News & Updates
June 1, 2016
When Google started using their algorithm to predict your tastes, it was freaky. You’d be writing an email in your gmail account about a trip to Mexico, then not long after there would be an advertisement for trips to Mexico in your searches. Weird – Google is reading my email…
We might not like the idea of having our tastes reduced to an algorithm – a formula. But you also sort of enjoy it when you buy something that you “knew you wanted anyway” from Amazon. Or find another season to watch on Netflix.
The reality is that there are mathematical formulas which predict our tastes. They also predict our behaviours.
But it didn’t start that way.
Let’s look at the path that knowledge takes:
- It starts as a mystery
- Then it becomes a heuristic
- Then it becomes an algorithm
The mystery is a bunch of information that has no connection. No seeming interlinkages. When someone starts to form conclusions about how the information is linked, it becomes a heuristic. A heuristic, or mental rule of thumb, organizes the information in a more meaningful way. This is where innovation happens. But the heuristic still has bias. So the next step is to systematically study the heuristic and make it more simple – reduce it to a mathematical formula.
Along the way, knowledge goes from exploratory to becoming exploited. What was learned in the heuristic stage gets exploited at the algorithm stage.
Not long ago, most of hockey was in the mystery stage. People attributed talent to inborn natural ability. They didn’t even try to do strength and conditioning. They would explain the success of certain players as literal mysteries.
Now we see some connections. Coaches are getting smarter. So are players. So are managers and parents. They’re starting to make connections in the patterns. There is still some bias. Bias is okay. It’s just part of the stage. And this is the stage is the heuristic stage.
I’d say that 90% of hockey is still in the heuristic stage. There’s nothing wrong with this. It’s just the way it is. Certain businesses like McDonald’s, Amazon, and Google are way past that. They’re exploiting the benefits of their algorithm. A sport like baseball is further along the continuum towards algorithm than hockey.
At the management level in the NHL, we see a shift towards algorithms. Managers figure out how to put together teams with advanced stats. They use algorithms to measure player value.
For a long time, I’ve been dissatisfied with my own role as a hockey development specialist. I have my own heuristics around what makes a good hockey player. I believe they’re more valid than other heuristics used by other coaches. For example, many coaches and scouts believe that big, smooth skating defensemen are the answer to any potential situation. When I stood on the bench and coached a team, I had a bias towards putting out my bigger defensemen. I’m not sure if that was because they were the best players, or that their size inclined me to think that way. Either way, I don’t know for sure. Neither does a scout. Neither does a manager. Neither does another coach. I used to think that the “bigger is better” heuristic was junk. But now I’m not sure.
As a hockey development specialist, my business survives on the perception that I’m making hockey players better. Once again, I truly believe that I’m doing the right things. I believe I’m giving my guys a bigger advantage than what other guys get with other coaches. But how do I know for sure? I can’t. You can’t. They can’t.
The earliest to the algorithm game in hockey was Anatoly Tarasov: the father of Russian hockey. He was using advanced stats before they were in vogue. The next to the algorithm game came Darryl Belfry. Whether he says this or not, he’s creating a formula for player success: an algorithm. He seems to measure many details, then sift through to find the ones that make the biggest difference. I’m not sure how his statistical process works, but it seems to work in the NHL. Good enough for me.
Many coaches, development specialists, and business people try to “measure things”. But the things they measure have close to no validity for predicting hockey success. “Oh ya? You measure shot speed to .00001km/h? Wow!!!”
For example: I was tagging a player the other day and on a shift he was -3 for Corsi. Except all three shots originated from the other side of the ice. The player I tagged was in position and had nothing to do with the shots against. He then retrieved a loose puck, and exited the zone. We tracked the shots against, the retrieval and the exit. So, which piece of data was most meaningful?
Obviously the loose puck recovery and exit showed what the player contributed on that shift. He wasn’t penalized for the shots against.
As a former pro and college player player, myself and my cofounder know which pieces of data to pay attention to. Then we use a couple simple, yet rigorous statistical methods to find the most important data points. To mathematically prove our heuristics. With that we’re building an algorithm. We’re building a formula for elite player development. And we’re getting data from all levels.
This is our drive to move from heuristic to algorithm. It greatly improves the rate of development of the average player. Very few in the space have the combined technical and mathematical expertise to do this. We count ourselves lucky to be working together on this.
