Responding to the Criticism II


I must have kicked the hornet’s nest with Friday’s piece, because there was quite the response both in the comments and on Twitter.

Upon reflection, I’m a bit embarrassed my stream-of-consciousness writing was sloppy...not that there were errors, but I could stand to flesh out my arguments explaining why, yes, I am correct.  Unfortunately I don’t get paid to write about hockey, and my actual job required some attention, so I had to rush through it.

I’m going to break down the two main points of contention in the original pieces: that injuries were not a legitimate excuse for the Wild’s failure and that the Minnesota Wild will be a below average team next year (and that can be proved with advanced statistics).

First, let me point out something I neglected to address in Friday’s article: I don’t expect the Wild to challenge for the Cup and made no boasts or guarantees about their success next season.  I firmly believe in intangibles such as chemistry and luck.  With hockey it’s incredibly difficult to predict the outcome of an 82-game season, let alone a best-of-seven playoff series.  Especially a year in advance.  So my expectations are that they will be much better offensively, somewhat better defensively, and likely grab a playoff spot (guessing 4-8 range).  This isn’t a team that is going all in for next year, but will be better over the next several with the growth of prospects.



Now, a lot of fun was had at my expense with the “boo hoo, the injuries” argument, via Ryan Lambert’s Twitter account (@twolinepass):

@twolinepass basically the gist of the post is "yes but injuries" and "the advanced stats prove nothing" and "lemaire doesn't coach here" which, lol.

I’m not sure that these people will ever be convinced, but it seems pretty clear to me, that if you lose four of your top six forwards for a significant amount of time, and you have nobody to step in and fill that role adequately, it’s going to devastate the team.

I’ve written a lot of words about this subject, and it never seems to get through to people.  Maybe a graphic will help.




I’ve plotted the team’s goal differential and win percentage (and a five-game moving average for a smoothed trend) through the season with injuries shown on top.  You can see within a couple weeks of each other, the Wild lost Devin Setoguchi to a knee injury, Guillaume Latendresse to a concussion, Pierre-Marc Bouchard to a concussion, and Mikko Koivu to a leg injury.  Lats and Butch both tried to play through the concussions but left for the season.  Koivu tried to play through his shoulder injury but was shut down again in February, then came back in late March to prepare for the World Championships (which Finland was defending on home soil).

You can see the team’s goal differential and win percentage both tanked in mid-December, which is when they fell from first in the league to about 12th-13th in the conference (and low-to-mid 20’s in the league).  They had a bit of success in early and late February, but were pretty much terrible until April, when they won a few games at the end of the season (and screwed up their draft position from about 3rd-4th to 7th).

Now is there a perfect correlation?  Of course not.  But you can see that, for much of the season, the Wild was missing three of its top six forwards, which had a trickle-down effect that I will cover later.

Here’s a comparison of the win percentage of the team with and without certain players (and the number of games they missed):


PlayerWin % WithWin % WithoutGames Missed
Mikko Koivu0.5640.37027
Pierre-Marc Bouchard0.5770.43043
Guillaume Latendresse0.5940.47766
Devin Setoguchi0.5070.46213


You can see these guys had a significant impact being in versus out of the lineup.  Koivu is the team’s leader, playing big minutes, on the power play, penalty kill, matching up against top defenders.  Bouchard is the team’s best playmaker, a big asset on the power play as well.  Latendresse was the team’s best goal scorer, one of the few skilled players who also played a physical game.  Setoguchi was one of the few skilled players with speed and scoring ability.

Who replaced these guys in the lineup?  Players like:

Kyle Brodziak, a solid third line center, who did a great job of filling in at the #1 spot for stretches, but clearly not a high skilled player.  He had some success playing with Dany Heatley with his aggressive forecheck and physical play.

Nick Johnson, who was picked up off waivers, a decent energy player who picked up some points playing with Heatley and Brodziak but wasn’t even qualified for a roster spot next year.

Cal Clutterbuck, a good energy and checking winger, but not a skilled player at all.  He has a good wrist shot but fairly poor hands and playmaking ability.

Essentially the Wild shifted some third and fourth liners up into the top six and brought in some journeyman AHL’ers to fill out the bottom six of the lineup.  This created a severe lack of skill and scoring up front, which is clearly evident in the goal differential graph.

Now Lambert mentioned Calgary and Pittsburgh as having suffered as many man games lost to injury as the Wild, yet having more success:

@twolinepass wild fans understand that calgary lost more man games to injury than minnesota last year and still finished nine points higher right?

@twolinepass another fun fact is pittsburgh lost the same number of man-games to injury as minnesota.

