My Thoughts


Odds and Ends

posted Nov 12, 2017, 10:17 AM by Robert Vollman

I think the Vadim Shipachyov story is the most recent turn of events recently. Three months ago, the Vegas Golden Knights gave him a contract with a big bonus and a $4.5 million cap hit, believing him to be the franchise's first No. 1 centre. Turns out, Cody Eakin is the team's first No. 1 centre, and Shipachyov didn't even make the team. That is such a strange turn of events. Were they make a horrible misjudgment three months ago, or a horrible one today? Either way, it's a fascinating story to follow. Here are the other stories on my mind these days.

#HockeyHelpeR

There's a new hashtag to follow on Twitter, thanks to Alex Novet (@AlexNovet). He, and others, are going to help people transition from Excel spreadsheets to R, which is a software tool that can help people do the same statistical work, but in half the time. However, there can be a bit of a learning curve for those without previous experience in coding.

Also, Hayden (@3Hayden2) said he'll similarly help people with Python, but I'm not sure if there's a hashtag for that yet. It's really great when people spend the time to help build up the community, and help more people get involved, and get ahead. To learn even more about programming, be sure to attend the third annual Vancouver Hockey Analytics event, March 2-4, 2018.

Defensive Defenseman

As you might know from following my blog, I've been playing around with defensive measurements lately (though not terribly seriously). To start my next model, I asked my Twitter followers for their subjective assessments of the best defensive defenseman in the game right now. Based on the top four suggestions, I put out the following poll, which surprisingly had Victor Hedman out front.


Based on your replies, Lindholm, Karlsson, and Giordano were also viable choices for that poll. Then, Suter, Tanev, Keith, Ekholm. I plan to study these defensemen more carefully to see if I can figure out what makes them tick, and find a way to rate others on the same basis.

Coaching

Based on history, the first coaching vacancy should be coming up soon. Based on your suggestions, I chose the top four choices for the following poll, and apparently Dave Tippett is the most popular coach to take over mid-season (among veteran options, at least). Apparently, Michel Therrien was a dubious inclusion on the list. My bad!


Injuries

Which team has hit been hit worst by injuries? Based on the data compiled at @NHLInjuryViz, it looks like Anaheim, Boston, and Buffalo.


Technically, Vegas hasn't been hit as hard, but they don't have any goalies! Fleury, Subban, and Dansk are all out, and Pickard was traded. They're using an AHL goalie named Max Lagace, and a 19-year-old named Ferguson to back him up. You'll notice my last blog post was about finding them an emergency solution.

As for Buffalo, their blue line has been hit so hard by injuries that it looks like the following. I think this is why the letters O, M, and G were invented.

I ran a poll, but sadly I neglected to include Boston in the list, which is a pity because I had a joke option in there. Nevertheless, Anaheim won this poll handily.


The NHL's Historical Data

Much was made of some recent glitches with NHL.com's latest statistical rollout, but don't ignore some of the great new stuff that may have been lost in the shuffle.

First of all, they have a fantastic new glossary. Also, there is now a plus/minus breakdown all the way back to 1955-56. That means we know how many goals were scored (for and against) when a player was on the ice. That's very useful information, because it tells us who plays on special teams, and can help estimate how much ice time they had.

I took a quick look to see who was on the ice for the most goals against, here's what I found:
185 Salming '83-84
182 Nylund '85-86
177 Larson '80-81
172 Lidster '86-87
171 Ellett '85-86

There's also save percentage going back to 1955-56. I used the data to see how save percentages have changed over time, so that we can compare players across eras, and here's what I came up with.

I also noticed in 1973-74 that Bernie Parent was .932, Tony Esposito was .929, and virtually nobody else was over .910. And, from 1993-94 to 1998-99, Hasek was .929, Roy and Brodeur were .915 and .914, and the league average was .902. Those are some dominant performances!

Assists

Looking back at all this great historical data reminds me of one of the little hobby horses in the world of hockey analytics, to reduce the number of assists. The stat would be more meaningful if it only included legitimate assistance in scoring goals, and not just someone who chipped away a defensive zone breakout, or had some incidental contact that had nothing to do with the play.

If we were to reduce the number of assists, what would be the most effective way? Most people think that it's just to allow only one assist per goal. Even if it's just someone leaving it behind the net for an end-to-end Bobby Orr, that is seen as the best way forward. I'm not sure I agree. I think I like option 3 the best. How about you?


What else?

