If you've been keeping score of my passing articles this is the sixth one (and it's by far the best.) Our new way of tracking, noting the shooter and time, has opened up a new world of ways to analyze the data and I believe we're onto something special. Again, I'd like to thank Ryan Stimson for starting and running the passing project.
If you don't buy that possession is the end-all, be-all measure to evaluate teams and players, because taking more shots don't necessarily mean they're dangerous shots, you're in luck. With this thought in mind, I believe I created a better metric to evaluate player and team performance, which you can read more about below. When restricting hockey to more controlled play, we see that the correlation with Goal For% increases from 0.23 (SAT% Close) to 0.37 (Weighted SAT% Close). Note this is for 2015 NY Islanders games only.
Before this post gets too long, let's jump into it.
Recap: What we track
We track all shot attempts (shots on goal, missed shots, and blocked shots) derived from passes. This means rebounds, attempts off the faceoff, and defensive giveaways are not captured as we aim to measure repeatable skills. On any given attempt, we track whether it was in transition (defensive/neutral zone), whether there was a secondary passer in addition to primary, if it was a scoring chance, one-timer, or crossed the Royal Road. Click here to see what a SC, One-Timer, Royal Road attempt looks like.
There are 11 significant categories of shot attempts that come from permutations of these categories (below).
After applying those permutations to the 2,900+ SAT's the Isles and their opponents have taken since January 4th (44 GP), I found the expected shooting rates of each:
So what's the big deal you ask? Well, using the NHL's 'On Ice' data from their RTSS sheets I was able to extract who was on the ice for the Isles for those events and after applying weights to each, come up with a Weighted SAT% for each player, among other stats.
Why It's Different
I know there seems to be a new stat that you must pay attention to everyday, but this is different because:
- The weights are not arbitrary. The weight calculations are all fluid in the workbook and will only get better with more data. (The numbers you see above are based on roughly 55% of the 2014-15 Islanders season.)
- It's measured in goals making it simple to understand.
- This sheds more light into the quantity vs. quality debate of anything you'll find on the Internet. While SAT% does come close to approximating Weighted SAT%, the market inefficiencies are in the difference.
- These charts 1, charts 2 are pretty cool too.
I'll refrain from adding my own commentary to the below table, and instead offer a way to understand it so you can form your own conclusions. First and foremost, the below numbers are on a per 60 basis so there are no time on ice effects. The 'Weighted SAT For' column is the weighted shot attempts for the Islanders while the player is on the ice. It's just like SAT For/60 except the attempts are weighted and restricted to those generated from passes.
Likewise the 'Weighted SAT Against' column are the opponent's weighted attempts while the player is on the ice. Here we can see that John Tavares and Josh Bailey were the best at driving offense, while Cal Clutterbuck and Nikolay Kulemin were the best at suppressing shots.
The Weighted SAT% column is simply Weighted SAT For / (Weighted SAT For + Weighted SAT Against), think of it as being similar to SAT or Corsi%. Not shown but still important is the Weighted SAT% Off column which is the Weighted SAT% of the team while the player is not on the ice. Finally the Weighted SAT Rel column is the difference between these two. Basically, a player who is positive in this column gives the Isles an edge in shot quality while he is on the ice.
The rightmost table shows the predicted per 60 stats of each player based on the quantity & quality of their shot attempts and passes. These correlated with actual goals scored with 0.857 correlation and actual assists at 0.726. It also passes the smell test as it claims Anders Lee was the best goal scorer, and Tavares, Ryan Strome, and Bailey were the best passers.
The below table is just like the one above except for the defense. While Lubomir Visnovsky's sample size may be small, he was the best defender on the team, in terms of offense and defense, in his last 25 games of the regular season.
Also of note is that Johnny Boychuk should have received Norris consideration over Nick Leddy. He is far and away the best defender on the team when taking into consideration strength of competition, and was the third best offensive driver on the team. Praise Bossy that Leddy and Boychuk signed those 7-year deals, and cross your fingers that Garth gets one to two more years out of Lubo.
Also Brian Strait is not very good, and those saying that Thomas Hickey is over-matched may have a point.
While the tables below may not be intuitively easy to understand, they do tell an important story. The first table is the final 44 games of the Islanders season. The second is games 1-22 of that time span, and the third is games 23-44.
Here we see throughout the season the Islanders were relatively equal in giving and taking harmless attempts (1st column in right tables). Harmless attempts are those that are not scoring chances, one-timers, or Royal Road crossers. The Isles and their opponents each took 879 of these attempts in stunning parity. A similar story is found for those one-timed attempts that were not scoring chances or Royal Road crossers (second column) which are also relatively harmless.
Where we really see differences are in the scoring chance categories. The Islanders out-chanced their opponents to the tune of an expected 3 goals over the final 44 games of the season (third column). While this may not seem like a lot, a scoring chance attempt has an 11.3% chance of scoring, which means the Isles must have had roughly 26 more scoring chance opportunities from passes than their opponents. Indeed, this is the case as the Isles led 186 to 160 in this category. However, the Isles got pummeled in the final 21 games of the season in the SC, One-Timer, Royal Road category, giving up an expected three more goals than their opponents. I thought that's why Strait was in there!
If you're still skeptical of the importance of Weighted SAT, consider the following:
- iWeighted SAT predicted goal scoring per 60 better than shots on goal (0.86 to 0.80 correlation) for the Islanders.
- Weighted Assists gives us an idea of the numbers of assists we can expect a player to generate (0.73 correlation). Going by passes alone gives only a 0.61 correlation.
- Weighted SAT was comparable to SAT regarding correlation with Goals For/60 (0.72 to 0.67)...
- ...but blew it out of the water when it came Goals Against/60 (0.46 to -0.14) and Goals For% (0.42 to 0.01).
NJ Devils and Other Data
Finally, as Ryan made the Devils and Isles passing data public about three weeks ago, we can see how the players stack up against each other. Click here to see how they compare.
Note that both tables are sorted by Expected Points/60 then by Weighted SAT% where players over 50% are green. I sure wouldn't mind taking a flyer on Scott Gomez or replacing Strait with Jon Merrill. Once the data from the other teams are released, I will look to see if they correlate in a similar fashion.
Weighted SAT% WOWY
Editor's note: As of June 11, 2015, the most requested item to be analyzed was Weighted SAT WOWY analysis. Please see this article for more information.