Right now, the larger hockey analytics community measures possession by shot attempt or scoring chance differential when a player is on the ice. These are commonly called Corsi or SAT (all shooting attempts) and expressed as a percentage, i.e. CF% or SAT%. Scoring Chances are expressed in the same way (SCF%). The thinking goes that if a player is on the ice for more attempts for his team than against, in the long run that team will win more games and score more goals than not. All of this makes sense and I agree with it. However, is this the best way to measure possession? This is not a post to argue against Corsi and this thinking. Instead, I’m simply arguing there should be a better way to measure possession by capturing the quality of possession plays a team makes and allows.
Scoring Chances account for some of this quality as they seek to capture shot attempts that are more dangerous than mundane. Scoring Chances, as defined on War on Ice, must fall under these guidelines: 1) all unblocked rebounds and rush shots from the low danger zone; 2) all unblocked shots from the medium danger zone; and 3) all shot attempts from the high danger zone, which is typically represented by a homeplate figure centered by the slot. You can read the entire post here for more details.
However, as more data is collected on shot quality and pre-shot movement, we learn that not all shots fit so nicely into those three categories to determine if they are a Scoring Chance or not. The movement prior to a shot can be amplified so that a low or medium danger shot can be higher than one high danger shot. Through tracking passing data, I've come up with the term "Shot Sequencing" to capture much more than just location data. I previewed this here, where you can find links to Steve Valiquette and Chris Boyle’s work on shot quality as well. Shot quality is on people’s minds a lot since the larger analytics community all agreed, for the most part, on the value of possession metrics.
On Monday, there was renewed debate on shot quality, namely between Steve Burtch and David Johnson, as they are wont to do. After my presentation at the D.C. Hockey Analytics Conference, which you can watch and read my slides here, I was planning on writing something up in greater detail soon. After seeing the two of them having this debate again spurred me to action. The biggest thing that people seem to miss - and this is not directed at either Burtch or Johnson specifically - is there’s much more to it than just a shot. If you’re not including what happened prior to the shot taking place, you’re not talking about shot quality. You’re just talking about shots.
Shot quality is not a singular event. It is the sequence before the shot that gives it quality. Accounting for the sequence accounts for any and all work that the goalie must do before the shot is even attempted. If you aren't accounting for both sides of the equation (goalie and shooter), you cannot possibly attempt to paint a clear picture of what is quality and what is not. Allow me to demonstrate.
Weighting Possession Sequences
The way we track data allows us to isolate twenty-six different sequences of shot attempts that are possible through passing. We weight each sequence on the likelihood of the shot attempt resulting in a goal (often referred to as "True Shooting Percentage" or Corsi/Fenwick Shooting Percentage) to gauge their true value. To illustrate why we do this, have a look at three sequences below. One is a single pass that is a one-timer from just above the faceoff circles, another is a wrist shot inside the scoring chance (home plate area) from multiple passes inside the offensive zone, and the third is a one-timer from inside the scoring chance area that crosses the Royal Road from multiple passes as well, though the secondary pass is in transition.
There’s a clear difference in perceived value to these shot attempts is there not? In fact, it’s not only perceived, but there is an actual value to all three sequences. The first sequence results in a goal 1.5% of the time, the second sequence results in a goal 8.2% of the time, and the third sequence results in a goal 27.1% of the time. These are not percentages of all shots, but percentages of all attempts. When we measure possession, we measure by attempts, so that’s what I’m weighing here. Now, let’s compare this to a player’s SAT%.
So, if two players are on the ice for fifteen shot attempts with 66.7% possession (ten for, five against), we’d glance at that and think, "Okay, they both had strong possession games." However, if one player’s SAT For is made up of mostly attempts similar to the first clip above and his SAT Against is made up of attempts similar to the second clip, his 66.7% possession number, when weighted for the quality of shot sequences for and against (10*0.015 For and 5*0.082 Against), becomes are pathetic 26.7%. Quality matters. Sequencing matters.