I added up the top six forwards (points per game) and top four defensemen (average time on ice) for the teams:

Calgary: 114
Pittsburgh: 161
Minnesota: 193

When you look at actual impact players being lost, Minnesota did suffer significantly more than the others.  Granted, the Wild didn’t lose Sidney Crosby for much of the year, but they also didn’t have Evgeni Malkin to step up.  It’s a little easier to bear losing the #1 player in the world when the #2 can step in.  Malkin’s no Kyle Brodziak...

But what about the highly touted prospects?  Couldn’t they step in?  No.  Mikael Granlund, Johan Larsson, and Jonas Brodin were playing in Europe.  Charlie Coyle, Zack Phillips, and Brett Bulmer were playing in the CHL.  Jason Zucker and Erik Haula were playing in college.  Literally all the Wild’s top prospects were not available to play this year, but all of them (except Haula) could be playing in the Wild organization next year.

So the Wild had injuries in the top six, no skilled players in the bottom six to step up, and no skilled prospects in the minor leagues that could be called up to the big team.  Adding Parise and Granlund to the top six, as well as multiple high end prospects to the minor league team, will provide some insurance against injuries next year.  Which is a major reason the Wild should be greatly improved.

-----

The other point I’d like to address in greater detail is the concept of Moneyball (or advanced statistics) as it applies to hockey.  Frankly, it does not work nearly as simply as some would lead us to believe.

Moneyball type statistical analysis works fairly well in baseball, but does not work with hockey.  The reason is that baseball outcomes are fairly easy to track.  There are hits, runs, errors, strikes, balls, outs.  Clearly defined.  In addition, competition is relatively static in baseball.  The batters face the same pitchers and fielders.  You can compare a #1 hitter with a #6 hitter easily, as they are the only offensive players on the field.

Hockey is a mushy and dynamic sport in contrast.  Likely the best indicator of skill and success, puck possession, is incredibly difficult to track and define.  Possession is measured indirectly through Corsi, but is muddled with changing defenders AND changing line mates.

With Desjardin’s article, he appears to make the assumption that the Wild are simply adding Parise and Suter to the lineup.  There’s no mention of the players who would be replaced in the lineup, who are journeyman AHL’ers (not even mediocre NHL’ers, or good AHL’ers, just mediocre AHL’ers).  Replacing a player who clearly struggles against NHL competition with a player who has proven to succeed in the NHL is a big deal, and will have bigger ramifications than just the player’s contribution superimposed on the team.

BUT, there’s another omission made by Desjardin: the trickle-down effect of making improvements at the top of the depth chart.

Parise and Suter won’t simply plug in where say Warren Peters and Justin Falk played.  They will be matched up against the top competition.  And unlike a player like Devin Setoguchi, who faced secondary competition with San Jose, they have already faced and succeeded against top defenders.  These players are much more of a “sure thing.”

And by inserting them at the top of the depth chart, other players and pushed down into roles they can be more successful.  In statistical terms, their quality of competition will decrease.  In layman’s terms, they will have an easier time playing hockey.  Skill is not an independent variable; it depends on the level of competition.

Using a personal example, when I play with my winter hockey team, I score a lot of goals.  We play at a lower level men’s league, and I’m able to skate and shoot better than my opponents.  I had 26 points in 20 games last year.  This summer, I’ve got one point in six games so far (wish me luck tonight).  My skill level is relative to the opposition.

Back to the NHL (quite a ways away from my league), players like Clutterbuck and Brodziak won’t be matched up against the Charas and Webers.  They will face secondary defenders, and have a relatively easier time playing hockey.  Their Corsi relative to the competition will increase because the competition is focused on other players.  And this trickles down throughout the lineup, improving the Relative Corsi of the bottom six or even bottom nine forwards.  That is why a high end player is so valuable; improving a bottom six player improves that position, but improving a top line player improves EVERYONE’S position.

The Wild adding Parise and Suter doesn’t merely improve the top line and the top pairing; it improves the entire lineup because everyone will face slightly easier competition.  Marco Scandella and Jared Spurgeon become the second pairing instead of the top pairing.  Their skill level didn’t decrease, but their competition will.

So until the advanced stats guys can tell me where they made all the substitutions and adjustments to account for all the changes in the lineup, I don’t see why we should take these claims seriously at all.

-----

As always, we enjoy your feedback, be it here in the comments, on Twitter (@FRBHockey), or on HFBoards.com where many of us are members.

9 comments:

  1. All you really need to know is that Minnesota was a bad Corsi team even when everyone was healthy and that Corsi correlates extremely well with winning hockey games. Would Minnesota have sunk as low as they did had they not sustained injuries? Of course not. But was their fast start a product of puck-luck rather than an actual reflection of the team's overall skill? Yeah, almost certainly.

