By The NumbersHockey Analytics... the Final Frontier. Explore strange new worlds, to seek out new algorithms, to boldly go where no one has gone before.
In 2008, I presented a goalie statistic called “Wins Added”. It attempted to record the number of wins a goalie earned, above and beyond those that an average goalie would have earned in the context of his team. I was able to improve this metric thanks to thoughtful comments from many people here.
I’ve calculated the statistic for the past four seasons. I’ve made a significant improvement over the previous version of the formula, as I’m now analyzing shots against by situation (ES, PP, SH) as opposed to simply looking at total shots against. The following table shows the save percentage during each situation:
Season
ES
PP
SH
WAvg
2009
91.9%
86.8%
90.5%
90.8%
2010
91.9%
87.4%
91.9%
91.1%
2011
92.1%
87.5%
91.2%
91.3%
2012
92.1%
87.5%
90.8%
91.4%
Unsurprisingly, power play shots are significantly more difficult to stop than even-strength shots. With the exception of PP shots in 2009, the ES and PP numbers have remained remarkably consistent over the past four years. The average save percentage on shorthanded shots (i.e. while the team taking the shot is shorthanded) has fluctuated significantly, however these shots are infrequent and the variation is probably due to randomness. The following table shows that relatively frequency of shots in each situation over the past four years:
Season
ES
PP
SH
Total
2009
77.3%
19.7%
3%
100%
2010
79.5%
17.6%
2.8%
100%
2011
80.5%
16.7%
2.8%
100%
2012
82.1%
15.3%
2.6%
100%
Although this should be obvious to anybody who has watched the NHL on a regular basis over the past several years, it’s still worth emphasizing. The number of power play opportunities continues to decrease. The average save percentage in the NHL has risen from 90.8% to 91.4% over the past four years, however half of that improvement is simply due to the fact that there are fewer power plays now. In other words, 0.3% of the increase in the league average save percentage is due to improved goaltending, and the other 0.3% is due to decreases in power play opportunities.
Step 1: calculate expected win percentage for a given team-year
Calculation: There is a very strong correlation (about 94%) between the Pythagorean formula, and actual win percentage. Thus, my estimate for how many games a goalie is expected to have won is based on that formula. To calculate it, we need the team’s goals for, excluding empty net goals, multiplied by the number of the goalie’s decision’s [A]. Next, we need the team’s shots against in each situation – even strength, powerplay, and shorthanded. Then, we calculate the number of goals an average goalie would have allowed in each situation [B]. For example, multiply the number of even strength shots faced by one minus the league average ES save percentage, repeat the calculation for PP and SH shots, and sum those three numbers. Plug the results into this formula: A^2 / [(A^2) + (B^2)]^2. That shows the expected win percentage based on how a goalie with a league-average save percentage would have performed, given the team’s level of offense and defense.
Example: in 2012, Miikka Kiprusoff played for the Calgary Flames. His team scored 190 goals, excluding empty net goals and he earned 68 decisions. Thus, we assume that the Flames scored 190 * 68 / 82 = 158 goals while he was in net. Kiprusoff faced 1,716 ES shots, 283 PP shots and 41 SH shots; based on the league average save percentage in those situations (92.1%, 87.5% and 90.8%, respectively), we’d expect an average goalie to allow 136 + 35 + 4 = 175 goals. Plugging this into the formula, we’d expect Kiprussoff to win 44.8% of his decisions.
Step 2: calculate the number of games the goalie was expected to win
Calculation: take expected win percentage from step 1 and multiply it by the number of his decisions, multiplied by the OT/SO adjustment for that season.
From 1918 until 1999, every NHL game was worth two points – either the winner took both, or each team took one in a draw. Thus, the Wins Added formula requires no adjustments for any seasons up to and including the year of Gretzky’s retirement. The formula is designed so that the sum of the Wins Added statistic for every goalie in any season adds to zero (which makes sense, since it’s a zero-sum game and I’m comparing goalies to the league average).
Starting in 2000, the league awarded teams one point for losing in overtime (effectively making some games worth three points, and others worth two). Over the next five years, teams averaged 86-87 points per season. The NHL eliminated ties and added shootouts in 2006 (I’ll refrain from editorializing my thoughts on shootouts). Over the next seven years, teams averaged 91-92 points per season. Without making an adjustment, most goalies would have positive Wins Added, since all of the “freebie” points from overtime and shootout losses are being awarded to the goalies.
I’ve decided to allocate the OTL/shoutout loss points to each goalie in proportion to the number of points they were expected to receive. In other words, in 2012, teams averaged 92 points over an 82 game schedule, so teams averaged 1.12 points per game. I simply awarded each goalie 1.12 points for every expected point.