The problem with looking at just the shot is that you’re ignoring everything that came before it. How teams generate offense, the ability of players to find open ice, and, perhaps more importantly, the ability of defensive systems to limit these opportunities, should all be measured appropriately. Now, I understand working with the data that is available, but you’re only going to get so far with surface-level data. So, let’s take a look at your New Jersey Devils skaters’ possession numbers and weight them to get a better idea of the quality of their on-ice possession.
Weighted Possession of Devils Defensemen
Before I get into the numbers for the defensemen, I’ll illustrate, again, how I arrive at these numbers. I’ll use Andy Greene as an example. In the chart below, I've included four of the twenty-six sequences that we track passes for. These are: multiple passes in transition, one pass in transition and one inside the offensive zone that yields a one-timer from the scoring chance area after crossing the Royal Road, two passes in the offensive zone and inside the scoring chance area, and a single pass from outside the scoring chance area.
You’ll see the number of times Greene was on the ice both for and against for each of the four sequences. The next number is the percentage chance of the sequence yielding a goal. This is taken from not only Devils games, but from our entire population (now over 10,500 generated attempts). The final columns illustrate the value of the sequences when weighted, followed by the Weighted Passing SAT%.
The second example chart shows Greene’s total events for and against for these categories, both weighted and non-weighted. The third column is his SAT% (Corsi) and his differential when weighted.
So, we see that when we account for the sequence that precedes each shot attempt, we can quantify the value of the team’s execution, or failed execution, in our percentages. Let’s get on to the rest of the defensemen.
Our sample size includes all players from 1/14 to 4/11 that played at least 300 minutes. The reason for this date range is that was when we changed our tracking sheet to time stamp events and pull on ice data. These are numbers from 5v5 situations.
Above you’ll see each defenseman’s SAT% and their Weighted PSAT%. Why is there such a big difference between a player’s SAT% and Weighted PSAT%? It’s because 75.3% of the shot attempts from passes that the Devils give up, are weak shots. Of the 954 passing shot attempts opponents generated since January 14th, 478 were in the non-Scoring Chance area within the offensive zone (OZ) and 241 were generated in transition (D/NZ). The likelihood of sequences generated in these categories (OZ and D/NZ) resulting in a goal are 1.8% and 3.1%. So, just over three-quarters of opposition offense is relatively useless and wasteful. Conversely, of the 861 attempts generated by the Devils, only 68% were of the weak variety. What does this tell us? It means the Devils lack scoring talent. Lots and lots of scoring talent. How do we know this? Take a look below.
What we see here is how the Devils stack up to the rest of the data we have in terms of sequence shooting percentage by category. The league converts all Royal Road attempts on a 17.6% clip, compared to only 7.0% for the Devils. You’ll see that the Devils are worse off in almost every category. But, that’s what happens when you pay $3.5 million to watch Michael Ryder hit posts every night, or watch Patrik Elias continue to decline and whiff on wide open chances, or witness Jordin Tootoo on a top line. I could go on, but why bother?
Let’s get back to the defensemen. What we know now is that while Marek Zidlicky generated lots of offense individually, his defensive woes allowed the other team to generate nearly as many quality possessions while he was on the ice than any other Devils defensemen. Well, we probably always knew that, but now we can quantify it. In the end, his offense barely outweighed his liabilities in defense, but just barely. Peter Harrold and Mark Fraser were the worst, which isn't surprising. Harrold actually had more quantity than quality, which is a rare feat. He should be jettisoned from the roster immediately. What’s intriguing is that Jon Merrill led the blue line at 54.7%, Damon Severson was right behind him at 53.9%, and Adam Larsson was third with a 51.1%.
