Analytics, or advanced stats, are here to stay. Though it may seem like a fad due to their explosion into the mainstream over the last year, the fact is, the NHL’s 30 teams are all, to varying degrees, creating analytics departments, hiring statistical experts. and generally attempting to get progressive about how they look at the game.
Predictably, there has been a lot of backlash against this movement. “It’s just a fad,” say some, while other say “I know what I see.”
Analytics aficionados scoff openly at those who are slow to grasp new information, and ‘old-school’ types lament their arrogance and double-down on ‘traditional’ ways to view the game, saying that they’ve watched hockey for 30 years and know everything they need to know already.
Of course, these are two extremes. For the most part, people who are into the stats side of things are patient and thoughtful and willing to share what they know and to explain it, while people who don’t know about them are eager to learn. It is my experience that people are not 100% on either side of the debate and that they are open to dissenting views and happy to argue them. Don’t let a few pompous jerks on either side close you off to an interesting and revelatory way to view the game. To wit: no one thinks you can learn everything about hockey without watching a game, just as most people – even if they are ‘old school’ to the bone – are open to new information.
The dichotomy of opposites gets blown out of control and in reality, there is no need to pick a camp. It is, however, necessary to understand that only a fool would turn down new information, that hockey analytics are not going anywhere and that there are legitimate lessons to be learned from the excess of information available due to the technology of our time.
One thing no one’s talking about when it comes to analytics and hockey, is that in every case where data is tracked and analyzed, no matter the subject, topic, or reason for tracking the data, things are revealed that surprise people who thought they knew what they were doing/watching/talking about. It isn’t just hockey: it’s the whole world.
The fact of the matter is that as humans, we are subject to a bias in everything we do. When we are assimilating information, every single one of us is guilty of confirmation bias – basically seeing what we want to. There is recency bias, where by we give credence to the most recent information and there are a whole bunch of other ones too. Then there are our likes, dislikes and our past experiences that we can’t shake in our evaluations, no matter how hard we try.
So it doesn’t matter if you’re just an average fan or an award winning head coach like noted analytics dismisser Ted Nolan, it remains that data will always reveal things you did not know.
With all this in mind, let’s clear up some misconceptions about advanced stats and list some of the things we have learned from them (besides the fact that Clarke MacArthur might be a better player than anyone previously realized).
1. Goals do not accurately measure performance
The fact is, there are just not enough goals scored to get a big enough sample size to make the data mean anything. A player can score two goals in a game, and if you look at the stat sheet, it appears he played great. But, what if he just got two lucky bounces and actually had a horrible game? If he’s a 20 goal scorer, that’s 10% of his entire output for the year and it’s already skewed, meaning that at the end of the year, we’ll have an inaccurate view of that player’s effectiveness.
If, however, we use shots as the metric to track in an effort to understand that player’s performance, we have a lot more data. A player could put up five weak wrist shots from the blue-line in a single game, but since he’s likely to shoot 150 odd times in a season, this doesn’t wreck the data like a lucky goal scoring game could.
Basically, a goal is what’s called a ‘low-occurrence, high-profile event’ meaning that it rarely happens, but when it does, such a big deal is made out of it that understandably incorrect conclusions are drawn. There is not a person alive who is not manipulated to some degree by low-occurrence, high-profile events. This is why it is important to question what you think you know about everything, not just hockey.
Over time, shots are a proxy for possession. You have to have the puck to shoot it, so by virtue of logic, if you out-shoot your opponent, you had the puck more. This, however, is an aggregate stat and I think that’s were the confusion comes in for a lot of people.
2. Advanced Stats don’t really matter in the short-term
In my experience in talking hockey with people, it seems that one of the major factors in their mistrust of these stats is that they are aggregates. This means that we are talking about a collection of data and the bigger the sample size, the more accurate the data is. This problem here is compounded by the demand for these new stats causing analysts and writers and TV guys to give out single game advance stats,which in turn contribute to the questions of veracity posed about them.
Single game analytics have their purpose but are dangerous to draw conclusions on because the entire reason for using shots as the metric in the first place is to get a read on the long term. Yes, some info can be gleamed, for instance: if a player had 9 shots in a game, he probably played well, even if he didn’t score. But, single game analytics need to be taken with a grain of salt because anything can happen in one game, but over time, things even out.
The fact that Colorado was so successful last season despite being a poor possession team, and going against the tendency of high percentage Corsi teams to win, causes a lot of people trouble. It shouldn’t. Because, in a long enough time-frame, anything can happen. It’s called an anomaly and it is not indicative of anything. If you won ten coin tosses in a row, it might seem like you were great at coin flipping, but keep flipping them, and eventually your winning percentage will revert back to 50%.
It seems to me that as analytics move into the mainstream, people are jumping to point out every little time they are wrong. But this is not right because the entire point of them is to get objective information and to see what is most likely to happen, not what will.
3. +/- is a useless stat
It drives me nuts to be watching hockey in 2014 and have paid analysts quoting a player’s +/- like it means something. It doesn’t. The problem with this stat is that it considers goals for and against equal. This is not quite accurate as it is very hard to score goals. We see that the same players year to year tend to be among the leaders in goals scored. This is because scoring regularly is very difficult.