Example: as per above, in 2012 Kiprussoff was expected to win 44.8% of his decisions. He earned a win, loss, or OT/shootout loss in 68 games (35-22-11), so he was expected to win 30.7 games (68 * 0.448). This is then multiplier by the adjustment factor of 1.12, so he was really expected to win 34.2 games after taking into account the points from OT/shoutout losses.
Step 3: compare expected wins to actual wins
Calculation: subtract expected wins from actual wins. Adjust to an 82 game schedule if necessary.
Example: in 2012, Kiprussoff won 40.5 games (35 wins plus 11 overtime and shootout losses), but a statistically average goalie would be expected to win 34.2 games. Therefore Kipper contributed 6.3 wins to the Flames, above and beyond what a statistically average goalie would have contributed.
[01] Does the analysis take shot quality into account?
No, it doesn’t directly take shot quality into account, in the sense that I haven’t done a rigorous analysis based on shot type and location, like Alan Ryder. That being said, shot quality is indirectly taken into account since I’m breaking up the shots by situation (even strength, power play and shorthanded). This is imperfect, but it’s a step in the right direction. For those interested, in 2012 Antti Niemi had the most favourable shot profile of any goalie with 50+ starts (85.1% ES, 12.9% PP, 2.0% SH), while Kari Lehtonen had the least favourable (78.6% ES, 19.4% PP, 2.0% SH).
[02] Why don’t you remove the goalie’s stats when calculating the league average? Can’t this skew the results?
Given that there are 30 teams, no goalie has a significant impact on the league average. To test this, I re-calculated Rinne’s Wins Added for 2012, excluding his shots and saves to calculate the three (ES, PP and SH) league average save percentages. I chose Rinne for this analysis because he faced more shots than any other goalie. The tall Finns’ Wins Added only changed by 0.1, which is trivial. Although the adjustment is valid in theory, in reality it is not material.
[03] Why do you use an exponent of 2?
To be honest, I no longer have the appropriate software to do a rigorous analysis of what the best exponent would be. Using a “guess and check” system, I found that something around 2.1 probably worked best, but it was an infinitesimal improvement over 2. For simplicity, I stuck with a nice, round two.
[04] Why do you exclude empty-net goals when calculating the amount of offensive support a goalie receives?
First, I’m interested in the amount of offensive support that a team provides before the game is over. The vast majority of empty-net goals are scored when the outcome of the game is essentially decided. From the perspective of seeing how much goal support a team provides its goalies, I don’t think that there’s any relevant difference between a team winning 4-3 or winning 5-3 with an empty net goal.
Second, if I didn’t exclude empty net goals from the offense, then the calculation wouldn’t balance (since empty-net goals are excluded from the goalies’ personal stats). This allows me to check and confirm that the total expected goals for equals the total expected goals against.
[05] Why did you use the average number of goals a team scored over the season, as opposed to the average number of goals the team scored when each goalie is in net?
Although the data exists, it would require enormous manual effort to compile the number of goals scored while each specific goalie is in net. I can’t imagine this having a material impact, but somebody should challenge me on this if they disagree.
[06] In 2012, isn’t Lundqvist at a disadvantage since his team is in the Atlantic, which features the top two offenses in the NHL?
I’ve been hearing this a lot lately, and I’m skeptical. The people who tell me that Lundqvist faces the top two offenses in the NHL six times a year never seem to mention that he also faces the anemic Islanders (third worst offense in the NHL) six times a year, which would at least offset facing the Flyers. They also fail to mention that he faces each of the elite offenses in the west (Canucks, Blackhawks, Wings) just once per year.
Still, I looked into this further. I calculated the weighted average number of goals scored by the teams the Rangers faced (excluding empty net goals). The Rangers’ opponents averaged 216.5 goals, compared to the league average 210.2 goals.
To see if this is significant, I adjusted Lundqvist’s expected goals against upwards by 3.0%. This is very generous to King Henrik – the Penguins and Flyers score so much largely because they shoot so often (ranked 1st and 4th in shots for, respectively) and the number of shots Lundqvist faces has already been accounted for. Even after this generous adjustment, Lundqvist’s results increase by only 1.0 Win Added. This is a much smaller change than some have suggested – Lundqvist shouldn’t win or lose the Vezina on the basis of this argument.
[07] Is clutch play taken into account?
