Corsi and Score Effects
by Hawerchuk on Apr 13, 2010 9:00 AM EDT
Email
PrintI'm actually going to look at what I believe are called "Fenwick" numbers today - total shots directed at goal excluding blocked shots. Because of rink bias, Fenwick numbers actually correlate better with winning than Corsi numbers do. First, a little rationale for why we want to look at Fenwick and Corsi numbers. This table shows how many playoff and non-playoff teams exceeded 50% Fenwick ratios on the road, at even-strength, with the game tied:
| Spots | Teams | Make Ply | |
| Playoff | 112 | 51 | 45.5% |
| Non-Playoff | 98 | 8 | 8.2% |
These numbers are for 2001-2009. For the years where missed shots are missing, I just used shots on goal. It's clear to me - and hopefully clear to you - that playoff-bound teams are much more likely to dominate the shot counts in the clutch minutes of their games. It's not a guarantee, but it certainly was the path Detroit and New Jersey used to succeed over the last decade, and a low percentage is certainly a red flag for any team with cup aspirations.
Here are the overall (home and away) Fenwick ratios for the playoff-bound teams in 2009-10 by lead:
| Team | -3 | -2 | 1 | 0 | 1 | 2 | 3 |
| chi | 63.0 | 70.4 | 63.2 | 58.6 | 53.9 | 55.0 | 54.8 |
| det | 55.4 | 60.5 | 54.2 | 54.0 | 48.8 | 50.2 | 39.6 |
| pit | 61.8 | 54.3 | 57.4 | 53.7 | 50.0 | 46.7 | 40.5 |
| bos | 48.3 | 53.3 | 55.5 | 53.6 | 49.9 | 46.5 | 48.3 |
| pho | 58.7 | 58.0 | 52.0 | 52.1 | 48.9 | 43.6 | 49.4 |
| nj | 55.5 | 60.3 | 52.2 | 52.1 | 49.2 | 48.2 | 51.5 |
| sj | 56.5 | 59.6 | 54.2 | 51.6 | 45.9 | 43.1 | 49.2 |
| nsh | 55.7 | 54.9 | 56.3 | 51.6 | 48.3 | 46.5 | 36.4 |
| phi | 59.8 | 57.8 | 55.4 | 51.6 | 49.9 | 47.4 | 46.3 |
| ott | 53.7 | 55.0 | 57.4 | 51.2 | 49.1 | 48.8 | 52.0 |
| was | 61.6 | 59.2 | 58.2 | 50.9 | 46.8 | 49.9 | 46.9 |
| la | 56.1 | 62.9 | 56.9 | 49.8 | 47.3 | 41.5 | 39.5 |
| buf | 46.5 | 55.6 | 53.8 | 49.6 | 47.5 | 44.4 | 47.3 |
| van | 61.5 | 66.4 | 55.9 | 49.2 | 46.1 | 42.3 | 42.5 |
| mon | 50.2 | 57.1 | 49.2 | 46.7 | 43.5 | 42.7 | 52.7 |
| col | 59.7 | 54.4 | 52.8 | 44.5 | 43.0 | 39.3 | 36.1 |
That's a lot to chew on, but there are two takeaways:
1) Chicago is dominant
2) Montreal and Colorado are in big trouble
One issue with this table is that because playoff teams have tended to win games, they're in the lead and tend to shoot a bit less than they would if they'd won fewer games. We can correct the total for score effects:
| Team | Actl | Corr | SE |
| chi | 58.3 | 59.3 | 1.0 |
| was | 51.3 | 52.2 | 0.9 |
| sj | 50.6 | 51.1 | 0.5 |
| col | 45.8 | 46.3 | 0.5 |
| nsh | 51.2 | 51.5 | 0.3 |
| pho | 51.2 | 51.4 | 0.2 |
| nj | 52.0 | 52.1 | 0.1 |
| pit | 52.8 | 52.9 | 0.1 |
| van | 50.9 | 51.0 | 0.1 |
| phi | 52.2 | 52.3 | 0.1 |
| det | 53.0 | 53.0 | 0.0 |
| buf | 50.0 | 50.0 | 0.0 |
| bos | 52.5 | 52.4 | -0.1 |
| la | 51.2 | 51.0 | -0.2 |
| ott | 52.1 | 51.9 | -0.2 |
| mon | 47.4 | 47.2 | -0.2 |
So the "true" Fenwick ratio for playoff teams is, for the most part, higher than the actual ratio. Similarly, teams with poor records needed to get adjusted in the other direction:
| Team | Actl | Corr | SE |
| cgy | 50.9 | 51.4 | 0.5 |
| nyr | 49.5 | 49.8 | 0.3 |
| ana | 47.2 | 47.5 | 0.3 |
| stl | 50.1 | 50.2 | 0.1 |
| fla | 45.8 | 45.7 | -0.1 |
| nyi | 48.1 | 48.0 | -0.1 |
| cls | 46.9 | 46.8 | -0.1 |
| tb | 47.3 | 47.1 | -0.2 |
| car | 47.7 | 47.5 | -0.2 |
| dal | 48.4 | 48.1 | -0.3 |
| atl | 48.0 | 47.5 | -0.5 |
| min | 48.9 | 48.3 | -0.6 |
| edm | 45.4 | 44.6 | -0.8 |
| tor | 53.0 | 52.1 | -0.9 |
Incidentally, for those who still don't believe even-strength shot ratio is important, note that only two teams with a ratio below 50% made the playoffs, while only three teams above 50% missed the playoffs. Two of those teams were Calgary and St. Louis, both of which would have easily made the playoffs if they were in the Eastern Conference.