    Parise and Suter will make the team better and I agree with your assertion that this is particularly true once their prospects are able to contribute in a meaningful way to the lineup. Not sure that this is Minnesota's year to even make the playoffs though, I'm afraid.

    ReplyDelete
  2. The Corsi figures you discount are a measure for individual player performance and the Player Usage Charts that display them as a graphic also take into account the quality of competition (QoC). When displayed in the player usage charts, these advanced statistics are very telling.

    The argument against injuries being the only factor for the decline of the Wild last season is not based on individual player Corsi numbers, as you seem to believe based on your arguments, but on team shots for/against. Even while the Wild were winning and completely healthy they were being outshot by a consistently significant margin.

    You claimed that "the best indicator of skill and success, puck possession..." This is not true. The best indicator of skill and success is 5 vs 5 shot differential. Teams can win temporarily while being significantly outshot, but only as an aberration that will regress.

    ReplyDelete
  3. The problem with the Corsi argument is largely begging the question.

    Corsi is not predictive. That's all that needs to be said on the point, but I'll go even further. Even if Corsi were predictive, the sample size for the portion of the season you're attempting to project over was completely worthless. The section of the season being selected for projection is biased and non-representative. The error rates as well as standard deviation for Corsi are very high. Put simply, any attempt to use these advanced statistics over the Wild's 2011-2012 season has absolutely no statistical merit, and is frankly a serious affront to any meaningful statistical analysis ever done.

    ReplyDelete
    Replies
    1. If you want to talk about predictive power, nothing we have is very predictive. But compared to goal ratio and winning percentage:

      "Corsi Tied is the best predictor of how a team will perform over the remainder of its schedule, regardless of the point in the schedule at which the calculation occurs."

      (http://objectivenhl.blogspot.com/2011/03/loose-ends-part-i-predictive-validity.html)

      So I don't think it's necessarily foolish to trust Corsi over winning% when the two metrics disagree.

      "Shot differential is flawed because it does not take into account shot quality or actual puck possession times."

      The NHL tracked zone time for a season or two about ten years ago. Shot differential correlated really strongly with zone time differential (http://vhockey.blogspot.com/2008/08/zone-time.html).

      As for shot quality, at a team level it's hard to sustain, and scoring chances had the Wild around 46% when the Wild were in first (which is about as good as Edmonton, and far worse than, say, San Jose, Pittsburgh, or Vancouver, or even Washington).

      Delete
  4. Shot differential is flawed because it does not take into account shot quality or actual puck possession times. A backdoor tap in is given equal weight as a dump-and-change from the red line that hits the net. That is why I say it is an indirect measure of success, and not a direct one.

    I do like Corsi as a measurement tool within a team, but across teams it becomes very problematic because of the difference in defensive schemes, and other errors. I see that you can isolate a lot of the effects in things like away 5v5 shot differential, but that becomes such a small sample size of the larger picture...

    But I'd agree the Wild were not as good as they appeared through December, but were not as bad as they appeared after December. Watching games though, it wasn't regression so much as it was a complete disaster of a roster.

    ReplyDelete
    Replies
    1. The easiest stats to judge a team by: Ws and Ls

      Delete
  5. As a preventative measure for the impending "regression to the mean" comments:

    The 2011-12 Wild did not regress to the mean. The term literally means that over a long enough period of time, the generalized average value will occur frequently enough that outliers will not have a significant impact on the overall mean. This is not what happened last season.

    What happened last season is that the Wild performed much better than should be expected based upon their statistical performance. After injuries struck, the Wild performed far worse than should be expected based upon their statistical performance. This is not a regression to the mean. In fact, it's about as opposite as you can get from regression to the mean. Statistically speaking, the Wild's second half performance last year was at least as unlikely as their first half performance. The team last year did not "prove" advanced statistics models. They actually provided strong argument against them twice, doing absolutely nothing to support them.

    ReplyDelete
  6. These six commenters are proving that your arguments are fantastic and much better thought out this time.

    I love the Win % chart, and I love your explanation that the Wild were missing depth last year. The Wild had a good team last year, add in some depth and they could have been a dangerous team. We've added depth (Konopka and Mitchell), we've added up front (Parise over Lattendress, Plus Granlund) and we've added to the back end (Suter). To me, that undoubtedly means good things.

    ReplyDelete
  7. We've added depth in Houston too, tho we don't know yet the exact names.

    To me, that is really the biggest thing, should significant injuries hit again this year. Real (if inexperienced) NHL prospects with real NHL-level skills will back-fill, not career AHL'ers hoping to maybe crack the 3rd/4th line or 3rd pair by their 26th birthday.

    Charlie Coyle and Zack Phillips, or Jeff Taffe and Warren Peters? Please. To ask the question is to know the answer.

    ReplyDelete