This metric possibly captures clutch play indirectly. The main problem with save percentage and GAA is that they don’t consider the impact that each goal/save has on winning –a goal allowed in the last minute of the third period in a tied game has a much larger impact on a team’s ability to win/lose a contest than a goal allowed in a 5-1 blowout. Thus, if a goalie is particularly good (or bad) in clutch situations, it might be captured in this metric. Then again, a goalie could unfairly receive credit for the clutch play of his teammates (and vice versa) – if a goalie wins more games than expected, it might simply be due to the team’s offense scoring goals at the right time.
[08] Why don’t you take goalie penalties into account?
I suppose one could argue that the goalie is responsible for any shots faced on penalty kills where the goalie himself drew the penalty. I think one should take penalties into account for legitimately crazy netminders like Smith, Hextall and Belfour. However, goalies are much better behaved post-lockout. Over the past seven seasons, only one goalie has reached 30 PIM in a single season (Ray Emery in 2007).
Go figure, a goalie named Smith led the NHL in penalty minutes! In 2012, Mike Smith was easily the most penalized goalie in the league, with 16 PIM (no other goalie had more than six). Let’s assume that none of these are coincidental minors. The average PK percentage last year was a bit under 83%, so we’d expect an average team to allow 1.4 goals on the PK as a result of Smith’s lack of discipline. However, they would have allowed some goals at even-strength anyway. If we assume that Smith increased his team’s expected goals allowed by 1.0 goal, it would negatively impact this metric by 0.2 Wins Added. Thus, this adjustment is immaterial.
[09] How do you account for traded players?
In a perfect world, I would split out the data for each traded goalie so I can analyze their performance by team, and perhaps add them together to have their total performance for the year. Unfortunately, I don’t have time to do this analysis. I’ve simply allocated all of their time to the team they played the most games in front of – in other words, I’ve assumed that in 2011 Brian Elliot played 55 games on the Senators, instead of 43 in Ottawa and 12 in Colorado.
[10] Is the way you account for OT/shootout loses really the best approach?
You tell me – I’ve struggled with this concept and am open to feedback.
[11] You're so biased, you did this to make my favourite (most hated) goalie look worse (better) than he really is.
NASA is in on the conspiracy too. Please pick up the tin foil hat and leave this thread.
Last edited by Hockey Outsider: 04-12-2012 at 05:17 PM.
Note: in each of the tables, I show the results for all goalies with 45+ decisions. The columns show the goalie, his team, and his actual wins, losses and OT/shootout losses. Next I show his total Wins Equivalents (wins plus half of any OT/shootout loss points), his Expected Wins, and finally his Wins Added. The total Wins Added above won't sum to zero because I'm excluding all of the goalies with fewer than 45 decisions.
Tim Thomas won the Vezina trophy, and he led the NHL in Wins Added. Thomas was exceptional at even-strength (stopping 94.0% of shots faced, leading the NHL) and while his team was shorthanded (stopping 90.5% of shots faced, third among goalies with 50+ games). Boston was a strong team and, as a result, WA expects Thomas to win a lot of games, but he performed even better than expectations (he was 36-11-7 and was expected to be something like 29-17-8).
Save percentage can be misleading. A quick glance suggests that Jonas Hiller was only slightly better at stopping the puck than Pekka Rinne (91.9% compared to 91.7% respectively). In reality, Hiller was clearly better at even strength (93.4% vs 92.6%) and while his team was shorthanded (87.7% vs 86.7%). They only appear to be close because Hiller played behind an aggressive, undisciplined team and faced a brutal amount of shots while his team was shorthanded (24.0%); only his teammate, Giguere, faced a greater percentage of shots while shorthanded. On the other hand, Rinne played on the disciplined Predators and faced just 13.1% of his shots while his team was shorthanded, the best among goalies with 45+ games played.
Backstrom, Luongo, Nabokov and Lundqvist finished 3rd through 6th in Vezina voting and ranked 4th, 5th, 6th and 2nd in Wins Added, so I think the GM made great selections this year. The notable exception is Steve Mason, the Vezina runner-up. There’s no question that his good save percentage from 2009 now appears to be a fluke – but even if we look at his 2009 on its own, his performance isn’t as impressive as many would think. Mason got a lot of credit for helping Columbus earns its first (and so far only) playoff berth, but the numbers indicate otherwise. Columbus had an equally stingy defense in 2008 and 2009 (improving slightly from 6th to 3rd) and their team’s save percentage was also similar both years (decreasing slightly from 91.0% to 90.6%). The real reason that Columbus made the playoffs was due to a significant improvement in their offense – they were dead last in 2008, and improved to 21st in 2009 (partially due to Rick Nash having the best season of his career, and partially due to improved depth). Mason was an above average goalie, but he got far too much credit for the improvement in the team’s offense.
