As usual with these sorts of things, they should get better as the season goes on and we get more data.
A handful of teams jumped out, that I'd like to comment on:
- Colorado has put up decent, much-improved possession numbers at even strength. In special teams they have been above average getting pucks toward goal on the PP and very strong suppressing shots on the PK. They have not gotten the goaltending or bounces at either end. I think this ranking overrates them, but a deeper look points to them being better than I initially thought. Perhaps mediocre is more accurate than bad.
- The Jets are another team that this ranking puts substantially above where they are in the standings. They have put up better than average possession numbers but whatever combination you like of bad bounces in their own zone and bad goaltending has done them in. They have also haven't been disciplined, ranking dead last in the league at times shorthanded. If the team hadn't moved, this would be familiar to Jets fans since they had a similar pattern in the second half of last season.
- They won't win the rest, but I think we can expect the Canucks to continue their rise and pull themselves up near the top of the standings.
- The Rangers are in trouble. They've had an unsustainably high save percentage and a ridiculous shooting percentage at even strength (9.9%), both of which are almost certain to drop. They are getting dominated 5-on-5 and are fourth worst in the league in shooting rate 5-on-4 (BTN). I don't see them making the playoffs.
Here is the ranking, sorted by Corsi%
| Rank | Team | Corsi% | Fenwick | Fen Rank |
| 1 | Detroit Red Wings | 57.9 | 57.6 | 1 |
| 2 | Vancouver Canucks | 56 | 55.1 | 3 |
| 3 | Pittsburgh Penguins | 55.3 | 54.8 | 4 |
| 4 | St. Louis Blues | 55.1 | 56.5 | 2 |
| 5 | Boston Bruins | 54.1 | 53.1 | 6 |
| 6 | Chicago Blackhawks | 53 | 54.4 | 5 |
| 7 | Colorado Avalanche | 52.8 | 52.4 | 8 |
| 8 | San Jose Sharks | 51.8 | 52.7 | 7 |
| 9 | Washington Capitals | 51.6 | 50.7 | 12 |
| 10 | Philadelphia Flyers | 51.4 | 51.6 | 10 |
| 11 | Winnipeg Jets | 51 | 51.8 | 9 |
| 12 | Florida Panthers | 50.8 | 50.2 | 13 |
| 13 | Montreal Canadiens | 50.7 | 51.5 | 11 |
| 14 | Ottawa Senators | 50.1 | 49.9 | 14 |
| 15 | Phoenix Coyotes | 49.7 | 48.9 | 18 |
| 16 | Los Angeles Kings | 49.7 | 48.4 | 19 |
| 17 | Columbus Blue Jackets | 49.5 | 49.6 | 16 |
| 18 | New Jersey Devils | 49.5 | 49.9 | 14 |
| 19 | Buffalo Sabres | 49.2 | 47.6 | 24 |
| 20 | Calgary Flames | 49.2 | 49.3 | 17 |
| 21 | Toronto Maple Leafs | 48.1 | 47.8 | 22 |
| 22 | Carolina Hurricanes | 47.9 | 47.7 | 23 |
| 23 | Edmonton Oilers | 47.4 | 48.1 | 21 |
| 24 | Dallas Stars | 47 | 48.4 | 19 |
| 25 | Tampa Bay Lightning | 46.3 | 46.5 | 25 |
| 26 | New York Islanders | 45.6 | 45.3 | 28 |
| 27 | Minnesota Wild | 45.1 | 46 | 26 |
| 28 | New York Rangers | 45.1 | 45.8 | 27 |
| 29 | Nashville Predators | 44.6 | 44.5 | 29 |
| 30 | Anaheim Ducks | 44.5 | 43.9 | 30 |
While I agree the Rangers are a bit of a paper tiger, it should be noted their Fenwick tied since the beginning of their standings surge (the SJ game) is 49.6, their Corsi is 48.0. Still not great, but at least closer to break even. PDO of 104.4 over that stretch, but who's counting really?
ReplyDeleteAnyway, I wouldn't mind seeing power rankings having more of a "what have you done lately" feel.
J,
ReplyDeleteFirst I think this is an awesome idea. It will be a great way to look at the strength of teams. I was wondering if you could publish your logit model, addressing why you choose specific variables. Im wondering if you choose varibles based on predictive value of future success, or for some other reason. also, I'm a little puzzled why you wouldn't include goal diff (ie. Pythagorean puck or something) in the model, as it is highly predictive (of course assuming your trying to predict future success). Looking forward to future articles though, this is great stuff
Thanks for the comments, guys.
ReplyDeleteGeorge - This is going to be a work in progress and adding a recency component is something on the agenda. After the first installment this won't be as bad because we'll be able to track movement in the standings. Triumph is about to do his monthly trends, which will touch on some of that.
Patrick - I have a variable for each team that takes on a value of 1 if they are the home team, -1 if they are the away team and 0 otherwise. There's a similar variable for five-on-four, five-on-three and four-on-three, which takes on 1 if the home team is on the relevant man advantage and -1 if the away team is. Finally, I have a similar variable for up one (1 if the home team is up 1, -1 if the home team is down 1), up 2 and up 3 or more. There's a constant term as well, which is necessary due to home-ice advantage. To avoid multicollinearity, I solve it with the constraint that all the team coefficients need to sum to 0. I thought about doing ordered logit and including all the no-shot time, but in the end just decided to go simple and only include Corsi or Fenwick events depending on which one it is.
As time goes on I'll definitely be tweaking the model, mostly to make it more predictive according to backtesting. The overall approach is based on the idea that including time and adjusting for the conditions should be better than just throwing out all info.
As for excluding GD, right now TBH I don't think it would help much at all. Late in the season I could see it becoming a significant factor even taking into account Corsi numbers. Like I said, I'll be putting in more work testing things and making changes as things go on. I could see it being the case that one model works well early on but another does better later in the season when there are more data to look at.
Thanks J,
ReplyDeleteI think I may have initially misread the top, going back over it seems to make a lot more sense, specifically when you account for score effects. Another confounder, aside from (I'm sure you addressed this) late game score effects when goalies are pulled would be zone%. Have you though about adding a variable for faceofff% by zone?
This makes me think of an overall "Game State Theory" concept I've been thinking a lot about lately. In which statistics are largely dictated by game state. The most important states being strength (eg. 5v5), score effects, time, and zone%. I think nearly every adv stat tracked right now (including Sh%) is highly game state dependent.