Puck Stops Here:

I think you’re mistaken in regards to my main point on plus/minus. Essentially, it’s a team stat and a good team stat.

If a team is +60 over a season at even strength scoring, you know for certain it’s a good team.

When you apply this team stat to individual players, however, it tells you much less, because an individual player is only 1/6th responsible for what goes on the ice when a goal is scored for or against his team.

This application of team stats to individual players is a fixture of hockey analysis, but when you have 30 or 40 per cent false positives and false negatives—as team plus/minus does when applied to individual players—it renders the stat unreliable. It’s neither fair nor accurate.

My goal, if nothing else, is to cut down that rate of false positives and false negatives in a plus/minus stat for players, to be more fair and more accurate.

The goal of hockey sabrmetrics may be to find objective standards, but in the work we’re doing the problems of subjectivity can be controlled if a team of people go by an objective set of standards in making a call. That’s what we do. If you want to review the standards, contact me and I’ll send them to you.

This method can provide useful information. It already does in other sports. It provides a stat like “assists” in basketball and “errors” in baseball. I know baseball stats guys hate “errors,” but it’s proven to be a durable and somewhat useful stat.

The new fielding stats in baseball also have an element of subjectivity and scoring chance stats in hockey, which are widely used, have a big subjective element.

Here’s another example . . .

There’s no end of controversy about referee calls in a hockey game, which are subjective calls governed by an objective standard.

But over a season, if we see Player A has four penalties and Player B has 40, we don’t go saying, “Hey, that means little, because this was too subjective.”

We say: “Player A plays a cleaner game than Player B.”

We say this because we know that most often the refs get it right, even if they make the odd mistake.

Another example is the strike in baseball, a subjective call made on the spot by someone trained on an objective set of standards. Now any one call can be hugely controversial, but baseball fans accept the results over the long term. They accept a pitcher who strikes out 250 guys over a season is a better strikeout artist than a guy who strikes out just 50, even if his strikeout total is governed by a subjective call (and one that doesn’t have the benefit of replay to get it right, a benefit we have).

If over a season, Corey finds finds Florida Player A has contributed to 50 goals and made mistakes on 40, making him +10, while Player B has contributed to 30 goals and made mistakes on 40, making him -10, I’m going to put a lot more weight on his finding—which is governed by his judgment but guided by guidelines—than I do on any kind of team stat crudely applied to either player.

So I don’t believe the issue you highlight is dire. I’ve been assigning errors for two years now. Some people don’t like it. Others do.

Perhaps it’s more on the line of an enhanced scouting system on a given team, a scouting system that has some built in objective standard to guide it along. If that’s all it is, that’s fine with me.

If others want to go a different route and keep on applying team stats to individual players, well, that’s their business, of course.

P.S. You write: “Perhaps all of those elements are not equal.  A goal scored is clearly more important than a missed shot.”

Now, it’s true that on occasion I’m given to over-statement. But, perhaps, this was a bit of an understatement.

Posted by David Staples on 10/20/09 at 01:32 PM ET

Dave

There are several reasonably well proven methods to handle the comparison of players +/- ratings on different teams.  Here is one method that treats +/- as a rate stat and here is another that treats them as a counting stat.

Your problem of team effects on individual players is largely solved under both methods - of course there are always other effects that one might then worry about.

I think a better solution than worrying about false positives and false negatives (and inputting the bias that comes in determining them) is to gather sufficient data that the false counts are not significant to the result and can be adjusted for, as is done in the +/- examples.

The problem in picking which players are important in goals scored is that there is too much subjectivity (at least as defined in your article).  For example, is the guy who carried the puck up the ice important in scoring a goal?  Does it matter if he brought the puck up the ice on the previous shift before getting a faceoff deep in the offensive zone and and going off on a line change?  In order to subjectively give out goals, +/- etc by the method you spell out you must assume answers to these questions.  Either you give them credit or you do not - the subjectivity of this judgment is built into your system.  I argue we need to gather the data and study this.  We can amass a list of who brought the puck into the offensive zone for each goal and how long before the goal etc and quantify these effects.  Instead, the system you describe subjectively chooses whether or not to include them and hence cannot ever gather results that go beyond the subjective inputs in the system.  In baseball with errors (for example) things are far better defined, because we know better what questions to ask and thus how to define things.  That said, conventional wisdom is that the number of balls a player successfully fields (his assists + putouts) is a better measure of his defence than his error total (for a given position - and with a couple other pieces of fine print).  A player who makes more outs and bobbles a couple balls getting errors is better than a player who makes less outs (has less range) but does not get as many errors.

Posted by PuckStopsHere on 10/20/09 at 02:22 PM ET

David, you make good points. 

If you read through his stuff, PSH is the slow kid in the class.

Posted by moore00 from Columbus, OH/Grand Rapids, MI on 10/20/09 at 02:41 PM ET

TSPH,

As the fellow who directed you to this post, I urge you to re-read the article.

Staples’ isn’t against hockey statistics; he seeks to develop more meaningful hockey statistics that explain rather than mislead. Moreover, the error assignment is Staples’ project (he’s been working it for the last couple of years on the Oilers). The interview is with a fellow who is going to take Staples’ methodology and apply it to the Panthers.

There is merit to a refined plus/minus that credits/debits players DIRECTLY involved in the play. Traditional plus/minus only reveals whether or not a player was on the ice at the time the goal was scored. It does not reveal if the player in question had any direct impact on the play or was in proper position of coverage.

For example, I can tell you that Shawn Horcoff and Ales Hemsky were minus 1 on the last play and that gives you the information that they were beat for a goal. But if I tell you that Hemsky was covering the right point and prevented the pass/shot from that zone, while Horcoff was in front of the net where he lost his assignment, screened the goaltender and the left point shot deflected off him into the net, doesn’t that provide you with more information on the cause? And now knowing this, why should Hemsky and Horcoff get assigned the same result (a minus)?

As this concept develops, I think you will find the hockey error is not subjective as much as it is interpretive.

Posted by Matthew McCallum from Redding, California on 10/20/09 at 02:50 PM ET

I notice between the time I started and finished writing—bookending a meeting that I had to run off and attend—David Staples posted.

Posted by Matthew McCallum from Redding, California on 10/20/09 at 02:57 PM ET