Quality of competition/teammates is a great, great indicator. Love it.

I might also say this: I use all those stats side-by-side; +/- isn’t being thrown to the trash in favor of, say, Corsi.

I like Corsi (which should really be called total shots differential or something) because, as far wiser men than me often say, it’s a good indicator of territorial advantage.

I like the number of faceoffs taken in offensive/defensive zones because it gives you an idea of a give player assignements.

As for the plethora of adjusted stats: well I’m still wrapping my head around them, so…

Behind the Net uses GFON/60 minus GAON/60 to give an advanced +/-ON/60. This is a great stat, but is also dependent on quality of competition and quality of teamates.

BTN also gives GFOFF/60 minus GAOFF/60 to calculate +/-OFF/60 to give an idea of how the team performs when that player is off the ice.

If you measure those against eachother, I think you can get a pretty good idea of how valuable a player is to his team. Of course, you have to account for sample size as well, as someone with great numbers in these areas might not be so great when you realize they only played 10 mins a night for 7 games last season. (However, this player might merit more ice time next season.)

BTW, I really like the Quality of Competition, Quality of Linemates metrics, but have never exactly understood how they’re calculated. Can anyone humour me?

The ones I like are the usual suspects (corsi, GF and GA/60, qualcomp, head starts (offensive – defensive zone faceoffs at start of shifts). I think EV shots/60 gets way too little credit. If corsi matters and pts/60 matter, why not put 2 and 2 together and use shots/60?

One thing I like that I’ve never seen anyone else mention is what percent of a player’s EV pluses that he got a point on. Basically, points divided by GFON, all at even strength. I think it’s usually a better indicator of how good a player’s linemates are than qualteam, at least for offensive-type forwards. It’s pretty interesting for defencemen, too, but in a different way.

For goalies, keep it simple. EV save%, career save%, and career EV save%.

1. Pts/60 at ES and on the PP, simple and efficient. I ignore SH scoring rates, seems to be 95% random, virtually no year over year consistency.
2. Quality of opposition/teammates. Only useful when comparing within a team context, and it has some methodology flaws but it’s a good look.
3. Zone shift/faceoff ratios… tremendous, this combined with Quality of opposition gives an extremely accurate look at the difficulty of minutes.
4. CORSI, broader sample size then goal rates. And greater/lesser shots virtually always lead to greater/lesser goals in the long run..
5. GF/GA while on-ice vs. off-ice.
6. Goalie EV SV%. Far more repeatability then SH/PP. My impression thus far is that it seems less influenced by team ability as well.
7. Penalties +/-. Nice stat, really shows whose drawing a lot of penalties and benefiting the team. Seems to be pretty repeatable over the long term either, so it’s a legitimate skill and not randomness.
8. PDO number. Seems like a good ‘first look’ to see whose riding the percentages and getting lucky/unlucky.
9. It’s a small thing but I like seeing the percentage of team goals a player plays a role in as well.
10. Adjusted RTSS stats to mititgate what seems to be a pretty significant amount of arena bias.

One stat I’m putting far less faith in as time goes by is shot quality metrics.
It definitely doesn’t seem like players have much impact on the SV% of the goalies behind them, and goalies switching teams seem to see their SV% far by far less then SQ SV% would imply they should.
I’m not quite to the point of considering it pure randomness like some have posited, but I do think there are probably only 2-3 genuine outliers in terms of Shot Quality and everyone else is probably fairly closely clustered.

I really think you have to look at stats in combination with others. CORSI without knowing quality of opposition tells little. Quality of opposition could be very misleading without taking into account zone shift. etc.
There is no one single stat that tells everything.

All those who have said hockey stats are in their infancy are on the right track.

Right now there’s a ton of information left on the ice. The goal of hockey statheads should be to collect that untracked information, so we have more valuable and useful measures of individual player performance.

One big problem right now is that we have all kinds of team stats, such as +/- and Corsi +/- which we clumsily apply to individual players.