The only real outlier in our entire sample is the Toronto Maple Leafs, who, for most of the season, had a goaltender who was so bad that they couldn't afford to stop being aggressive even when they had the lead. Or something like that - Toronto, for some reason, outshot their opponents when they had a two-goal lead on the road. I can't come up with a reason why they played like that, but nothing is rational in Toronto these days.
19 comments
|
0 recs |
Do you like this story?
Comments
Is there any metric that takes shot quality into account? For instance, weighted based on the distance from the goal. One or two trigger happy defensemen on a team can generate a lot of shots but they are extremely poor quality shots if they’re from the blueline at even strength. Ideally, that shot would be weighted differently from a shot 3 feet away from the crease. I’m not sure if this is widely available or not.
by ThrashersRecaps on Apr 13, 2010 10:15 AM EDT reply actions
Shot quality is really something tiny, around the edges, plus (for the most part), it’s not repeatable.
I doubt it would do much, but I suppose an easy way to measure shot quality would be:
a) shots taking by above average shooters (above average shooting %)
b) shot ratio by forwards to defenceman
Again, I don’t know if these would bear any fruit. Well, logically point a should. Logcially, a shot taken by Ovechkin, Stamkos or Crosby is more dangerous than one taken by Jason Strudwick.
Hockey blogging can't get any flatter.
In general, shots by forwards are 20% more dangerous than defensemen. But most shots are taken by forwards, and most shots are taken by good-shooting forwards.
It’s not that shot quality is unimportant – it’s just that with the data we have (location, rebound, shooter ID), it’s just inconsequential compared to shot volume over the course of a season. And it does not appear to be much of a true talent at the team level. Vic Ferrari and Likens’ posts on how much of shooting is luck are very good on this issue.
Where does that 20% come from? Location? Or is that controlling for location?
by DoctorMyBrainHurts on Apr 13, 2010 11:23 PM EDT up reply actions
This gets at a question I’ve had for a while. I realize what you are saying that most of the effect of shot quality can be drowned out over large samples and ignored. I also realize that part of the problem is that distance is the proxy for quality, and that these data are considered pretty shaky. But, given two teams (or more) comparable teams in terms of corsi (or Fenwick), has anyone checked to see if a difference in winning (b/w those specific teams) can be shown to be partially correlated to a difference in quality? It might get a finer grained analysis.
Looking backwards, yes, it can explain some differences. But looking forward, shot quality isn’t really a skill that teams seem to have, so their core shot total is more important.
Okay, if I got you right: you’re saying that a hypothetical GM who wanted to evaluate players for their effect on winning would not be able to take previous shot quality (proxied as distance) as a repeatable for future shot quality?
It’s because the Leafs were such an effective shutdown team that their opponents, facing a two goal deficit, simply gave up and stopped trying.
I've been looking at the sky
Would the players you are giving the territorial edge to have a significant effect on results? For example, Vancouver, SJ and LA perform relatively poorly when the score is tied but are both relatively strong when they are trailing. Looking at their rosters, it seems that most of that is given up by their fourth liners mostly against other fourth liners while their top-9 players are generally Corsi positive. Would that have a tangible effect on the number of scoring chances or the expected goal differential (based on a lower expect SH%)?
Nice work here, Gabe.
A question:
Where did you did find EV shot data from 2001-02 and 2002-03? I remember trying to scrape the data from the play by play sheets on NHL.com a while back, but wasn’t able to because too many individual games were missing.
Fascinating stuff, Gabe. I’ve been keen to see information of this type for quite some time. Thank you for this.
My biggest concern with Corsi/Fenwick and shots data generally has been w.r.t. score effects, but when you sort the data by score differential in the manner you have done, the relationship is clear enough.
Because of rink bias, Fenwick numbers actually correlate better with winning than Corsi numbers do.
That’s interesting. Is the rink bias that much bigger for blocked shots than missed shots (or shots on goal for that matter)? How much worse is the correlation to winning for Corsi than for Fenwick?
For the years where missed shots are missing, I just used shots on goal.
How much better do Fenwick numbers correlate to winning than shots on goal? How much more do we learn from the extra info of missed shots? I’m particularly particularly cuz in the past we only have shots data, so the Fenwick option doesn’t exist when analyzing games from previous eras.
Writer for The Copper & Blue and primary shareholder of Zorg Industries
"Never be ashamed of who you are" -- Jean-Baptiste Emanuel Zorg
Comments For This Post Are Closed
Secondary Sidebar
User Tools
FanPosts
Community blog posts and discussion.