Marty Turco had a terrible year. I don’t understand why Dave Tippett keeps playing Turco so much; he’s the only goalie to have two seasons with a sub-90% save percentage and 60+ decisions. From 2001 to 2004, Turco was tied for the best save percentage out of the 28 goalies who appeared in 150+ games (although this was largely as a backup on a disciplined and very strong defensive team). Post-lockout, Turco has the second-worst save percentage out of the 18 goalies who appeared in 300+ games. The Stars were a playoff-calibre team (20th in goals scored; 5th in shots allowed). An average start would have given the Stars around 96 points, good for 6th in the West.
Last edited by Hockey Outsider: 04-12-2012 at 05:14 PM.
Ryan Miller won the Vezina and Ilya Bryzgalov was runner-up. Both were strong choices according to Wins Added, finishing third and first, respectively. Both netminders had excellent even-strength save percentages (both were at 92.8%; T-4th among goalies appearing in 50+ games). Miller’s strong performance in 2010 was largely due to his staggering, unsustainable 91.9% save percentage on the powerplay, easily the best in the NHL. Overall, I give Bryzgalov the Vezina due to putting up essentially the same W-L-T record as Miller while facing a tougher shot mix (19.3% on the PP compared to 15.3% for Miller) and playing on a team that was around 10% worse offensively.
Martin Brodeur was the second runner-up for the Vezina; he was very good (10th in Wins Added), just not Vezina worthy. Although one could possibly argue for him on the basis of a good save percentage (10th among goalies appearing in 50+ games) combined with a heavy workload, that statistic fails to take into account the fact that he was sheltered by an incredibly disciplined team (just 13.3% of the shots he faced were on the powerplay). His save percentage was only slightly above average at ES and quite a bit below average on the PP.
Jimmy Howard didn’t get much respect for a very strong season (he finished a distant 8th in Vezina voting). His 37-15-10 record was quite strong given that Detroit had an unusually bad year offensively (surprisingly, they scored 3% fewer goals than the average team).
Thomas won the Vezina in 2009 and 2011, but had a surprising off year in 2010. This was due to a poor 90.8% even-strength save percentage (it was 94.0% in 2009 and 94.7% in 2011).
The 2010 Philadelphia Flyers made it to the Stanley Cup finals, despite the fact that their starting goalie has one of the worst Wins Added scores of all-time. Aside from goaltending, the Flyers were actually a very strong team (8th in goals scored, 5th in shots against). Aside from Boucher, their troupe of backup goalies went 32-17-3 despite having a very pedestrian 91.1% save percentage, which is almost exactly what we’d expect given the strength of the Flyers’ offense and defense. Still, one could argue that Boucher’s regular season was simply due to bad luck – despite posting an 89.9% save percentage in 2010, he had a 91.7% SVP in 2009 and a 91.6% SVP in 2011. He also looked average in the playoffs before getting injured (90.9% SVP).
Antero Niittymaki played behind an undisciplined Lightning team. A brutal 22.0% of the shots he faced were on the powerplay.
Last edited by Hockey Outsider: 04-12-2012 at 05:15 PM.
In 2011, Tim Thomas won the Vezina, and, for the second time in three years, Wins Added agrees that he was the best goalie in the league. Thomas had a staggering 94.7% save percentage at even strength, easily the best in the NHL. He was also 7th in save percentage on the PK. Thomas’s 8.8 WA puts him very close to Hasek during his phenomenal stretch from 1997 to 1999 (he averaged 9.5 WA per year, though that was using a slightly different version of the formula and is probably not directly comparable).
Rinne and Luongo were the other finalists for the Vezina trophy. I have them ranked 3rd and 5th among goalies appearing in 50+ games, so they were strong choices. Rinne in particular was a good selection (3rd at ES save percentage and 2nd at PP save percentage). Interestingly, Rinne was the highest-ranked goalie in Hart voting (4th overall) despite being the runner-up in Vezina and all-star voting.
Bryzgalov was a distant 6th in Vezina voting, but Wins Added has him ranked 2nd. Despite playing in a brutal situation (playing on a below-average offensive team and facing the highest percentage of shots on the powerplay of any goalie), he won a lot of games (36-20-10). An average goalie would have been just barely over 50%, which would have caused the Coyotes to tumble from 6th to 12th.