On November 5, I took a look at which players were out-performing and under-performing last year's scoring rate by the largest extent. Here are the results.
+10.3 Couturier
+9.4 Schwartz
+8.3 Namestnikov
+7.8 Gostisbehere
+7.4 Kopitar, Larkin, Stamkos
(...)
-5.0 Burns, Hanzal
-5.1 Galchenyuk
-5.2 Guentzel
-5.3 Spezza
-6.0 Crosby
-6.9 Sheary

Also, I looked at who was getting the most extra ice time:
+84:43 Schmidt
+78:07 Wagner
+76:56 Dorsett
+72:26 Sissons
+69:57 Tennyson
+69:16 Scandella

In terms of those who have lost the most ice time, it's Kulikov and Enstrom on Winnipeg's blue line, the Sedins in Vancouver, Chris Kunitz in Tampa Bay, and Radek Faksa in Dallas.

I had a poll suggested by one of my followers, on a particular game tactic. Feel free to reach out to me to suggest your own poll questions!


There's a new book on the market, self-published just like my beloved Hockey Abstract. It's called Hockey Analytics, by Stephen Shea and Christopher Baker.

It is great to see another book on the market, and I hope many others follow suit. It's a lot of hard work to write a book, and for very little reward. If anything, it just makes you a target for a lot of haters. Over the years, I've learned that very few people can even manage to write a chapter, so we really need to support the work that's out there.

And, you may have heard that Sunny Mehta, who was one of the first public hirings in the field of hockey analytics (if not THE first) is no longer with the New Jersey Devils, as of last Sunday (h/t James Mirtle). I'm not sure what happened or what it means, but it's something to think about.

In closing, there were two impressive records set on October 28. First, Arizona tied the record for the worst start to a season, by going 0-10-1 like the 1943-44 New York Rangers (h/t Craig Morgan). And, the Red Wings set a record with their 12th consecutive shootout win (h/t Prashanth Iyer)

Finding Vegas an AHL goalie

posted Nov 7, 2017, 10:11 AM by Robert Vollman

The Vegas Golden Knights have already lost Marc-Andre Fleury, Malcolm Subban, and Oscar Dansk to injury, shortly after trading Calvin Pickard away to the Toronto Maple Leafs (oops!).

At the moment, they are using Max Lagace in nets, who is backed up by 19-year-old Dylan Ferguson. I think they can do better than that.

To find a better goalie, I grabbed all the AHL goaltending data from 2005-06 through 2016-17, and specifically highlighted those goalies active in the AHL today. Rather than compare them using straight-up shooting percentage, I used a stat known as SV%+.

What is SV%+? Inspired by the ERA+ statistic in baseball, Cam Charron introduced an era-adjusted version of save percentage five years ago that allows you to measure a goalie's performance relative to league average, and then compare it to other seasons. 

As described by Kurt R of Broad Street Hockey shortly thereafter, SV%+ is calculated by subtracting the league-average save percentage from one, and then dividing by one minus the goalie's save percentage, and then all multiplied by 1000. That means that a League-average goalie will have a SV%+ of 1000, and the degree to which it is higher or lower is the percentage difference between that goalie's save percentage, and the League average. It was later added to Hockey Reference, but soon re-deployed as GA%- by swapping the numerator and the denominator.

Why go to this extra trouble? In essence, SV%+ allows me to compare goalies across different seasons. AHL save percentages have varied over the years, from a low of .904 in 2005-06, up to .913 in the peak period between 2011-12 and 2014-15, and then back down to .908 last season. It may be a minor adjustment on save percentage, but it might be an important one.

The results? Well, Lagace has a SV%+ of 929, which ranks 41 among the 44 active AHL goalies for which there's at least 20 games (or so) worth of career data. Only Kristers Gudlevskis and Alex Nedeljkovic rank lower. Basically, Lagace ranks last among those with names that I can pronounce.

Technically, the best result is Anders Lindback, 1232, but he only played 19 games before this season. That's obviously not enough data. Same thing for Kevin Boyle, who ranks fourth with 1197 in the same 19 games. The picture is a little bit better for Casey DeSmith, 1228 in 35 games, and David Rittich, 1196 in 31, but let's raise the threshold a little higher.

First place is probably Garret Sparks, with a SV%+ of 1197 in 78 games. That means that his save percentage is 19.7 percent higher than the league average over his AHL career. He's currently playing for the Marlies, and he's having a fantastic season. Maybe Toronto will cut Vegas a break, given that they're the ones who took Calvin Pickard off their hands.

Next is Jared Coreau, 1148 in 110 games. He's off to a so-so season with the Detroit Red Wings. Since they have both Jimmy Howard and Petr Mrazek on the big club, and they also have Tom McCollum in Grand Rapids (1010 in 244), it's possible that Vegas could shake him loose.

