Statistics in Hockey; What is Their Place?

Malkin4Top6Wingerz

Can you like, shutup
Mar 14, 2009
5,032
9
Heck, take CORSI. If you're a team that runs a counter-attack system and does not mind getting outshot 35-20 every game, you won't stack up favorably with teams that really limit the shots against. But your team might be winning 60 percent of its games while their team wins 40 percent. Who is the better player? CORSI would probably say the player on the losing team. It's limited because it places a value on a stat that is not necessarily predictive.

Anybody who is familliar with evaluating advanced statistics in hockey wouldn't put that much weight into raw Corsi numbers, as they are often very team dependant and can be greatly impacted by other things such as zone starts and quality of teammates / opposition.
And there's a huge danger of assuming there will never be outliers. For instance, there was a terrible post on an SBNation Oilers blog recently basically saying the Dallas Stars were lucky to be at their current points totals because they had won a lot against the East and because they won a lot of one-goal games. He had all sorts of numbers to back this up.

But both points were fairly ludicrous. For one, all Western Conference teams have basically equal opportunity to rack up points against the East (and vice versa). That the Stars are more effective than most tells you nothing than.... the Stars are more effective at that than most. Nothing further. The second point was even dumber. The Stars do win a disproportionate number of one-goal games, but that doesn't mean it will equalize or that they're "lucky." Being an outlier sometimes just means you're an outlier.

I recall that article, and while I didn't think it was particularly great, it raised some interesting questions about how much of Dallas's success can be attributed to luck. You're right that there's always the possibility of an outlier, but it's usually not wise to assume that it is without a reliable sample. It's not as if Dallas has an excellent defense or past success that would indicate they are a team that would thrive in close games. Time will tell if Dallas was simply a beneficiary of favorable percentages or not, but I wouldn't write the article off that quickly. I do think the part about beating up on the East was a little bogus though, as it seemed there was no difference in their underlying numbers against either conference.

If a team has a 38 percent power play, at this point of the year, I'd say chances are greater than not that it will stay much, much higher than the statistics will tell you it should because we're well more than halfway through the sample set. True numbers nerds, however, will try to make an argument that their PP numbers say that team's power play will drop dramatically over the last half of the season because no team has ever had an average that high or whatever. Outliers will always exist, no matter how much the numbers might say things will equalize.

Regression to the mean is a pretty basic statistical concept that almost always holds true when the percentages are out of the ordinary. There are some instances of outliers, like goaltending performance under Jacques Lemaire, but these are relatively easy to spot and disect. Not many of them exist, either, as evidenced by you doing a lot of handwaving instead of illustrating real examples of current outliers.
 

Lard_Lad

Registered User
May 12, 2003
6,678
0
Kelowna
Visit site
Anybody who is familliar with evaluating advanced statistics in hockey wouldn't put that much weight into raw Corsi numbers, as they are often very team dependant and can be greatly impacted by other things such as zone starts and quality of teammates / opposition.

And it's fine to try to improve on Corsi by modifying it for those factors, but they've got their own problems (gross problems, in the case of QoC/T) and are hardly comprehensive in addressing all the issues with trying to draw useful conclusions from shot totals.

But people work with what's available. The biggest problem is the NHL recording stats in ways that aren't terribly useful for analysis. Pure individual time/change-of-possession stats would be more valuable than biased, subjective giveaway/takeaway numbers (and would remove the need for doing things like using shots as a poor proxy for missing possession data.)

It'd be nice to have a stats board on HF.
 

iceless

Guest
My take is that numbers are a great tool for informing (supporting) a point of view. I follow stats, but the depth to which some today (including on this very board) drill down numbers is truly admirable.

All that said, statistics are not a de facto substitute for an opinion. Unfortunately, one sees that misperception played out here regularly - sterile numbers presented, bereft of context and perspective.

Likewise, investing in a pair of skates and a stick, as well mastering the ability to observe, formulate and articulate thought beyond simply reciting numbers, are wonderful things from which everyone, number cruncher or otherwise, can benefit greatly when analyzing and discussing the game.

^Agreed 100%.

Numbers are only part of the story when it comes to analyzing hockey players. Plus/minus is a particular indicator that should be either re-vamped or removed completely, because when analyzed on its own, it is a sketchy indicator of said player's defensive prowess -- which is essentially what +/- is supposed to indicate. More often than not, this is a stat that says 'this player played on a team that is stacked offensively.' Check out Zdeno Chara or Andrej Meszaros from 05/06 to now for recent examples.
 

Sensfanman

Registered User
Jan 27, 2006
10,184
1
Los Angeles, CA
I'm not sure exactly which part of which metric you're referring to here. I'm going to assume it's the offensive contribution to GVT vs the defensive contribution.