Congratulations to Nikola Khabibulin. He recorded the worst single season in NHL history, measured by Wins Added. (Again, the previous version of Wins Added is probably not directly comparable to this, but any adjustment to the previous calculations would be small). Thanks to the advent of freebie points for losing in OT and shootouts, every goalie but two had a win percentage of at least 40% (with Elliot narrowly missing at 38.2%).Khabibulin was far behind everyone else- he had a brutal 26.1% win percentage, which is something you’d expect see on a 1990’s expansion team, not in the parity-obsessed post-lockout NHL. Of course, the Oilers were a bad team and Khabibulin didn’t get much help, but he was brutal at even strength (90.5% save percentage, the second worst among goalies with 45+ decisions) and even worse while his team was shorthanded (81.5% save percentage, also second lowest in the league).
Last edited by Hockey Outsider: 04-12-2012 at 05:16 PM.
It seems like most people believe that the Vezina is a two horse race between Lundqvist and Quick, with Rinne likely finishing third. All three goalies have had excellent seasons and rank in the top seven according to Wins Added. I’d vote for Quick – both netminders earned 83 points in the standings for their teams, however Quick faces tougher shots (the 2nd highest percentage of PP shots faced out of all goalies with 50+ starts) and is on a clearly inferior team (although similar in shots allowed, the Rangers are 11th in goals scored while the Kings are 29th). I debunked the “Lundqvist faces tougher opponents” argument earlier. That being said, there’s another goalie who had a very similar season to Lundqvist and Quick, helping drag a team to unexpected heights…
I’m disappointed to see Mike Smith receive very little support for the Vezina. He has the same save percentage as Lundqvist (and similar flashy but meaningless stats like wins and shutouts), and maintained this excellent level of play over more playing time. Lundqvist has a better GAA, but that’s simply a product of him facing fewer shots (Rangers are ranked 6th, Coyotes are 28th). Smith had the best even-strength save percentage in 2012 (93.6%) among goalies who played 50+ games. If the Coyotes had an average goalie instead of Smith, they would have lost around 18 points in the standings, which would send them tumbling from their division title down to 13th place. I would easily make Smith a nominee for the Vezina instead of Rinne (better save percentage and more Wins Added, and that’s before taking into account the fact that Rinne plays in front of arguably two of the top five defensemen in the league).
Miikka Kiprussoff quietly had a very strong year. He’s a likeable player and I was glad to see him bounce back after a very disappointing stretch between 2008 and 2011 (ranked 14th out of 16th in save percentage for goalies with 200+ games played).
Marc-Andre Fleury was second in the NHL in wins, but he’s slightly below average in terms of Wins Added. He was fortunate to play in front of a very strong team in 2012. Pittsburgh had the league’s best offense by a comfortable margin, and allowed the fourth fewest shots on goal; both factors increase the number of games he was expected to win. Fleury also faced a very favourable shot distribution – out of the 21 goalies who earned 50+ decisions, only three faced a lower percentage of PP shots than Fleury. He’s won a ton of games for a 27 year old, but the numbers suggest that many goalies would have done so, had they played in his situation.
I wonder if this was Brodeur’s last season. He had the worst even-strength save percentage (91.1%) among goalies appearing in 50+ games.
Last edited by Hockey Outsider: 04-13-2012 at 11:34 PM.
Very interesting stuff! I do something similar (Support-Neutral Wins and Losses), although I don't do some of the things you do (like split results by power play situations) because I want to apply it further back.
My guess is that my less-refined calculations overstate the goaltender's contribution (for good or bad). For coarse comparisons, I've got Tim Thomas at:
(2008-09) SNW/L 35-19 (suggesting a wins added of 8.0)
(2009-10) SNW/L 22-21 (suggesting a wins added of 0.5)
(2010-11) SNW/L 37-18 (suggesting a wins added of 9.5)
This is awesome but one thing that seems unaccounted by looking at the results and not dissecting the method is team system. Otherwise it looks quite odd to see Ilya Bryzgolov at the top of the charts for his years in Phoenix and then fall off the map in Philli. Similarly, Mike Smith comes out of nowhere and tops the charts once on Phoenix.
I guess it adds credibility to the old adage "show me a great goalie and I'll show you a great coach"
Elliot also seems to be a good example of this in St. Louis.
However, maybe it is luck that accounts for the variences as well, just like Thomas' down year in 2010.
As a Canuck fan, how did Shneider due in his limited playing time over the last two years. It would be interesting to compare the two goalies rates of WA. statistically, it has not been too difficult to pick up wins in Vancouver over the last couple years.
And finally, I believe that I have seen you do numbers like these for players and their offensive/defensive contribution (unless I am imagining it), but can't remember if it reinforces the notion that the goaltender is the most important position in hockey via Wins Added?
