Tackling Analytics: Part Two
As promised, part two of this look at hockey analytics will address some questions and concerns that are often heard about certain statistics and the assumptions they make about the games.
As a quick refresher (or introduction), the Corsi statistic is actually incredibly simple. If I take a shot, no matter what happens to it, it counts as one shot attempt, or Corsi. The one important thing to note is that a given player does not need to take the shot to have it count. If Alex Ovechkin shoots the puck and it misses wide, all five Capitals on the ice gain +1 Corsi For and all five of the opposing players gain +1 Corsi Against. A high Corsi For number is good while a high Corsi Against number is bad.
Today, more and more things are being measured — there are people measuring how teams enter and exit zones, how often certain types of passes lead to goals and many other things. Still, the “anti-stats” crowd seems to be focused mainly on Corsi.
With that said, let’s jump in and respond to some of these questions and concerns:
“You’re just looking at numbers and not watching the games”
This is one of the most common issues people have with those who look closely at analytics. If I’m looking at a team’s possession (read: Corsi) numbers and come to a conclusion about a player, I’ll surely be told by some other fan that I don’t watch any of the games. In reality, there is no way for that person to know how many of the team’s games I’ve watched. Oftentimes, when it’s someone who isn’t involved professionally with the league, they’ve probably watched fewer games than I have. But here is where the statistics really shine through: I’m a college student, and while I have a busy life I probably have more time to watch hockey games than most people. Last year I watched roughly 45 Nashville Predators’ games and 30 New York Rangers’ games during the regular season. The next highest team? Maybe I saw Washington play five times.
I’d be willing to bet that there is an incredibly small number of people who can say they’ve watched at least 20 games of three different teams, yet I’ll inevitably talk to someone who tells me “Jonathan Quick is one of the top five goalies in the NHL” when they haven’t watched a Kings’ game all season. The likely scenario is that the only evidence for that claim is something that they heard somebody say on NBC or NHL Network. Now how many games has THAT person watched? I’d assume more than I have, but I highly doubt that a given NBC analyst has watched enough of each team in the NHL to confidently talk about every one of them with legitimate evidence. It’s a vicious cycle of reputations and hearsay that vaults certain players into the realm of “elite” while others are dismissed because one time someone saw them miss a defensive assignment.
Let me be clear that I am not criticizing the effort that on-air personalities put into their jobs. I’m just saying that it’s nearly impossible to be well-versed on every team in the league. If that’s the case, why not use available tools to enhance your knowledge of players and teams that you aren’t as familiar with? Analytics allow us to get a quick point of reference on these players that we otherwise might not know much valuable information about.
“Player X’s value can’t be quantified/he has intangibles”
In my post where I discussed separating the player from the person, I talked about how analytics have been portrayed as weapons used to discredit certain types of players. In addition to the examples cited in that article, the recent performance of Team USA at the World Cup of Hockey has highlighted some of these issues. Players such as Brandon Dubinsky, Justin Abdelkader and Erik Johnson were criticized for their inability to play strong possession hockey but were picked for the team because of their grit, intensity and leadership.
I’d first like to point out that leadership, while certainly a factor, is one of the most arbitrarily decided attributes in all of sport. Sometimes a player who shows fire and emotion and urgency is considered a great leader, while other times it’s the athlete who seems unfazed and cool even in the most intense of situations. Leadership seems to retroactively be applied to players who were part of some huge success, while others who are on the other side of that game are labeled as having poor leadership or character.
Now, do I believe leadership plays a role in the outcome of games? Absolutely. I think effort also does. But somewhere along the line it became assumed that analytics in hockey only represent the skill aspects of the player. If Shea Weber is a great leader and also intimidates other players on the ice, would that not factor into statistics which, at their core, represent a bird’s eye view of his play on the ice? Weber’s physical intimidation should mean players are more likely to dump the puck in than carry it in, which would lead to better possession stats for him. It should also mean that he clears out the crease well, giving up fewer rebound shots and fewer goals against. In reality, Weber gives up more shot attempts than anyone on his team except for his partner.
All of these things manifest themselves in the amount of shots taken by or against a team. If a player truly makes the other players around him better, then the numbers that represent all of those players should rise. Shot attempts aren’t a measure of how fast you are or how aggressive you are, they are a measure of how well you facilitate your team attempting to score and how well you prevent the other team from attempting to score. I would imagine leadership and grit are supposed to work toward scoring and preventing scoring against.
