In Part I, we looked at the most common way “advanced” statistics are used to evaluate team performance. In Part II, we get to the much trickier part: judging individual players.
An important part of evaluating individuals—and one that is often overlooked in traditional media—is context. Two statistics in particular can help shine light on something that should be of keen interest to any fan of the game: how a coach uses his players.
Quality of Competition/Teammates
NHL coaches pay deep attention to their matchups, so some players constantly face the other team’s best and others see them only rarely. Naturally, this matters a lot; it’s easier to look like a quality player against Craig Adams than it is against Sidney Crosby.
By the same measure, it’s a lot easier for a player to score a goal with Sidney Crosby as his centre than it is with Craig Adams in that role.
Various ways of measuring quality of competition have been calculated (using plus/minus, points, time on ice or shot metrics) but they all produce basically the same results and show which players have been taking the toughest matchups.
For fans who want to know whether their coach prefers a power-vs.-power matchup or employs a checking line, or whether a certain defence pairing is facing top opponents or weak sisters, these numbers can be extremely valuable.
And while a fan may know just by watching how the coach of his favourite team runs his lines, that knowledge can still be valuable—not just as a check on his team, but also as a way of evaluating a player acquired in trade or the preferred strategy of a given opposition team.
Some players start lots of shifts in the offensive zone, while others are asked to start 200 feet from the opposition net. Naturally, it’s a lot easier to look good doing the former rather than the latter.
To pick the most famous example, in a little over two years with Vancouver, Manny Malhotra was on the ice for 257 offensive zone faceoffs in five-on-five situations (as recorded by behindthenet.ca). He was on the ice for 1,086 defensive zone draws.
That sort of disparity makes it a lot harder to out-score the opposition.
It's also the kind of number that can help explain why a certain player looks bad but keeps getting ice time. A fan might be more disposed to cut a checking centre slack if he knows that he's starting at the wrong side of the ice over and over again.
But those numbers only tell us how a coach is using a player, not how said player is performing in the role. That matters, but a player still has to produce.
Traditional statistics like goals and assists can still be informative, but it’s helpful to modify them slightly.
For example, instead of looking at point totals, an analytics-based approach is more likely to look at points per 60 minutes, and then to further divide that into five-on-five and five-on-four situations.
This compensates for factors like increased or decreased minutes and shows exactly where a player’s offense is coming from.
Shots and Shooting Percentage
One of the big items that can sway a player’s totals is percentage-based fluctuation. As at the team level, sometimes the pucks either go in or don’t go in, and a lot of times that can fluctuate from year to year with very little influence from the player.
As an example, let’s look at individual shooting percentage.
On his career, Alexander Ovechkin is a 12.3 percent shooter. In 2010-11, he took 367 shots; at his career level we would have expected 45 goals, but instead he scored 32. In contrast, in 2007-08, he took 392 shots; at his career level we would have expected 48 goals, but instead he scored 65.
Good scorers and bad scorers alike fluctuate like this.
For veteran players, career shooting percentage is the number to bet on in the long-term; for young players it’s just important not to assume that a brilliant or brutal shooting percentage is going to continue indefinitely.
As with the team numbers, taking shooting percentage into account can help calm fan worries during a cold stretch and help temper excitement during a hot stretch.
It's a valuable tool for someone interested in keeping his or her expectations reasonable.
The same thing happens for total shots with a player on the ice. In 2010-11, with Ryan Getzlaf on the ice, the Anaheim Ducks scored on 12 percent of the shots they took at five-on-five. Getzlaf put up 76 points in 67 games.
The next year that number dropped to less than 7.5 percent, and Getzlaf fell to 57 points in 82 games. In both seasons the Ducks dominated the shot clock with Getzlaf on the ice; the difference was just that one year the pucks went in, and in another they didn’t.
As at the team level, shot metrics—most commonly five-on-five Corsi per 60 minutes relative to team performance (abbreviated CorsiRel)—are seen as critical in player evaluation.
In a lot of ways, Corsi is extremely useful, since shot differential drives team performance, and good players drive shot differential.
But it’s also dangerous. A good player can post poor results (and a poor player good results) depending on usage, linemates, coach’s system, playing out of position, injury and a hundred other factors.
One important tool is WOWY (with and without you) analysis, which shows how a player performs with and without another.
For example, according to HockeyAnalysis.com, the Boston Bruins out-Corsi their opponents by a 3-2 margin with Zdeno Chara and Johnny Boychuk together on the ice; that number falls to below break-even when Boychuk is out without Chara.
There is no one right way to blend all those numbers together to evaluate an individual player.
While attempts have been made to unify the statistics above into a single number, the risk is losing important details. The important thing is to use the best data available, even if what each person considers the best data varies significantly.
Then there is that other position: goaltending.
There are two basic rules to the statistical analysis of goaltenders.
Firstly, of the commonly used statistics, save percentage is the only one that really matters. All the other numbers regularly cited, from wins to GAA to shutouts, are a factor of save percentage and items beyond a goalie’s control. Even-strength or five-on-five save percentage is the key metric here, as it helps reduce the effects of good special teams play or a disciplined/undisciplined team.
Secondly, always bet on the long-term track record. Goalie performance fluctuates wildly up and down, and a single week or month, or even season, often doesn’t reveal the player’s true quality.
As an example, in the last two decades four goalies have been named NHL Rookie of the Year: Martin Brodeur, Evgeni Nabokov, Andrew Raycroft and Steve Mason. Raycroft and Mason were every bit as dominant as Brodeur and Nabokov, but that quartet has enjoyed very different careers.
There is a lot to take in above, and in Part I, but we can boil it down to some basic principles.
At the team level, the single biggest predictor of team success to date is shots. That means out-shooting the opposition at even strength, generating a lot of shots on the power play and preventing shots on the penalty kill.
At the individual-skater level, context is critical. Without knowing the situations a skater plays in, who his linemates and common opponents are and where and when the coach deploys him, it’s very difficult to arrive at a legitimate conclusion.
Using that context, performance can be evaluated.
Traditional numbers like goals and points should be expressed as rates. The team’s performance (as measured by shot metrics) with a player on the ice is critical, but every effort should be made to determine whether that performance is being driven by the player or his environment.
For fans, those numbers can help in a lot of ways. They allow the fan to check his or her observations against objective statistics and to gain insight into teams he or she doesn't get to see very often.
The numbers not only help keep expectations reasonable, but they can also reveal new facets of the game for new and old fans alike.
Above all, the best thing to remember when looking at statistics is this: More information is better than less information. Sometimes that information comes by eye, and sometimes from the data, but using both is better than leaning on either alone.