Of course, acquiring Coreau would take a pretty good offer. Maybe they need to consider those who might be available on the cheap, like Dan Taylor, who is back in North America after four pretty strong seasons in Europe. Thanks to his solid AHL career from 2007-08 to 2012-13, he has a career SV%+ of 1147 in 134 games. However, he's 31, has a lousy .892 save percentage for the Belleville Senators, who just traded away their other goalie, Andrew Hammond (494 in 80). So, this might not be a viable option for Vegas.

Would the Tampa Bay Lightning give up AHL mainstay Michael Leighton? He's 36, and has a lousy .867 save percentage for the Syracuse Crunch this season. However, it was .921 last year, and Leighton has been a consistent AHL performer for years. He even has 111 games of NHL experience with four different teams. His SV%+ is 1134 in 318 games. He might be the ideal solution, if they can get him.

Among active AHL goalies, here are all the goalies with a career SV%+ above 1000, but with a minimum of 100 games:
1148 Coreau
1147 Taylor
1134 Leighton
1116 J. Smith
1110 Hutchinson
1085 Berube
1079 Gibson
1065 McKenna
1056 Bachman
1048 Pasquale
1045 Thiessen
1039 Campbell
1024 Grosenick
1019 Tokarski
1010 McCollum

What can be expected from these goalies? Well, when the Los Angeles Kings got into trouble last year, they used Peter Budaj, but his SV%+ was an incredible 1235 in 79 games. Other NHL-calibre goalies are even higher, with 1440 in 72 games for Matt Murray, 1307 in 51 for Frederik Andersen, 1284 in 53 for Scott Darling, 1277 in 119 for Jonathan Bernier, and 1269 in 96 for Jaroslav Halak. 

Falling clearly below that level, obviously expectations need to be much lower for Leighton, or whoever the Golden Knights might pursue from this list. Recognizable goalies in the 1134 range include Jason LaBarbera and Brian Boucher, 1133, Philipp Grubauer, 1129, and Josh Harding, 1128. While that level of goaltending won't help the Golden Knights steal many games, it will hold the fort better than Lagace, and certainly better than Ferguson. When the injury bugs eventually clear up, he could be moved down to the AHL to back up Dansk (or Subban). What do you think?


Best Defensive AHL Players

posted Nov 4, 2017, 12:18 PM by Robert Vollman

Who are the best defensive players in the AHL?

According to the process that I'll explain momentarily, it might be someone like Chris Breen of the Providence Bruins and Brian Strait of the Binghamton Devils on defense, and Colin Campbell of the Grand Rapids Griffins, Cole Cassels of the Utica Comets, and Zach Sill of the Hershey Bears up front. Or, they might just be weak offensive players who were lucky enough to play on strong defensive teams. So, let's back up and let me explain the approach.

In my early years, when others were working on more fruitful endeavours like the development and analysis of shot-based statistics like Corsi, catch-all statistics like GVT, or shot quality work like expected goals, I was trying to develop a defensive statistic. 

At first, I looked at the plus/minus statistic, under the mistaken belief that if I could adjust it for the various factors that impact it (which was my first co-publication, with Iain Fyffe, in the Hockey Research Journal in 2001), and then removed the offensive component, then I would be left with the defensive component. 

Sadly, I learned that the best defensive players allow the most goals, because they're up there against top opponents, while the weakest defensive players allow very few, because they're only out there against the fourth lines, and tasked with playing low-event hockey. 

For example, in Ken Daneyko's prime, he was -22 with the New Jersey Devils in 1988-89, while the defensively suspect Tom Kurvers was +11. Now that the NHL has this information online, we know that Daneyko was on for 74 goals against at even strength that season, and Kurvers was on for 71. So, essentially any use of plus/minus to measure defensive play is a dead end.

Iain proposed another way to measure a player's defensive play. He figured that every player was on the ice for a reason, and if that opportunity couldn't be explained by his scoring (or something else that was measurable), then it must have been because of his defensive play, or some other intangible. That formed the basis of the defensive component of his own catch-all statistic, Point Allocations.

So to answer today's question, let's apply Iain's ideas to the AHL. I've got all the player data from 2005-06 to 2016-17 handy, so here's what I did.

1. First, I figured out how many goals were prevented, league-wide. Under the assumption that offense and defense are equally important, the number of goals prevented is equal to the number of goals scored. Bear in mind, this isn't an actual count of goals prevented (which would be a highly subjective exercise anyway), but rather a "thing" that we're going to call goals prevented.