“Can’t playing styles skew the numbers?”
“How do you know Corsi/Fenwick really represent possession?”
These questions have a lot of overlap, so I’m choosing to answer them together. Some teams, such as old Soviet Union teams, would pass the puck around and wait for the perfect chance to take a shot, therefore suppressing their Corsi numbers even though they dominated possession. In the NHL, you could argue that teams would begin to take shots from everywhere they could in order to inflate these numbers if they truly led to winning. Teams play at different tempos and as a result will generate and give up different amounts of shots.
When looking at teams, it’s best to look at Corsi as a percentage: shot attempts for divided by the total shot attempts by both teams. A percentage above 50% means that a team is taking more shots than it is allowing. The top five teams by Corsi For % over the last five seasons: Los Angeles, Chicago, Detroit, Boston, St. Louis. A percentage below 50% means that team is allowing more shots than it is taking. The bottom five during the same period: Buffalo, Calgary, Colorado, Edmonton, Toronto. The first list features six of the last seven Stanley Cup champions while the second list features six of the last seven teams to pick first overall.
If you look purely at the Corsi For per 60 minutes rankings, LA ranks first over the last five years while Detroit ranks only 19th. However, the reason for Detroit’s success is similar to that of the USSR: while they may be generating shots at a lower rate, the time that they spend with the puck on their stick means their opponent is generating even fewer shots. That is why the numbers can’t be skewed: perhaps your offense will suffer as a result of your playstyle, but if you play good possession hockey then your defense will soar. Detroit is third in Corsi Against per 60 minutes over that time, one of the best defensive teams in the league.
Players, on the other hand, need to be looked at differently. Last season, the worst Corsi For % player on the Kings (playing at least 500 minutes) was Andy Andreoff with 50.42%. The Rangers’ best player in that regard was Keith Yandle with 50.35%. Does this mean that every single player on the Kings is better than every single player on the Rangers? Absolutely not. For this reason, many analytics proponents today use relative numbers. They take the Corsi numbers of a player’s linemates while he is on the ice and compare them to when he is not on the ice with them. The result, although imperfect, captures the positive or negative impact a player has on his team’s offense or defense. For this exercise, let’s take a look at the Penguins. Chris Kunitz has a whopping relative Corsi For per 60 minutes of 10.17. This means that for every 60 minutes Chris Kunitz plays, his team generates 10.17 more shots than they would if every member of the team (and opposition) stayed the same except for Kunitz being replaced by the other wingers on his team. Defensively, Brian Dumoulin led the way with a -3.6 relative Corsi Against per 60. For every 60 minutes of play, Dumoulin allows 3.6 fewer shots to the other team than his replacement defenseman would.
These numbers, as mentioned before, are imperfect. Brian Dumoulin’s relative numbers would skyrocket if he played on a weak defensive team such as the Colorado Avalanche. However, using these numbers in conjunction with more raw Corsi stats or even other numbers entirely helps to paint a better picture of a player.
“Some of the most connected people in the hockey world don’t support analytics”
That doesn’t mean they are right. Many people said Phil Kessel was a loser when he was in Toronto and could never win a Stanley Cup. Many people said the Sharks had the experience from playing in the rough and tough Western Conference that the Penguins couldn’t compete with. The analytics said otherwise in both cases.
That isn’t to say that these people don’t understand hockey at all. The thing is, many people in front offices as well as on TV screens across the world played and have been involved with the game for a long time. Hockey, especially in the NHL, is very different than it was even 15 years ago. Physicality and intimidation simply don’t have the impact they used to, while speed and skill reign supreme. Sometimes it takes an influx of youth in the management team to come to this realization, like in Arizona or Florida. For others, such as David Poile and Doug Wilson, two of the longest tenured General Managers in the league, it simply requires an understanding that the sport is always progressing and you can either stay in front and be successful or lag behind and struggle to adapt.
I would like to once again finish with the caveat that these numbers do not represent everything. Keith Yandle’s possession numbers last season blew Ryan McDonagh’s out of the water. That does not mean I believe Keith Yandle is better than Ryan McDonagh (although the gap is smaller than you might think). What these numbers do represent is a new way to look at things, and in many instances a less-biased look. Take a closer look at these statistics and ways of thinking for yourself and don’t be afraid to question conventional wisdom. That’s how you uncover what might be the next Cup-winning formula.
