about teammates? what about those guys who always play against the top lines?), that doesn't give leave to dismiss the whole thing. The odds are very good that the objection has already been thought of by analysts, examined, and either had a metric introduced or was dismissed as not making enough difference to matter on aggregate.
great point indeed lol! can't believe the number of times I read this as objections to analytics


Quote:
Originally Posted by Et le But View Post
I'm still not entirely convinced statistical analysis can work for soccer, concepts like shooting percentage and balls lost/recovered are just far less meaningful than they are in basketball or hockey. Hockey is closer to basketball here, there's far more stoppage and less players meaning more emphasis on individual variables. Perhaps someone will prove me wrong someday but I have yet to see proof that soccer can be quantified.
honestly I haven't done my homework on soccer yet. I have a decent amount of articles I can't wait to read but haven't done so yet. However, from the small amount I have read I'm led to believe it can actually be done, but in general, the maths are more involved and complicated. For instance, the standard pythagorean formula won't work, or at the very least, doesn't give as good an indication of team performance than it does for hockey, baseball and so on. But I've seen a different formula which computes a pythag expectation, except there's more maths behind it. I'll see if I can get the link back.

there it is

http://www.soccermetrics.net/categor...agorean-tables

I can't find the link back on the actual paper that explains how the formula works, I have printed it but can't find the link

so if goals for and against can explain team winning%, then I'm led to believe there must be some sort of analytics that can be used


Last edited by Mathletic: 03-30-2012 at 03:48 PM.
Mathletic is offline   Reply With Quote
Old
03-30-2012, 03:42 PM
  #25
Goldthorpe
Meditating Guru
 
Join Date: Jan 2003
Location: Montreal
Posts: 4,367
vCash: 500
Quote:
Originally Posted by Poulet Kostopoulos View Post
Since hockey is a free flowing sport, I'd imagine to analyze and quantify a given situation/play, we'd need to first divide the hockey rink into cells of a certain size. Then, obviously, look at the location, position, condition (e.g. tied up by another player), speed, type, size, reach, style (defensive-oriented, "goon", etc) of each player and the position of his stick at the time when the subject is about the make the play. We'd also take into account how the subject was positionned (e.g. facing the board with puck on the backhand). The state of the puck (bouncing? deflected?), the time on the clock, the score, the officials on the ice (their position and tendencies) would also be factors.

It's possible but it's much more complicated than baseball.
The microstats (in hockey, this tend to be the term for "sabremetrics", I don't know why) that already exist and have been compiled in the last few years don't go that far, but they still go farther than the traditional stats we all know (goals, assists, hits, etc.). Some of them have been shown to be pertinent over a long period of time (like many years).

For anyone who is just starting on the topic, I highly suggest the links provided by Mathman above.

Goldthorpe is offline   Reply With Quote
Reply

Forum Jump


Bookmarks

Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off



All times are GMT -5. The time now is 10:38 AM.

monitoring_string = "e4251c93e2ba248d29da988d93bf5144"
Contact Us - HFBoards - Archive - Privacy Statement - Terms of Use - Advertise - Top - AdChoices

vBulletin Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
HFBoards.com is a property of CraveOnline Media, LLC, an Evolve Media, LLC company. ©2014 All Rights Reserved.