2. Then, I figured out how many goals were prevented by each team. Assuming that the opportunity to score is equal in all games, that means that this can be calculated by adding together the league average goals scored and goals prevented (which is just double goals scored), and then subtracting the actual goals that a team allowed. In theory, this could result in a negative number, but in practice it should result in a distribution that is exactly the same as goals scored.

3. Then, I set aside the goals that were prevented by the team's goalies. But, just how many is that? First of all, the common perception is that goalies are responsible for roughly half a team's goals prevented, or maybe a third.


However, most of the models I've seen peg a goalie's contributions somewhere between 20 and 30 percent. In this case, I arbitrarily decided to assume that a quarter of all goals prevented should be credited to the goalies. 

Now, that's not the same on every team, that's just the average. To break it down on a team-by-team basis, I modified the goals save above average (GSAA) statistic until it added up to a quarter of the goals prevented, league-wide. 

In its standard form, GSAA will add up to zero, because it multiplies the league-average save percentage by the number of shots a goalie faced, and subtracts that from the goalie's actual saves. Through trial and error, I discovered that subtracting 0.0241 from the league-average save percentage before making that GSAA calculation generated results that added up to about a quarter of the league's goals prevented. So, a goalie with a .899 save percentage will still prevent a few goals, but not very many.

4. Next, I figured out how much credit for those goals prevented should be awarded to defensemen versus forwards. 

I proceeded on the assumption that forwards and defensemen were equally valuable to a team. Therefore, the number of goals created plus the number of goals prevented for an average forward should equal that for an average defenseman, and vice versa.

Goals created is a simple formula: add together a player's goals plus his assists divided by the league-average number of assists per goal (so that assists are worth the same as a goal), and then divide by two. If you add up all of a team's goals created, it should equal the team's goals scored. 

Over this entire time span, a forward averaged 0.1882 goals created per game, and a defenseman averaged 0.1059. So, I reversed these numbers for the average number of goals prevented for each position. That is, I assumed defensemen prevent 0.1882 goals per game, and that forwards prevent 0.1059. In practise, I calculated these numbers on a season-by-season basis.

Side Note: There is the argument that defensemen are more valuable than forwards, because they play more minutes. So, you could add up all the goals created by forwards and assume that an equal number of goals were prevented among all defensemen. The results will be 50 percent higher, because those goals were created by three forwards and the equal number of goals prevented are being dividing up among two defensemen. That will make defensemen far more valuable than forwards. While I don't necessarily disagree with that assessment, I opted for the simpler approach this time around.

5. Ok, so now I finally assigned a team's goals prevented to its forwards and to its defensemen. Here, I did so on an equal basis, with the exception that forwards and defensemen were being weighted differently, as computed in step four. 

For strong defensive teams, like the 2015-16 Albany River Rats, or the Wilkes-Barre/Scranton Penguins from 2010-11 through 2013-14, that meant multiplying the given average (0.1882 or 0.1059) by just under 1.2, while it got as low as 0.65 for weak defensive teams, like the old Binghamton Senators, or the 2007-08 Lake Erie Monsters.

Side Note: For fun, I used this to figure out which players were frequently on particularly good or bad defensive teams. All I did was add up a player's goals prevented over the entire time span (2005-06 to 2016-17), and divided it by their games played to create a "base team index" that might become useful later. At the very least, it indicates whether a player has played on particular strong or weak defensive teams, which is useful information in and of itself. For example, defensemen Ryan Lannon and Aaron MacKenzie often found themselves on good defensive teams, as did forwards like Matt Mangene and Nate Thompson. They were either really lucky, or doing something right. On the flip side, scoring-line forward Derek Grant and enforcer Darren Kramer were on some awful defensive teams, as was defenseman Ryan Murphy -- that's either bad luck or bad defense.

6. Ok, so the team's goals prevented have been assigned, but on an entirely equal basis. Ideally, I'd like to see the same spread in goals prevented that we see in goals created. For example, strong offensive defensemen like T.J. Brennan and Brendan Montour average up to 0.3 goals created per game, while stay-at-home defensemen like Nathan McIver are below 0.03. I'd like to create that same spread in terms of goals prevented, except scaled up to reflect an average of 0.1882 instead of 0.1059. So, the range would be between 0.05 and 0.47. 

To do that, I assigned more of a team's goals prevented to the shutdown defensemen, and fewer to the enforcers and/or one-way offensive defensemen. So, I divided a player's goals created relative to the rest of the team's players at the same position. Again, this goes back to Iain's premise that everyone is in the lineup for a reason, so if they're not there to score, then they were probably there to defend.