It might be valid within sets (I'm not conceding this at all, but for the sake of argument), but it's not necessarily valid if you're going to combine the two sets. If you're going to combine the two sets, then the scales should be equal...not just in equal proportion. After all, a goal against is just as damaging as a goal for is beneficial. (As a tangent, in baseball there's a trend away from thinking that a run is a run is a run. After all, if a team scores a run in the top of the ninth to make the score 10-1 they haven't significantly added to their probability of winning the game at all. Scoring a run to make the score 1-0 in the same situation, on the other hand, increases their probability a great deal...but I'm ignoring that sort of thinking for the moment just for the sake of conversation).

Anyway, according to this metric, if there was a universal draft then approximately 60 skaters and 23 goalies should be drafted before the first defenseman is taken because they provide more value over replacement level than a defenseman does. Also according to this metric the top defensive players in the league are worth about 5 goals more than a replacement level player (to this point in the season, anyway) on the defensive side of the puck. I'm sorry, but that doesn't pass the smell test at all.

So I looked into it. I think the difference is that there is a lot more spread in offensive talent. Since it's relative to a threshold, it's possible there isn't much difference in defensive acumen from the threshold; that's why the scales are different.
 

EastonBlues22

Registered User
Nov 25, 2003
14,807
10,496
RIP Fugu ϶(°o°)ϵ
So I looked into it. I think the difference is that there is a lot more spread in offensive talent. Since it's relative to a threshold, it's possible there isn't much difference in defensive acumen from the threshold; that's why the scales are different.
Therein lies the problem. Offensive contributions are measured from individual stats, so it's easy to distill the difference between those who are deemed to contribute among the five players on the ice and those who aren't.

Defensive acumen is directly tied to Plus/Minus (or a variation thereof), which means that the blame for each goal is originally split equally between the two defensemen on the ice (2/7 apiece) and to a lesser degree among the forwards on the ice (1/7 apiece)...even if the "blame" lies consistently with one person among each defensive pair, or one person on each forward line, who happens to be the obvious defensive weak link.

In other words, this effectively regresses the defensive contributions portion of this metric toward a common mean among all players who regularly play together (i.e. a great defenseman plays with a crappy one, their ratings will be obviously be more middle of the pack than a great defensemen who plays with another great defenseman, or a crappy defenseman who plays with another crappy defenseman...yet the true talent levels of the defensemen have not varied at all). The same is true for forward lines that play together, and this is why there isn't as much "difference" observed between the defensive contributions of this metric and the offensive ones.

A secondary byproduct is that the metric chronically undervalues defensemen relative to forwards. As I implied above, the credit for every goal for is split among 1 to 3 players...not 5. This concentrates the value of the credit that's assigned for offensive contributions in general, and particularly among forwards since they're going accumulate more points in general than defensemen simply as a byproduct of playing their position. Since, as we just pointed out, the method for distilling defensive contributions effectively dilutes defensive contributions among teammates, there is less observed (and thus credited) value there. It also leaves defensemen more succeptible to fluke variations in Plus/Minus swings than forwards because they are deemed to be twice as responsible for such things. It all stacks up to give forwards a huge advantage over defensemen when offensive contributions are combined with defensive ones to come up with one final number for ranking players...and this is exactly what we observe when we look at the ranking list.

Where does that leave us?

For one thing, value over replacement should be measured against positional peers. That is, offensive contributions for each forward should be compared to replacement level for forwards in that category, and offensive contributions for each defenseman should be compared to replacement level for defensemen. That removes some positional bias, because now you're comparing apples to replacement level apples and oranges to replacement level oranges.

For another, the defensive contribution portion of the stat is fatally flawed as it stands right now. It should be left out completely, IMO. All it does is skew our overall impression of the metric, and give the false impression that the difference between the best defenders in the league and a replacement level defender is a goal every ten games or so. That's obviously not true.
 

Sorge Georos

Registered User
Apr 28, 2009
3,026
260
LI
Herein lies my biggest problem with these type of stats. Stats should be used to supplement your argument, not to form an argument because of a stat. Put it this way, I wouldn't want someone to look at that and say "well, Lidstrom has the best CORSI Rel QualComp therefore he should win the Norris".

While I can appreciate stats (or sabremetrics) in a sport like baseball, I truly don't believe those types of stats are the best indicators in the sport of hockey. Ironically, hockey is such a face paced game that it's difficult for a viewer to form an opinion as there are so many different things going on (as you've mentioned). Maybe I'm just old-school but I will never believe stats can give you the fullest and most complete analysis of the game that you've never watched.

Yo are spot on. Baseball is a slow paced game with things happening one at a time. Therefore it is easy to use Sabremetrics to gauge your argument. Everything that is happening is logged down and there are numerous stats that are good indicators of the luck facor and how the layer will perform in the future.

I'm sure sometime in the future, something will be developed for hockey, but as of now you just have to watch games on a consistent basis. I long for that day when people realize that number of Cups is not a good way to judge an individual player, like the RBI is not a good way to judge a baseball player.
 

Ad

Upcoming events

Ad

Ad

-->