Now, in between performing larger studies with multiple teams’ data as my trackers finish up, I’ll be doing video work on various players and these events to get a better understanding of who exactly we should applaud and blame for what occurs on the ice. Merrill seems like a logical first choice. Though, when I looked at the defensemen last season, Merrill's underlying metrics looked very good here and here. Perhaps we can't properly measure defensemen that well with current data? After all, evaluating defense is trying to measure the absence of something (opposition shot attempts), so it may be a while before we properly evaluate them. If you'll permit me a soccer example: when evaluating defensive players one of the metrics often looked at is how many tackles a player made and how successful they were, but some players read the game so well that they don't even need to tackle because of their awareness and positioning--the ball is never ever passed into their area or they intercept it. A few other numbers stood out when we delve deeper in what sequences they are on the ice for. Have a look.
Well, that certainly explains some of the value that Merrill is on the ice for. Now, in terms of direct involvement in those twenty-nine attempts, Merrill has attempt two shots himself, generated two from primary passes, and two more from secondary passes. But, he has only allowed nine while on the ice, so I’ll be taking a look at some of these and compare them to other defensemen. While defenders may not have much to do with setting up Royal Road events, they certainly have a lot to say about defending them.
Weighted Possession of Devils Forwards
Moving to the forwards, I included players with at least 200 minutes to include Jagr and Havlat, both of whom were just on the edge of that threshold.
What’s interesting is cases like Tuomo Ruutu and Jacob Josefson. Both players had nearly identical SAT% (49.4 for Ruutu and 49.3 for Josefson), as well as identical WPSAT%. For more evidence of this, look at the original top line of the 2014 – 2015 Devils: Jaromir Jagr, Mike Cammalleri, and Travis Zajac. All of these players had significant jumps in quality possession compared to their standard possession numbers. That makes a lot of sense. Tootoo’s numbers are evident of playing with top players.
Elias’ numbers don’t improve as much as others, as he equates to Scott Gomez and Steve Bernier. Considerable woe unto Dainius Zubrus as he literally added the least quality while on the ice. The notion that he is elite defensively needs to be put to bed. Adam Henrique looks about average, so not only do we need to look at why Merrill looks so good, but also why Henrique doesn't look quite as good. So, John and I will likely start with Merrill and Henrique. If any readers and other ILWT writers wish to help out with this project, don't hesitate to let me know.
So, Zajac gives up a lot, but also is on the ice for the most Royal Road attempts for. Here we also see that the opposition is able to set up Royal Road chances with Henrique on the ice more so than any other forward (Henrique’s RR A/60 is 1.81, worst among forwards). Ruutu’s -5 Royal Road differential explains while his quality possession barely moves. Zubrus has a poor differential as well (-6).
Concluding Thoughts and Summer Plans
So, as I mentioned, we'll be going through the video for Merrill and Henrique to start as they represent the most obvious candidates for further analysis. When we can attribute a quantifiable value to each sequence of possession, we can get a better idea of which players create better chances and defend against better chances. Think of it like a football game. SAT and Corsi essentially measure how many plays a team ran. They weigh ten three-yard runs the same as ten fifteen-yard passes. However, we know that if you’re moving the ball fifteen yards at a time as opposed to three, that’s usually the better play. That’s what I’m doing here: separating the noise (shots from outside the scoring chance area and those in transition for the most part) from the quality sequences and weighing them appropriately.
Put another way, do we care if a player gets "shelled" in Corsi if most of the attempts he gives up are weak and from far away? Wouldn't we be glad if that player then was able to create a few, high-quality chances going the other way? To use another sport for comparison, this is similar to soccer teams defending well and then attacking on the counter: absorbing pressure and maintain discipline to defend against a more skilled opponent and then taking your chances going forward. This sums up the Devils, doesn't it? Structured defense to soak up pressure and then only get a handful of shots going forward. The trouble is that they simply didn't have the forward talent to execute effectively when going forward.
Anyways, that’s all for now. Sound off below and we'll be back with some video on Merrill and Henrique.