So, rather than assigning 0.1882 per game for defensemen and 0.1059 for forwards (or whatever the averages were that season), I divided that number by the result I achieved here. This actually worked out nicely, because the average results varied from a low of around a third, to a high of three, which meant that it would make roughly the six-spread distribution I wanted. 

Oh, and two more details. I also did something similar for penalty minutes, just to filter out those who were in the lineup as enforcers. And, for those who didn't play very many games, I regressed the results towards 1.0, up to 200 games (which was arbitrarily chosen). For example, if someone played 50 games and had twice as many goals created as expected, his end result of 2.0 was regressed down to 1.25. That's 50 games at 2.0, and 150 games at 1.0.

In terms of all-time results, the leaders among defensemen were from 2005-06 through 2016-17 were:
Joey Mormina, 154.8 goals prevented in 670 games with an average team weight of 0.179 (average is 0.1882)
Ryan Lannon, 146.9 in 318, with 0.219
Andrew Campbell, 142.0 in 595, with 0.164
Joe Piskula, 140.6 in 576, with 0.170
Brian Sopitz, 138.1 in 294, with 0.190
Maxime Fortunus, 133.2 in 803, with 0.176
Chris Breen, 131.2 in 400, with 0.175
Breen Palin, 127.1 in 367, with 0.205
Corbin McPherson, 112.7 in 287, with 0.185
Jaime Sifers, 110.9 in 544, with 0.183

Up front, the leaders were:
Andrew Joudrey, 83.9 in 492, with 0.103 (average is 0.1059)
Rod Pelley, 77.7 in 491, with 0.104
Zach Sill, 76.2 in 407, with 0.116
Mike Keane, 72.2 in 365, with 0.105
Brett Sutter, 72.0 in 678, with 0.096
Francis Wathier, 71.1 in 577, with 0.100
Harrison Reed, 70.9 in 208, with 0.092
Warren Peters, 68.8 in 578, with 0.106
Ryan Garlock, 68.1 in 321, with 0.096
Carter Bancks, 68.0 in 426, with 0.103

To arrive at the five players I listed at the top of the piece, I just computed this on a per-game basis, and chose the five top active AHL players.

7. (Future work) Ok, so the most glaring problem is that this system just rewards players for failing to score, and for being on great defensive teams. Sadly, it's impossible to tell one-way defensemen and two-way defensemen apart using just the basic stats available in the AHL. Likewise, it's impossible to separate the shutdown defensemen from third-pairing options who simply can't score. 

In terms of traditional stats, both sets of players look identical. In the NHL, for example, Marc-Edouard Vlasic has 91 points and 358 shots in 212 games from 2014-15 to 2016-17, and Trevor Daley has 85 points and 327 shots in 206 games. In terms of higher-scoring defensemen, Duncan Keith has 141 points and 484 shots in 227 games over this time span, and Kevin Shattenkirk has 144 points and 476 shots in 208 games. In both instances, we subjectively know that one player is far superior defensively than the other, but there is just no way to establish that using basic stats. 

So what I need to do is to make a second pass through this data, and try to get a distribution of goals prevented among the different types of players. That is, even among the higher-scoring players, there needs to be a wider distribution of goals prevented. Yes, their average goals prevented per game should still be lower than the average among lower-scoring players, but the distribution should allow for lots of overlap. Having done this first pass, this could be achieved by simply widening the distribution among each tier of players. 

Furthermore, there are a ridiculous number of reasonable but completely unproven assumptions throughout this process, not to mention a number of reasonable but completely arbitrarily chosen numbers. Each step along the way can be improved by proving these assumptions, and calculating the correct numbers.

But, what I have done is enough for the first pass to kill time on a snowy Saturday, so I'll just leave this here for now, pause to reflect on the next step, and save the hard work for later.

Thursday's Thoughts

posted Oct 27, 2017, 8:39 PM by Robert Vollman

Today, we start with a fun video graphic from Cole Anderson (@CrowdScoutSprts) that shows where in the world all of the NHL players have come from, and how that has changed over the years. 

Unfortunately, my limited skills prevent me from properly embedding the tweet or the image, but you should be able to click here to see it: pic.twitter.com/nDW39mviGi

This week's poll was about the teams sharing the basement with the Arizona Coyotes in the early going. According to all of you, Edmonton is only there temporarily, but either Montreal or the Rangers could be there for the long haul. 


My NHL piece this week was about the hot start from Kucherov and Stamkos, so it involved looking at all of the NHL's top two forwards, some of whom play together, and some of whom are split up onto different lines. 