I'm telling you right now that Fleury being in the negative on this list makes me almost not even look at it (just this season, it was fully deserved in 2010). Fleury has been awesome this year, and that strong defense really wasn't all that strong this year. If you watch the guy, he was the linchpin of our defense this year. Letang missed like half the year, Martin sucked, Michalek wasn't that great. Hell, Shero even made the statement after the season "is it possible for the league MVP to be Malkin and the team MVP by Fleury?" He then qualified his statement saying how ridiulous it was but his point was to articulate just how good Fleury was this year. He was the second best player on a top 5 team in the league (better than Neal). To say he actually didn't add any wins is awful (and I understand the process of a formula, the work you've done here is fine HO, it's just that in this circumstance the numbers don't tell the story).
Having Smith at the top makes me question the stats.
Phoenix has quality goaltending year after year. It makes more sense that its the system then that they consistently get new goalies that are best in the league.
Having Smith at the top makes me question the stats.
Phoenix has quality goaltending year after year. It makes more sense that its the system then that they consistently get new goalies that are best in the league.
How is that any more confusing than some franchises getting bad goaltending year after year? Some teams get lucky. Some teams get unlucky.
Is it safe to assume that Brodeur suffers from this analysis?
Yes. It includes an ES/PK/PP breakdown which hurts him vs traditional save percentage, but doesn't account for goaltender shot prevention or home scorer bias, both of which would help him.
Having Smith at the top makes me question the stats.
Phoenix has quality goaltending year after year. It makes more sense that its the system then that they consistently get new goalies that are best in the league.
Yes, there are always certain teams that cycle through goalies that always have a good save percentage for that team and a much poorer one for other teams. Nashville and Florida in the past; seems Phoenix has joined them.
Nashville in particular seems to have a very trigger happy scorekeeper when it comes to recording shots; one reason why I take Rinne's official save percentages with a grain of salt
What do you (or others) think are the biggest points of doing this study and publish its results?
I suppose it is supposed to help give a picture of about much influence different goaltenders have over their team's ability to win? Apart from that, isn't this mainly a curiosity stat?
I think refining save percentage, by isolating situational goalie stats and then combine them is great. (Yes, I've done it for a while mysef.) One can then sort the goalies according to that "combined" save percentage.
But is it wise to include things like team offense in the formula?
Why not just calculate the "combined" save percentage for each goalie, and then the average "combined" save percentage for the league as a whole. Then normalize the results, so that the average goalie gets 1, and the other goalies get results relative to that. Very easy, and one can instantly know exactly how good a goalie performed compated to league average. Let's call it save ratio (I'm not North American and am not familiar with all the official terms used.)
Then find out the league average GF and GA (which is the same) per team, for example 226-226 in a season where there are scored on average exactly 6 goals per game.
Then insert the numbers into your pythagoran win formula.
Team has scored 226 goals.
Team has allowed 226 / "save ratio". (Or 226 * "save ratio", depending on how you want above 1 to be better or worse than average.)
If you want to account for playing time, you just multiply the "pythagoran win %" of the goalies with the share of time they have played.
Shouldn't this simple method give you even "fairer" results than the one you use? (Fairer to the goalie, as we eliminate both his skaters' ability to score and his skaters' ability to prevent opponents from shooting. Without having thought carefully about it, I think you don't even have to exclude EN goals too, which if so would be another benefit.)
Finally, don't forget that home and away matters a lot, as team's generally win more at home than on the road. If you're not able to account for that in your formulas, you can list number of home games in a column and let the readers make their own adjustments. (There may have been cases where certain goalies played unproportional shares of home games.)
If you'll continue to prefer your current method, maybe you also should use pythagoran math for both expected and factual, to avoid mixing apples and oranges (and perhaps also to avoid having to deal with OT wins/losses).
Now I just saw another thread by you about save percentage. (I have now only looked at the top 2 posts.) http://hfboards.hockeysfuture.com/sh...d.php?t=718221
Obviously different persons prefer different methods and presentations, but that seems like another case where normalized save percentage would be more informatiive. (Make the seasonal average equal 1.000 and calculate all save percentages according to that. If doing so, one will always, instantly, be able to see if a goalie has performed above of below league average. You will also instantly easily see which are the "best seasons" ever, just like with - for example - hockeyreference's adjusted scoring stats. Of course, which "everyone" including me knows, the number of teams also matters.)
One major advantage something like this has over save percentage is that it credits goalies for playing more games at a high level. One of the biggest issues with straight up save percentage is that it's an averaging stat - closer to points per game than it is to points at the end of the season (to make an analogy with skater stats). "Wins added" is an attempt to determine a goalie's seasonal value and is an improvement over straight save percentage. Some of the issues inherent to officially recorded save percentage are still there, though
One major advantage something like this has over save percentage is that it credits goalies for playing more games at a high level. One of the biggest issues with straight up save percentage is that it's an averaging stat - closer to points per game than it is to points at the end of the season (to make an analogy with skater stats). "Wins added" is an attempt to determine a goalie's seasonal value and is an improvement over straight save percentage. Some of the issues inherent to officially recorded save percentage are still there, though
I basically agree on that. But basically, it's simply a way of combining (a normalized) save_percentage with games_played.
