Taking a Shot: What Matters and Who Should Take It?
In the Taking a Shot series, we’ve been exploring shot attempts and what "funnels" can stop them: blocks, misses and goalie stops. The first few parts looked at variations by team, then by player focusing on the Preds, then those players against the whole league. All analysis has been based upon a 2 million row play-by-play database for the 2016-2017 season. I then process all of that into a form for these posts.
If you want to see the entire resulting spreadsheet of every player in the NHL who met the criteria, it’s here as a google spreadsheet. Yay, free data! *runs around tossing gold confetti*
You get free data!
And you get free data!
In the last post, I looked at shot attempts and their obstacles abstractly. Now we want to know: Which funnels matter? Which ones are key to actually getting goals? And does it tell us what sort of player should be shooting?
When I started this post, I thought I’d just quickly toss some statistical models at the data, see what was important, report it, and be done. Just tossing all the data, however, told me a story that was… wrong. Never throw math at data you don’t understand, because you could be mathing the wrong things.
To make all this concrete, let’s just start examining patterns. What we’ll do is just look at shot attempts, blocks, misses, and stops against goals. As one goes up, do goals go up?
Figure 1. Corsi (Shot Attempts) per 60 minutes of play vs Goals for all players
Figure 1 looks pretty straightforward. The more you shoot (the x-axis), the more you score (the y-axis). A goal isn’t guaranteed, of course. You can find players who shot more than 10 times per 60 minutes and never scored and players who shot more than 10 times and scored 15 times. Those are the vertical lines of dots in the data. Similarly, looking horizontally, you can find people who had to shoot twice as often as others for the same goal total. That poor fellow who shot more than anyone else in the league, per 60, but only scored 9 goals (far right)? That’s Montreal’s Brendan Gallagher. One dot to the left is Norris trophy winner Brent Burns with 20 goals. He shot a LOT! We’re going to come back to him.
One further nuance is revealed by the blue line on top of the data. It’s not exactly a line. It bends up a little steeper around the 11 C60 mark. I put a dashed line there to mark it.
Regardless, a clear pattern is present. The more you shoot, the more you score. Let’s move on to the funnels: Blocks, Misses, and Stops.
Figure 2. Proportion of Blocked shots vs Goals. Forwards in blue; Defense in Red
Ignore the colors for a second. If you just look at the dots, it kind of looks like there’s a trend down in goals as the block percentage goes up, which makes sense. But it’s more like two groups of data, one high and one low, and that becomes clear when you code the data for position. The blue-green points are forwards; the red ones are defensemen. Shots from the defense are blocked far more often than those from the forwards. They function as two very different groups with little intermixing.
Indeed, once we see that there are two groups, the very large trend down from left to right is actually two slightly down patterns. The fact that both groups trend down in about the same way indicates that blocks do indeed predict goals.
However, if there are clear groups for blocks, maybe we should look at Corsi again. So here is Corsi against Goals with Forwards and Defence marked separately (Figure 3).
Figure 3. Corsi per 60 vs Goals grouped by Offense (blue) and Defense (red)
Now it’s clear that defense scores a lot less that offense, no shock, but also that, while shooting more increases goals for both groups, it has an even bigger effect on forwards. The “curve” marked in Figure 1 turns out to just be the position of the players and where forwards start to dominate.
Showing defense separately lets us also see just how exceptional Brent Burns’ season was. His dot is that red one on the far far right at 20 goals with no other defensemen anywhere near him. To get that many goals, Burns had to shoot 2nd most in the league. The next closest defenseman shoots almost 10 shots less per 60 and scored about half as many goals. Could those shot attempts Burns took have gone to someone else who might have put them in at a higher rate than a defenseman can? Looking back up to Figure 2, Burns’s shots are blocked almost 40% of the time, average among the defense.
Moving to Misses, we see a similar pattern to Blocks. First, look at all players.
Figure 4. Misses vs Goals for all players
This looks like a strong downtrending pattern, where the more you miss, the fewer goals you score. One interesting landmark is that little group of dots at the top right where they missed a lot but still scored a lot. One of those is Forsberg who missed more than any other forward on the team.
However, be careful of the line on this figure and what it shows. There’s a lot of data that’s quite far away from the regression line. That’s a huge amount of noise not represented by the regression line. A better picture is again provided by separating offense and defense.
Figure 5. Misses vs Goals grouped by Offense (blue) and Defense (red)
Now we can see that Misses make only a slight difference for defense. The line is almost horizontal. It does trend down for the forwards, but there’s still a lot of variation. Finding Brent Burns again, the only red dot above 11 goals, we again see that not only are his shots blocked a lot, but they miss a lot. He’s near the far right even among defensemen and you don’t want to be there.
And finally… something very very clear. No nuances with Stops (the goalie stopping the puck). It has a huge effect on everyone.
Figure 6. Stops vs Goals grouped by Offense (blue) and Defense (red)
If all patterns were as clear as this, it would be summertime and the living easy for hockey analysts. The less you’re stopped, the more goals you score, and the lines are far steeper than anything else shown so far. The steepness of the line shows you the power of the relationship. What we’re basically asking is, “if I know the value on the x-axis, can I better predict the y-axis (and vice versa)?” If the line is not horizontal, the odds are yes. The more steep, the more yes.
Let’s look at some specific dots on Figure 6.
