Last week we looked at each of the Winnipeg Jets skaters and their impact on shot location. We also looked at the save percentage in different regions in the defensive zone for each of the Jets’ goaltenders.
These articles show that the Jets skaters have experienced different distributions in shot location, both defensively and offensively. Does it matter though? If so, by how much?
We investigate how player evaluations change when adding shot location to shot differentials.
Why shot location?
Shot-attempt differentials -Corsi and Fenwick- give a pretty solid understanding on which players are improving their team's goal differentials through shot volume talent. What about shot quality talent though?
Talent in shot quality exists. We know this list has Sidney Crosby at the top and Zenon Konopka near the bottom for a reason.
Existing and having a major impact is two different factors though. For one, there is a lot of noise (variance and luck) that masks the signal (talent) in small sample percentages.
Hockey today is a game of bounces. Dirty goals scored after bouncing off of three players or an unintentional rebound to a wide-open Chris Thorburn on one knee can lead to goals that are extremely unlikely to happen again even if the player wanted it to. It is why we observe about 66 percent regression over team level shooting percentage over a full season.
Luck and bounces in shooting percentages are not concepts understood solely by hockey-numbers types, as Adam Lowry showed earlier this month:
I think we knew sticking with our system, and early on we had been generating a lot of chances and our shooting percentage wasn't necessarily where it would be, and some guys were just missing some opportunities that they normally don't miss and I think, as of late, those chances are starting to go in as well as our power play is starting to produce.
There is some indication that shot location is the major driver of sustainable, non-luck shooting percentage. By isolating shot location, we are separating a large chunk of the signal in shot quality from the noise.
There are other factors such as rebounds, screens, deflections, and cross-ice passes, but we will take shot quality one step at a time.
But still, existing does not mean the factors play a major role. The extent shot quality talent impacts goal differentials in addition to shot quantity also depends on the distribution of talent. If the distribution in talent for quality is small relative to quantity, then the impact factor for quality will be less.
Also, talent in hockey usually is not extremely specialized. Finishing talent, defending talent, and shot differential talent are not mutually exclusive skills. If there is a strong relationship with those that are talented in quality versus quantity, then some of the quality factor is being accounted for in quantity.
What about the Jets?
The shot location regions in the previous articles were separated by scoring success probability. The three locals contained a nine, four, and two percent chance of a shot being a goal scored. Using those percentages as weighting and each player's shot differentials, we can then estimate each player's average expected goal differential for 2011-15.
| Player | Exp GD/60 | Player | CD/60 | |
| Mathieu Perreault | 0.020 | Andrew Ladd | 7.8 | |
| Andrew Ladd | 0.019 | Mathieu Perreault | 7.4 | |
| Blake Wheeler | 0.016 | Michael Frolik | 6.4 | |
| Paul Postma | 0.016 | Bryan Little | 5.8 | |
| Alexander Burmistrov | 0.012 | Paul Postma | 5.1 | |
| Bryan Little | 0.011 | Blake Wheeler | 4.7 | |
| Tobias Enstrom | 0.009 | Alexander Burmistrov | 4.7 | |
| Michael Frolik | 0.007 | Tobias Enstrom | 3.9 | |
| Evander Kane | 0.006 | Dustin Byfuglien | 3.2 | |
| Dustin Byfuglien | 0.005 | Adam Pardy | 3.2 | |
| Adam Pardy | 0.003 | Evander Kane | 3.0 | |
| T.J. Galiardi | -0.001 | Mark Scheifele | 0.6 | |
| Mark Scheifele | -0.001 | Grant Clitsome | 0.6 | |
| Grant Clitsome | -0.002 | Jacob Trouba | -0.5 | |
| Zach Bogosian | -0.007 | T.J. Galiardi | -0.9 | |
| Jacob Trouba | -0.010 | Zach Bogosian | -1.7 | |
| Mark Stuart | -0.016 | Mark Stuart | -3.2 | |
| Matt Halischuk | -0.019 | Matt Halischuk | -10.8 | |
| Chris Thorburn | -0.032 | Chris Thorburn | -11.7 | |
| Jim Slater | -0.046 | Jim Slater | -13.9 |
Now keep in mind that none of these numbers have been adjusted for player usage which would factor into these differentials.
From an observational level, the order looks fairly similar in both without any major movement. At a surface level, shot location does not seem to add a substantial amount of information that was not already being relayed by looking strictly at shot volume.
The statistical tests match the previous observations. There is a statistically significant correlation between the two as shown by the 0.96 correlation coefficient. The R-squared tells us that the purely shot quantity model explains 92.5 percent of the variation in the model that includes shot location.
The studentized residuals show only Matt Halischuk being a substantial outlier, with Corsi differentials underestimating his shot location impact. Michael Frolik is almost on the threshold and Jim Slater sits on the line for being outliers with Corsi differentials overestimating their impact. Otherwise there is not much difference for players.
This analysis does not prove whether or not implementing shot location into a model improves it. What this analysis does say is that adding shot location does not add much in terms of new information… at least for these Jets.