Which two forwards would you like to have from this point forward? I think there's some recency bias at play, because most people chose the hot hand, Kucherov and Stamkos. Personally, I think Crosby and Malkin will have them beat in terms of both scoring and shot-based metrics. 


Those who prefer scoring might also entertain options like Benn and Seguin in Dallas, Backstrom and Ovechkin in Washington, Tarasenko and Schwartz in St. Louis, or even McDavid and Draisaitl in Edmonton (assuming Draisaitl is ok). Many fans privately suggested Matthews and Marner -- but I think that's premature. Some fans might even go off the board and suggest Giroux and Voracek. Those who favour shot-based metrics might suggest Bergeron and Marchand, or even Pavelski and Thornton. However, it's hard to seriously argue for most of these options. I really think it boils down to Kucherov and Stamkos vs Crosby and Malkin.

My next poll is going to be about how to fill the inevitable coaching vacancies that will soon start coming our way. If you have suggestions about who to place on the twitter poll, be sure to let me know.

In terms of new updates, one of my favourite sites, Hockey Reference, adds new features all the time. The latest is a "What's Happening" column on the far right side of their main page. This is a great way to see interesting upcoming dates. Of note, there's an NHL GM meeting November 17, and the 100th anniversary of the first NHL game is coming up on December 19, 2017.

Finally, I want to make sure that you hold open the weekend of March 3-4, 2018. That's the tentative date of the next hockey analytics conference, in Vancouver. For details, follow Josh Weissbock (@joshweissbock) on Twitter, or the #VanHAC hashtag, or the Meta Hockey website, which should have the details available once they're confirmed. 

My Favourite Hockey Analytics Books

posted Oct 24, 2017, 6:25 PM by Robert Vollman

Today's post is in video format, enjoy!

Knights, Eyeballs, and RITHAC

posted Oct 20, 2017, 1:53 PM by Robert Vollman

I guess their 5-1-0 start isn't really fooling anyone!

There's still plenty of time left in this poll, but so far the pessimistic view of the Golden Knights is quite overwhelming. Only 7% of respondents feel that they're a possible playoff team, despite starting off with a 10-point head start in the first six games. 

I suppose the pessimism is well-founded. They are an expansion team after all, and they're schedule includes two wins over Arizona, one over Buffalo, and one over a Bergeron-less Bruins team. Their special teams have been atrocious, ranking 27th on the power play and 18th on the penalty kill. Their shot-based metrics are middling at best. And, it's hard to imagine a team with a top pair of Nate Schmidt and Luca Sbisa, and a second pair of Deryk Engelland and Jason Garrison (or Brad Hunt) as any kind of contender. But, we shall see.

Eyeballs vs Analytics

I'm normally not one to get involved in philosophical debates, but Matt Henderson made an interesting point that I chose to retweet. "Eyeballs vs Analytics is only a thing for people who don't want to learn about analytics. It's eyeballs AND analytics for everyone else."

Matt doesn't presume that there is anything fundamentally unreasonable about using your eyeballs. In my view, there's nothing unreasonable about that perspective. Analytics aren't for everyone, the same way not everyone cares about training techniques, or equipment, a player's personal life, or any other number of areas that can have an impact on the game.

While there are people who use their eyeballs and nothing else, Matt is correctly pointing out that nobody does the opposite. That is, nobody uses analytics and nothing else. The choice is between using your eyeballs, or using your eyeballs plus analytics. It's not eyeballs vs analytics, it's eyeballs vs eyeballs plus analytics. Sure, some people use more analytics than they do eyeballs, but nobody uses analytics all alone.

Interestingly, the assumption that you have to choose between eyeballs and analytics is a phenomenon exclusive to analytics. If I wrote about all the latest equipment, nobody would tell me to just watch the game. That is, nobody would assume that I was looking at a player's skates and sticks exclusively, and not the game itself. Same thing for training techniques, a player's personal life, or anything else. Analytics is the only facet of the game where people presume that someone would study it to the exclusion of anything else. 

RITHAC

While on the topic of those who like analytics, RITHAC is this weekend, which is the third annual Rochester hockey analytics conference.

For me, it is very rewarding to see these conferences take off, since I held the first one just over three years ago. On one hand, it is wonderful to see these conferences prosper on their own, without me. There are so many other things that have yet to fully catch on, like writing hockey books, helping people get front office positions, getting more people on TV and radio, and so on. So, it's great when something does catch on, and fly without me. 