Will this stymy innovation in hockey? Yes. An algorithm naturally does that. It trades the search for validity for reliable outcomes. With an algorithm, you can get reliable development. Killing innovation in hockey is not a good thing if you’re killing Darryl Belfry’s ability to innovate. We aren’t trying to do that. But killing “innovation” in hockey is a good thing if you’re killing an inexperienced coach in your association’s experiments in “chip & drive methodology” vs the “get it out methodology”. In that case, you want a formula. You want to drive out heuristics and amateurish experimentation. You want reliable development.
This may not be exciting to you. But it is to us. When you are paying for results from coaches, and coaches have nothing to be accountable to, you’re essentially paying for a mixture between hypnosis and snake oil. And that includes me and what I do. I just happen to be really good at hypnosis. And I have the best snake oil.
But now that all changes. If we have data to meaningfully show the improvements that a player makes, we have something to be accountable to. You can have that too.
P.S. If you liked this article because it was different than most Drone Coach advice, and you’d like to get to work on becoming a Hockey Wizard, then click here to check out the benefits of becoming a Train 2.0 Member.
April 19, 2016
We know that coaching doesn’t quite work the way it should. We also know that there’s a lot of technology out there. Today, we discuss how technology could actually help hockey players and solve the coaching problem.
Ben Schmidt is an old teammate of mine, an engineer who has one of his design patents in use by the Bill & Melinda Gates foundation, and is an ex-professional hockey player. He is here to get you up to speed on the technology available for hockey players right now.
Statistics in hockey are advancing. Have you noticed? For years people complained that goals, assists, plus/minus, didn’t show the whole picture of a hockey players value. Now we can look deeper with statistics like Corsi (shots attempts %), Fenwick, zone starts, etc. The statistically inclined fan may be using these fancy new metrics to evaluate their favourite NHL players or fantasy hockey teams. NHL teams are using them to scout, evaluate, and make roster decisions with big payroll/cap implications. If you have never seen them before visit NHL.com/stats/enhanced
In hockey, analytics has uncovered some great insights, notably that puck possession is a significant contributing factor to wins. The Los Angeles Kings and Chicago Blackhawks in recent years have been analytics darlings for their possession numbers. Here is their rank in terms of 5 on 5 Corsi in the regular season when they went on to win the Stanley Cup:
- Chicago Blackhawks, 2010 1st
- Los Angeles Kings, 2012 2nd
- Chicago Blackhawks, 2013 4th
- Los Angeles Kings, 2014 1st
- Chicago Blackhawks, 2015 2nd
Now this is not absolute relationship. However, if we think rationally about Corsi it is effectively measuring shot attempt % at even strength. The best teams at generating shots at even strength in the regular season are good ones to bet on in the postseason. Side note: The top two teams this season were the Kings followed by the Penguins, just don’t bet the house;)
** For our more statistically inclined readers check out: http://www.eightleaves.com/nhl-stanley-cup-champions
Now I am not here to tell you about how great these are. To be completely honest in terms of player evaluation I am not impressed. As a hockey player I know there is a large evaluation gap between what’s happening on the ice and the statistics that are describing hockey. I am sure we have all had that game where you played great; you contributed to generating offense, you were sound defensively and you had the puck on a string, but didn’t register anything on the scoresheet. Maybe these new stats will show your worth. In my opinion they may expose an outlier in a particular game, or an undervalued player on a bad team, but they are not revolutionary. Connor McDavid wasn’t discovered from the 4th line of the Erie Otters because he had great underlying peripherals. He was a prolific goal scorer.
To me these new ‘enhanced stats’ only describe 10% more of the game. Call me ignorant but as a baseball fan the statistics in hockey to evaluate, predict, and explain pale in comparison. This is no fault of hockey. Hockey is a complex game with no deliberate and repeatable mechanic. It is a free flowing team sport without the discrete variables that allow us to isolate those statistics in baseball. To put it simply a hockey game is a complex system that requires complex analysis.
Enter computer vision. From the footage that you or I will be watching the Stanley Cup playoffs there is a company gathering millions of data points. This company is SportlogiQ. They are based out of Montreal. The first time you as a hockey player/fan visit sportlogiq.com your heart rate will spike. There it is! Instead of fishing together a new statistic why not collect all the data. Time stamped x-y coordinates of every player and the puck.