Often times, a player will score a lot of goals but will also be on the ice against the other teams top line. In fact, it’s fairly common for teams to match their best lines against each other. This means that even though they are still scoring a lot of goals, players like Kessel and Ovechkin often are on the ice for goals against because over time, it’s unavoidable.
Also, scoring a goal is a relatively individual event. By which I mean that Phil Kessel scores most of the Leafs goals. If they were spread around more, we could adjust our thinking, but the fact is, the best players rack up the most goals. However, defense is different. A goal scored against is rarely the fault of one individual players – there are five guys on the ice and a goalie. A goal scored by Kessel is – over time – far more likely to be solely because of him than a goal scored against is likely to be his explicit fault and, as such, it’s not fair to count a goal he scores as equal to one scored against him.
There are other factors too: A player can get a minus on a line-change, he can be at the end of his shift, or his goalie can have an unusually low save percentage when he is on the ice. Finally, if he is employed in tough situations, there is just a better chance he will end up scored against.
A player who plays on the third pairing, has offensive talent and plays against weaker opponents is often going to have a high +/- than a better first pairing player who plays tougher minutes. That’s why we should just retire the stat.
4. Face-offs aren’t as important as people think
To qualify this statement, please understand it’s not an extreme view. I am not saying that face-offs are unimportant, just that their importance is overblown. Clearly, if the game is on the line and there is a defensive zone face-off, you want to win it. But, as an aggregate, they barely matter.
For one, possession – actual possession, not the proxy – changes very quickly in hockey games. This means that even if you lose a face-off, you can still get the puck back. Furthermore, the math shows that you would need to win approximately 45 faceoffs to equal a goal, or win 245 more than you lose over a year to equal a win.
There is another factor in face-offs no one talks about: Say a team wins the 17 face-off and loses 7. After the game, people talk about how dominant they were on the dot, and act like this was a major factor in winning because they won by 10 face-offs. The thing is, this isn’t accurate. They didn’t win by 10 face-offs because there is only two outcomes to a face-off and there are only two teams playing. That means that if the second team won only 5 more face-offs, the teams would have tied at 12 a piece.
To conclude on this subject, one more thing advanced stats tell us is that it’s a poor roster decision to allocate a spot for a face-off specialist. The days of the David Steckels are over – it just isn’t worth skating a slow, untalented (relative to the players on the ice) player to win a few extra face-offs per game.
5. Speed and offense win games
As a Canadian, I have been raised on the “right” way to play hockey. Hit, have heart, be tough, fight if you have to, play defense. Advanced Analytics, however, call into question this nationally ingrained psychology.
Did we ever stop to think that maybe the reason the Russians were so close to us, despite Canadians having a far bigger talent pool to choose from, was because they, and not us, played the game the “right” way?
I know this is almost sacrilegious, but maybe the reason they could compete so closely despite all their disadvantages, was that they played the game in an offense first fashion? This is just a guess, but analytics have forced me to ask the question.
This is because scoring goals wins games. If you are playing defense, you don’t have the puck. If you are a defensive specialist, you specialize in not having the puck. Yes, there are times when a defensively strong player is important to the teams success, but over time, a player who scores and takes shots and has the puck is more valuable.
As is definitely the case with face-offs, the media tends to make a bigger deal about the value of defense than is perhaps warranted. This is not to say it doesn’t matter, but to say that perhaps if it’s between two players making the team, and one is bad at defense but highly talented with the puck, it’s probably better to take him over the conservative, responsible player.
A look at the stats shows that offensive defensemen like Jake Gardiner have hidden value because despite their lack of defensive awareness, they drive possession and thus contribute to winning games.
To my mind, it’s worth pointing out that while a sea-change is occurring with the way we look at hockey, all the current coaches got their jobs by doing things the old-school way. This doesn’t mean it’s necessarily wrong, but it does mean that if Randy Carlyle makes Jake Gardiner a healthy scratch, that this could be a symptom of Carlyle’s outdated approach to the NHL and not Gardiner’s play. It’s really easy to spot that one mistake he makes in the corner that leads to a goal and then to remember it. But, it’s harder to notice and quantify the countless times he advances the play by making a low-profile play, like hitting his man with a pass through the neutral zone, or out-skating an opponent to the puck.
Advanced stats allow us to put these kind of plays into context and find out their importance over time, for instance, Gardiner appears to make enough low-profile good plays that the team significantly gets more shots when he’s on the ice.
I want to make clear that you shouldn’t be offended by any of this stuff. It’s not necessarily factual, it is just the conclusions I have drawn based on stats, things I’ve read, conversations I’ve had, research and my own observations. In every case where I am wrong, it’s on my own failed logic and where I am right, I don’t pretend to take credit for the idea, only to put it out there for discussion. I don’t know everything and I am sure there are many holes in some of my arguments. It doesn’t matter – the whole exercise is about re-thinking the game and trying to learn all we can. I don’t consider that advanced stats can tell you everything without question, but I do think it’s silly to ignore new information when it’s readily available.
Thanks for reading.