Very interesting stuff! I do something similar (Support-Neutral Wins and Losses), although I don't do some of the things you do (like split results by power play situations) because I want to apply it further back.
My guess is that my less-refined calculations overstate the goaltender's contribution (for good or bad). For coarse comparisons, I've got Tim Thomas at:
(2008-09) SNW/L 35-19 (suggesting a wins added of 8.0)
(2009-10) SNW/L 22-21 (suggesting a wins added of 0.5)
(2010-11) SNW/L 37-18 (suggesting a wins added of 9.5)
I haven't done 2011-12 yet.
Thanks! I have Thomas at +6.7, -1.5 and +8.8 from 2009 to 2011, which is prety close to what you have. Different approaches, but it's good to see that we're in the same ballpark. How do you calculate SNWL?
I think that it's useful to split up shots/saves by ES, PP and SH situations, though of course that means we could only go back to 1998.
If I had far more free time than I do (the only reason I had time to do this is because I've been really sick the past week!), I'd calculate the estimated percentage of ES/PP/SH shots faced for each team, going back to the early 1980's, and do some type of analysis (Wins Added and/or adjusted save percentage). I think that with some effort I could come up with a regression formula to accurately calculate the percentage of a team's shots at ES, PP and SH based on certain inputs (ie penalty minutes). It would be a big, ambitious project but maybe I'll give it a try at some point.
Quote:
Originally Posted by LeBlondeDemon10
Very revealing. Good work. Could you post the WA stats for the Habs in 2010? That is the year Price and Halak split duties. Thank you.
Thanks. In 2010, Halak was +6.6 and Price was -2.5.
The formula expected both goalies to roughly win half their games, but Halak was 26-13-5 and Price was 13-20-5. For whatever reason, Halak was far more effective at stopping the puck at even strength (93.3% vs 92.1%) and while the Habs were on the penalty kill (89.2% vs 86.6%).
Quote:
Originally Posted by Dennis Bonvie
Is it safe to assume that Brodeur suffers from this analysis?
Brodeur suffers in two ways - I think one way is a legitimate strike against him, the other isn't.
Brodeur has been lucky to play on a disciplined team, therefore he faces a low percentage of powerplay shots, which are much harder to stop than even-strength shots. Over the past four seasons, thirty goalies have faced 4,000+ shots. Brodeur has faced the 2nd lowest percentage of powerplay shots and this analysis corrects for that distortion.
Percentage of powerplay shots faced
2009-2012; 4,000+ shots faced
Chris Mason
19.5%
Jose Theodore
19%
Kari Lehtonen
18.8%
Nikolai Khabibulin
18.6%
Marty Turco
18.6%
Jonathan Quick
18.3%
Roberto Luongo
18.2%
Craig Anderson
18.1%
Miikka Kiprusoff
18.1%
Dwayne Roloson
18.1%
Steve Mason
18%
Jonas Hiller
17.9%
Evgeni Nabokov
17.6%
Brian Elliott
17.6%
Ilya Bryzgalov
17.5%
Marc-Andre Fleury
17.3%
Mike Smith
17.2%
Carey Price
17.2%
Ondrej Pavelec
17%
Jaroslav Halak
16.9%
Ryan Miller
16.7%
Niklas Backstrom
16.6%
Tim Thomas
16.4%
Cam Ward
16.2%
Jimmy Howard
16.1%
Henrik Lundqvist
16.1%
Tomas Vokoun
16%
Antti Niemi
15.6%
Martin Brodeur
15%
Pekka Rinne
14.2%
On the other hand, New Jersey's home scorers are very stingy and have traditionally under-recorded shots. Let's assume that his shots are understated by 5%.
Before taking the above consideration into account, Brodeur is +0.1 over the past four years (23rd out of 30 goalies facing 4,000+ shots). After making the shot count adjustment, Brodeur would be +3.1 over that time period (20th out of 30). These aren't great numbers, but keep in mind that this is the tail end of Brodeur's career.
Last edited by Hockey Outsider: 04-13-2012 at 11:52 PM.
Excellent work, and I find the results very interesting. You have Halak and Theodore down in the minus section though (2012 analysis), which I assume is a cut and paste/formatting error of some sort.