But, of course, it can be bittersweet to see a bird leave the nest. This is only the second conference for which I wasn't invited (the second annual Vancouver conference was the first). I'm not needed anymore! That's a great thing in most respects, but it's also sad when I'm not needed anymore. You know, I doubt that the Ottawa conference needs me anymore either, but it's nice that they always ask me to come anyway.

If I was in attendance, I'd be eager to hear more about the following presentation by one of the newest stars in the analytics scene, Namita (@nnstats). She tweeted a sneak peek of her study, in which she discovered that it takes 1.5 seasons for 50% of first-round picks to make their NHL debuts, but 3.5 seasons for 50% of second-round picks to do the same. It would be interesting to integrate some of these ideas into my team-building model.


Odds and Ends

Last team to score a power play goal this season? Overwhelmingly, we thought it would be Montreal, but it turned out to be Anaheim, who was the most unexpected results.


Lastly, were my questions about franchise players. Seven years ago, Benjamin Wendorf (@BenjaminWendorf) asked us "If you had your choice of any of today's NHL players to build your franchise around, which player would you pick?  Why?" (Link: Arctic Ice Hockey)

Surprisingly, the most common answer was Claude Giroux, followed by the more sensible (and probably wiser) choices of Sidney Crosby, Alex Ovechkin, Steven Stamkos, and Drew Doughty. 

Well, let's repeat the exercise today. First, I asked the question in an open-ended way, to see what the most common answers were. Then, I grabbed the four most common ones and put them into a poll. The results were overwhelmingly in Auston Matthews favour. (Note: I obviously had to remove the obvious answer McDavid from consideration).


Are we correct, or will Matthews be the future Giroux? Some of it might depend on how much he signs for. If it's much higher than the $10M/yr they're giving Eichel, then I'm not sure that he's the right selection. But, time will tell -- we'll re-visit this in seven years!

Measuring Defensive Performance

posted Oct 16, 2017, 11:54 AM by Robert Vollman

In 2002, Iain Fyffe posited the idea that you could crudely estimate a player's defensive performance by whatever portion of his ice time couldn't be explained by his scoring.

To take a practical example, consider the defensemen on last year's Stanley Cup Champion, Pittsburgh Penguins. Justin Schultz outscored Brian Dumoulin 51-15 in the regular season, and yet in the playoffs when the games matter most, Dumoulin led the team's blue line with an average of 19:03 minutes per night at 5-on-5, while Schultz was mere seconds out of last, with 16:16. Clearly, Dumoulin's defensive contributions were considerable to have earned that much ice time with such modest scoring, and Schultz's must have been very slight.

Yesterday, I decided to play with Fyffe's idea. Here's what I did:
1. I grabbed all the relevant 5-on-5 data from XtraHockeyStats (in minutes!), from 2008-09 to 2016-17.
2. I calculated each player's points per 60 minutes, and their ice time per game.

Now, I did not just compare these, because a player's scoring rate and average ice time is highly contextual. To truly test Fyffe's idea, a player's scoring rate isn't what's important -- it's his scoring rate relative to the team's other options. For example, scoring a lot of points would get a centre a lot more ice time on the Arizona Coyotes than it would on the Pittsburgh Penguins. So I need to keep going.

3. For each team and for each season, I added up points, ice time, and games played, by position (forward or defense).
4. That allowed me to calculate the average points per 60 minutes, and the average ice time per game, by position, for each team and season.
5. Then, I divided a player's actual points per 60 minutes by his team's average points per 60 minutes at his position (NOT counting his own, of course). 
6. I repeated step 5, but for ice time per game.

So, at this point, we know how many points per game a player scored relative to the team's other options at that position. If a player averaged 1.6 points per 60 minutes, and the team's remaining forwards averaged 1.3, then his scoring rate is 1.23 higher. That means that his ice time should also be 1.23 higher than the team average, all things being equal. If it is higher, then he may be very good defensively, and if it is lower, then he may be weak defensively.

7. I summed this information up, and selected everybody who had played at least 200 games over this time span. I have always found that single-season sample sizes are far too small for virtually any statistical purpose, and I doubt this would be an exception.
8. I put it on the following chart so I could see it visually, and confirm that there was a relationship between a player's points per game and average ice time. As one increases, so does the other, exactly as one would expect.


9. I calculated the correlation between points per 60 minutes and ice time per game, and it's 0.48. Crudely, that means that 70% of a player's ice time can be explained by his scoring rate.
10. Using the formula identified by the trend line, I calculated what someone's relative ice time per game should be, based on his relative scoring per 60 minutes.