Mark Cuban the flashy owner of the Dallas Mavericks funds SportlogiQ. But it is led by some very smart computer science and hockey minds. Their CEO, Craig Buntin, is a former Olympian. Recently, I spoke with Craig over a WebEx meeting. Craig explained to me that they are gathering time stamped data points at 30 times per second and using these to tag events in a hockey game. They employ machine-learning algorithms, which basically recognize patterns of events in a hockey game and group them . From this new wealth of data they can create metrics for hockey that are only limited by our own creative hockey minds. Here is a screen shot of their recent evaluation of Nazem Kadri:
What do you gather from this information? What sticks out to me is that the Leafs are a bad hockey team. They are recovering pucks in the OZ and making OZ defensive plays, but not generating many chances versus the league AVG. What this means is they are chasing the puck and forechecking, but they are not able to hold onto the puck and make offensive plays, except Kadri, he’s clearly playing well on a bad team. Further evidence is the Leafs inability to make passes on the rush. This requires passing skill (saucer pass), hockey IQ (predicting how rush will develop) or deception (making defender think you are shooting when you are passing). The Leafs are below league average, thus Kadri excelled this season with players of inferior skill level…… and I would say that analysis passes the eye test.
Computer vision is so clearly the most effective way to describe the complex game of hockey. Now what can we do with all this data. It’s a hockey game on a spreadsheet. Do we evaluate players, use it to scout the next Connor McDavid, or predict the next Stanley Cup champion. I have an idea and I’ll share that with you next.
September 22, 2014
I’ve read two books recently (“The Rise of Superman” and “Smart Cuts”), that talked about surfing. So that’s probably what has me using a wave metaphor to discuss the what is happening right now in hockey, in relation to analytics.
Not long ago, analytics were once the crest of the wave. Just last year, analytics were discussed on TSN, CBC, and Sportsnet in a flippant fashion. In the first few NHL talk shows that I’ve watched, I’ve already heard discussion of analytics in a more detailed and critical manner.
Remember that analytics were designed in baseball, and now in hockey, to take advantage of market inefficiencies. This means that they were designed to look at players statistically to see who made contributions to their team’s success in a way that is undervalued by team managers. If someone uses analytics to find these market inefficiencies, and the market inefficiencies exist, then a manager get away getting more value out of a player than they are putting in. The caveat, here, is if the market inefficiency exists.
If everyone is talking about analytics, and everyone is using them, then the market becomes efficient again…there is no sources of untapped value for a manager to tap.
So then, as this wave gathers energy, if using analytics becomes just another manner to keep up, what is a manager supposed to do to get ahead?
Here are two areas to look:
- Evaluative metrics that inform a development coach how to best improve a player. If a coach can use advanced stats/evaluative metrics/analytics to design a program for their improvement, the coach can identify areas of weakness. We assume that by identifying and working on areas of weakness, we can increase the rate of a player’s improvement. This is like what Darryl Belfry is doing with his players.
- Physiological measurements. I am making the hypothesis, that there are physiological markers that can predict a player’s performance. Back in the Vancouver Canucks’ Stanley Cup Playoff Run, I had the opportunity to hear from and talk to Dr. Len Zaichowsky. He was their director of sports science. He had the team tracking many physiological values: heart rate variability, sleep quality and quantity, multi-object tracking, T-wave (I’m not sure what that was). I think that they were tracking these values and using them to inform how the players practiced and played. They also had their most successful season as a team…ever. The next year, Dr. Zaichowsky was let go, and they lost in the first round of playoffs. They haven’t made it that far into the post seasons since…
- What values could you look at and why?
- In-game heart rate and heart-rate variability. It might be possible to determine what heart rate and heart-rate variability values a player demonstrates when they play at their best. It therefore might be possible to design interventions to more consistently get a player to obtain these values in-game, thereby improving their performance
- Resting heart rate variability and adrenal stress. Players can sometimes play well when stressed for 1 or 2 games, but their performance may drop off if they remain stressed for games 3 and 4. Perhaps it’s possible to put find and put players in a sweet-spot where their stress levels are in balance to provide optimal performance.
- Brainwaves and transient hypofrontality. As you know from my previous article on finding flow, turning certain brain structures off is important for getting players into flow. A player who is predisposed to being in flow with a certain neurological state, more often, will be a more effective player. Perhaps by monitoring and informing an athlete, coaches and managers can design processes to more consistently get their players into flow.
- What values could you look at and why?
I’m suggesting that crest of the wave is a place where there are market inefficiencies. The market is becoming efficient in the sphere of analytics, but might still be inefficient when it comes to measuring physiological values and evaluative metrics that can be used to design a development program.
What do you think? Are there other areas where there might be market inefficiencies in the game of hockey?