I might adopt this as my primary statistical measuring "tool" for a goalie's season on the whole. I like that SV% plays a fundamental part, and that games played and special teams influence the results further as one should "expect".
This is awesome but one thing that seems unaccounted by looking at the results and not dissecting the method is team system. Otherwise it looks quite odd to see Ilya Bryzgolov at the top of the charts for his years in Phoenix and then fall off the map in Philli. Similarly, Mike Smith comes out of nowhere and tops the charts once on Phoenix.
I guess it adds credibility to the old adage "show me a great goalie and I'll show you a great coach"
Elliot also seems to be a good example of this in St. Louis.
However, maybe it is luck that accounts for the variences as well, just like Thomas' down year in 2010.
As a Canuck fan, how did Shneider due in his limited playing time over the last two years. It would be interesting to compare the two goalies rates of WA. statistically, it has not been too difficult to pick up wins in Vancouver over the last couple years.
And finally, I believe that I have seen you do numbers like these for players and their offensive/defensive contribution (unless I am imagining it), but can't remember if it reinforces the notion that the goaltender is the most important position in hockey via Wins Added?
Great work
Thanks for your comments. You're right, I can't capture certain intangibles, like coaching, in the calculation. I was also surprised to see two different Phoenix goalies with excellent scores in three consecutive years - a lot of that might be due to Dave Tippett, who's always been a very good defensive coach.
That being said, the formula tries to address factors like coaching indirectly by taking the mix of ES/PP/SH shots into account - I think that good defensive teams are generally disciplined and don't leave their goalie out to dry on the penalty kill.
Interestingly, the formula has Schneider at +4.2 and +4.3 both in 2011 and 2012, and Luongo at +2.3 both years. I was surprised to see the backup finish ahead both years. They have the identical ES save percentage over the past two seasons (93.2%) and Schneider is well ahead in PP save percentage (90.9% vs 88.4%). As a Canucks, you'd be in a better position to interpret these numbers than me.
I haven't done any analysis like that - you're probably thinking of Alan Ryder at hockeyanalytics.com. His analysis shows that goaltending is the most important position by a wide margin.
Quote:
Originally Posted by vecens24
I'm telling you right now that Fleury being in the negative on this list makes me almost not even look at it (just this season, it was fully deserved in 2010). Fleury has been awesome this year, and that strong defense really wasn't all that strong this year. If you watch the guy, he was the linchpin of our defense this year. Letang missed like half the year, Martin sucked, Michalek wasn't that great. Hell, Shero even made the statement after the season "is it possible for the league MVP to be Malkin and the team MVP by Fleury?" He then qualified his statement saying how ridiulous it was but his point was to articulate just how good Fleury was this year. He was the second best player on a top 5 team in the league (better than Neal). To say he actually didn't add any wins is awful (and I understand the process of a formula, the work you've done here is fine HO, it's just that in this circumstance the numbers don't tell the story).
I didn't watch the Penguins nearly as much as you this past year, so I can't comment specifically.
The formula sees that the Pens had the best offense in the league, and one of the best defenses (both in terms of allowing few shots per game, and also in terms of allowing a low percentage of shots while the team is shorthanded). Therefore it expects Fleury to win a lot of games (he won basically as many as he was supposed to).
Here are some intangibles that would probably bump up Fleury's numbers (feel free to elaborate): the Penguins are a fast, skillful team that takes a lot of chances. Although they don't allow a lot of shots, perhaps the shots that they allow are more dangerous than average. They might give up a lot odd-man rushes and breakaways - I call this the "Grant Fuhr argument". Also, their blueline is surprisingly weak given how good the Pens are overall, especially since Letang missed around 30 games. These two factors would suggest that Fleury is better than these numbers indicate.
Quote:
Originally Posted by TheDevilMadeMe
One major advantage something like this has over save percentage is that it credits goalies for playing more games at a high level. One of the biggest issues with straight up save percentage is that it's an averaging stat - closer to points per game than it is to points at the end of the season (to make an analogy with skater stats). "Wins added" is an attempt to determine a goalie's seasonal value and is an improvement over straight save percentage. Some of the issues inherent to officially recorded save percentage are still there, though
Thanks. Another benefit is that this puts goaltending in the same currency as the standings - wins, or more accurately wins and half of OT/SO losses, are what matter at the end of the day.
Agreed - although I never use save percentage anywhere in my formula, some of the issues inherent in save percentage would still impact the Wins Added calculation. For example, a strong defensive team that generally allows easier-to-save shots would artificially inflate a goalie's save percentage, and it would overstate his expected goals against (therefore inflating his Wins Added).
Last edited by Hockey Outsider: 04-13-2012 at 11:58 PM.