A player is on the ice for a goal against or a shot against, but he’s just one of five position players on his team at that time, yet we give him a plus or minus mark as if he were completely responsible for the event, for the goal against or the shot against.

On every goal scored, and on every shot on net, about one third of the players on the ice get a plus or a minus mark that they simply do not deserve, not in the least. They had nothing to do with the play, yet still get a mark, and we still refer to their overall mark as if it were meaningful. I do it myself, but I’m going to stop.

While traditional +/- and Corsi +/- have real use when it comes to evaluating teams, which makes sense as they are collective stats, they have limited use otherwise.

How do get past this problem?

I think we should study the most important events in the game — goals scored and scoring chances — and figure out which players were truly involved in the play, giving credit on an individual level only for the good plays made by each player and the bad plays made by each player. We need to figure out individual plus/minus, individual Corsi plus/minus, individual scoring chance plus/minus.

This can’t be done by data mining, it can only be done through careful observation of each game, and many won’t find that appealing, I understand. But it’s the way to go to get more precise plus/minus information.

The most useful stat, IMO, would be to count up scoring chances and assign plus marks to players who created them and minus marks for players who caused them.

A set of objective criteria can be drawn up and used to help cut down on the subjectivity of this exercise.

So that’s what I’m going to be doing. If others want to keep mining the existing data and relying on team stats such as traditional +/-, Corsi +/-, and QualComp based on traditional +/-, I’d agree that that’s as good as we can do right now, but it doesn’t get us very far.

Dennis did great work all year and made reference to his work many times. But, as I thought more about what he was doing, I still had one problem with it: Dennis was also applying a team stat to individual players.

So on every Oilers scoring chance, all five Oilers players on the ice would get a plus mark, even though, almost always, one, two, three or four of them had nothing to do with the scoring chance being created.

For instance, a guy like Horcoff might be out for 300 scoring chances, but on 100 of those chances, he had nothing to do with the play (I’m just making up the numbers here, BTW, I didn’t track this). And Horc might have been out for 290 scoring chances against, but on 150 he wasn’t a culprit on the play. He made no mistake, he actually made a good play, but he was still given a minus.

So the team plus/minus number when he was one of five players on the ice was 300 plus, 290 negative, making him +10. But if you tracked each player, if you focused in on Horc’s real, individual contribution, he would be +200, -140, for an overall +60.

Dennis tracked a team stat, or a unit stat, not an individual stat. So while I think he’s tracking the right thing — scoring chances are significant, important events in a game — to turn this into an individual stat, you need to break it down more, get more precise information

I don’t know if Dennis plans to do that this year. That’s his business, of course. I might do it myself, but it will take a ton of work. I’d certainly be willing to work with others on such a project.

Staples, you’re wasting your time. It’s a team game and your arbitrary attempts to break down scoring plays are simplistic and heavily biased.

Furthermore, I’m not convinced there’s truly a problem there anyway. This is just something for people to fixate on because they’re too impatient to wait for a sample size to develop.

Not to mention there are other ways to deconvolute the individual contribution to team stats if so desired. That’s what all of the Desjardins stuff has been looking at. If he switched it from goal events to Corsi (or Total Shots, which I don’t mind) then it would be even more powerful, IMO.

@RiversQ.

If you’re not convinced there’s a problem, why don’t study ten games worth of Corsi shots, in depth, in detail, and find out how many false positives and false negatives there are?

I suspect I get it wrong in my plus/minus assignments about 10 per cent of the time, as arbitrary and simplistic as they are. But I suspect team Corsi plus/minus when applied to an individual player gets it wrong at least 30 per cent of the time.

@Derek.
I think Jonathan’s method of figuring out QualComp is a helluva good idea, mainly because it’s based on an individual stat of some mert, points scored. When Gabe Desjadins recently applied Jonathan’s method to the Oilers, he came out with a superior QualComp number, IMO.