The dot on the far left, away from everyone else? That’s the Capitals’ Oshie. That dot in the middle at the top with more goals than anyone else? That’s Auston “the other Watson” Matthews.
The fact that Oshie is so alone suggests that percentage is going to come falling back closer to the group soon -- unless you have some reason to believe Oshie’s shots are categorically better than anyone else’s. It might be dangerous to pay him solely for being an outlier like that. Matthews on the other hand is an outlier in scoring lots of goals, but his shooting percentage is entirely in line with other excellent shooters. A safer bet.
Picking on Brent Burns again, he does indeed have one of the best Stop Proportions among defensemen; however, it’s still average among forwards. All of this indicates that you ideally do not want your defense leading in shot attempts. Stops and blocks are both worse for defensemen. Brent Burns scored 20 goals shooting, almost double any other defensemen in the league, and it earned him a Norris. But to do so, he had to shoot more often than almost anyone in the league. 20 goals is also typical for a solid top 6 forward, not something to earn a trophy for.
If that rate of shots from Burns takes away higher percentage shots from the forwards, it’s not as great as it seems at first. “Give it to Brent and it’s heaven sent” (as far as I’m aware, no one says this; I made it up just now.) has the potential to lower the team’s overall offense. Indeed, while San Jose was 6th in the league in shot attempts, they came out 2nd worst in the entire league in the Taking Care of Business metric, which was a weighted sum of all the funnels. They were 6th in the league at shooting, but bottom third in actual goals.
I’d like to argue then that the best way to get the defense involved in offense is passing to forwards, with only periodic shots from the blue line. This argument is pretty weak so far, though, so let’s study it directly.
I calculated the proportion of shot attempts taken by the defense out of all shot attempts (Defense Corsi / All Corsi) for each team. I then compared it to the proportion of shot attempts that actually went in (Goals / Corsi). There’s an obvious trend (Figure 7).
Figure 7. Defense Proportion of Corsi vs Goals over Corsi
The more the defense takes over shooting, the fewer goals the team scores. The pattern is also significant statistically (ANOVA, p = 0.02).
If so, you can actually make the argument that Burns’ goals might have suppressed San Jose’s offensive potential. Burns scored so many goals (20) by shooting 545 times (even strength only), which is a goal rate of 3.7%. The average San Jose forward’s goal rate was 5.4%. This isn’t a huge difference, but if the average Shark forward was shooting Burns’ shots rather than him, San Jose would have scored 9 more goals. The average forward shot 160 times compared to Burns’ 545 attempts. He was dominating their offense. Burn’s attempts were more than 3 average forwards together. If their best forwards were shooting (people like Pavelski, Couture, and Marleau) the team would have scored 15 more goals: 35 goals versus Burns’ 20. (This is based off the average of the best 6 forwards.)
Now, Burns actually scored more than 20 goals. He put in 8 more on the power play. Terrific! However, San Jose’s power play ranked 25th out of 30. I didn’t see enough Sharks games to know, but if the power play was oriented around getting it to Burns, it could have been limiting the team’s scoring potential. The best defenseman scores at the rate of a fine top 6 forward, not at an elite rate.
If there was an over-reliance on Burns, it’s of course not Burns’ fault for being so good. It’s the coach’s / team’s fault for falling into that tempting strategic trap. Burns had the best Goal / Corsi rate of the Sharks’ defense, so he stands out, but there were 9 forwards with higher rates. One doesn’t want to purposefully give the shot to the 10th best scorer on your team, even if he’s exceptionally good for his position.
Some giant caveats for this argument must be stated:
- If shot attempts from the blue line don’t go in, but do set up juicy rebounds, that’s valuable.
- If shot attempts from the blue line open up space for forwards because the other team must respect their shot, that’s valuable.
- If it’s not a zero sum game, so that shots from the defense are NOT taking shots from the offense, then my argument doesn’t work either. However, the fact that Burns shot more than 3 forwards combined suggests it did happen for the Sharks.
What I really think the data show is not that the defense should not shoot. Rather, the team’s strategy should rarely be based on setting up the defense to take the shot. There are other reasons to take a shot from the blue line. However, forwards passing up a decent shot to set up the blueliner, even when they are a stud blueliner, could be shooting yourself in the foot. The shots are just much lower in probability.
Apart from the defense vs offense question, can we quantify how important each funnel is? From the plots, we can see that all the funnels have some importance, except for Misses for defense, where the line was almost horizontal.
To make it precise I ran regression models for offense separately from defense.
For forwards, all funnels and Corsi are highly significant, but their power varies considerably. Stops are 4 times as important as any other funnel. Surprisingly, blocks are as important as shot attempts. This means that being blocked less is just as important as shooting more. Misses are half as important as either blocks or shot attempts. If one has to choose a strategy to maximize success at one of these funnels, then you want to concentrate on the goalie, then blocks, then more shots, and then finally the misses. In fact the goalie is 8 times as important as the player missing the net.
Statistical analysis for Defense is basically the same except that the effect of misses is even smaller and only bordering on significance.
In the end, high danger shots are the key.
When the Taking a Shot series returns, I will look at more data on what makes shots more dangerous (angle, distance, etc.), but the series is going to take a break for a bit. Variety is good. Instead, I will start a new series concentrating on patterns over time within games. Are there patterns regarding when things happen? That’s a real question. I don’t even know yet. A-hunting we will go.