At last, I sorted them by the widest difference, looking to identify which players had the most extra ice time, and which players had the least. Let's look at each group.

Players With Extra Ice Time

In theory, those with a lot more ice time than expected were bringing extra contributions to the table. Primarily that would be defensive contributions, but it could also be grit, leadership, and who knows what else.

On defense, the greatest outlier was Greg Zanon, a former NHL defensemen for four different teams. Go look up his data, and you'll see why he's here. He took only 333 shots and scored just 62 points in 493 career games, and yet he averaged 19:51 minutes per game. His scoring rate was less than half (48%) the team's other defensemen, and yet his ice time was 11% higher. So, it stands to reason that he was playing well defensively to get all that ice time.

Other defensemen high on the list include Jonas Brodin, Ryan Suter, Jay Bouwmeester, Mattias Ohlund, Drew Doughty, Eric Brewer, Brenden Dillon, Robyn Regehr, Braydon Coburn, Andrew MacDonald, Marc Staal, Alexei Emelin, and Francois Beauchemin.

Most of these defensemen can be safely classified as strong defensive defensemen. Others simply bring to light the question of coaching error. A player's ice time is a function of many things: his offense, his defense, his grit/leadership/intangibles, the team's other options, and the coach's assessment. In some cases, we have to wonder about the coach's assessment -- but I'll leave that as an exercise for the reader. 

Among forwards, Ilya Kovalchuk surprisingly led the list. His scoring was 33% higher than expected, but his ice time was 40% higher. That may not seem like a big deal, but look at the trend line. Getting 40% more ice time corresponds to less than 20% more scoring, not 33%. Plus, it's hard to get much more than 40% extra ice time, no matter how much you score.

Kovalchuk's high result may also be the function of playing for some relatively talent-thin teams that really didn't have any option other than just to play Kovalchuk all the time. In essence his scoring rate was so high that no amount of ice time was too high. Besides, the entirety of what Kovalchuk brought to the table might be understated, especially given how dramatically his teams plunged down the standings when he left town.

Other forwards on the list include Sami Pahlsson, Ryan Kesler, Dany Heatley, Jordan Staal, Ryan Nugent-Hopkins, Olli Jokinen, Ryan O'Reilly, Boone Jenner, Chad Larose, Jarome Iginla, Paul Stastny, Mikko Koivu, Ryan Callahan and Ryan Zetterberg. Again, I think it's plain to see which of these are strong defensive players, and which ones merely played for teams that were seriously lacking in other forward options. 

Players With Limited Ice Time

On the flip side, some players get far less ice time than their scoring justifies, because they are serious defensive liabilities. One obvious group of examples are enforcers, which dominate the bottom of the list; Paul Bissonnette, Cam Janssen, Darcy Hordichuk, George Parros, Jody Shelley, and so on and so forth.

Skipping such players, and we see the lowest-ranked defenseman is Marc-Andre Bergeron, which is absolutely no surprise. He was an excellent scorer, but was used as merely a depth option because he wasn't trusted defensively at all. Cody Franson was the second-lowest, which isn't a surprise given that massive disconnect between his above-average stats, and the fact that he is used as a third-pairing guy, and often finds it hard to find a free agent contract at all. Obviously, he is seen as a considerable liability in one respect or another.

Other defensemen on the list include Kurtis Foster, Justin Falk, Dougie Hamilton, John Scott, Matt Gilroy, Ryan Wilson, Brent Burns, Tyson Barrie, Yannick Weber, Steve Montador, Nate Prosser, Torey Krug, Matt Dumba, and so on. Perhaps you view some of these as coaching errors, but their presence on this list is consistent with most of their reputations.

Up front it's mostly just tough guys, but there's also Tyler Toffoli, Sven Baertschi, Dale Weise, Erik Haula, Kyle Chipchura, Patrik Elias, Ryan Spooner, Tanner Pearson, Derek MacKenzie, and others. 

Others

I also wanted to look for players who were good scorers and great defensively, like Duncan Keith. Because Fyffe's theory doesn't really help us identify the defensive skill of those who have great offensive skill. I mean, Keith's ice time can be justified either way. As it stands, Keith's scoring rate was 41.5% higher than the team's other defenseman, and his ice time was 23.1% higher. Which is actually just a slight bit below the trend line. He's mixed up with several strong two-way players. To use another example, Pavel Datsyuk's scoring was 37.4% higher, and his ice time was 25.9% higher.

I'll be exploring this more in the future, but I thought that I'd share my preliminary assessments. 

And we're underway!

posted Oct 5, 2017, 3:58 PM by Robert Vollman

The 2017-18 season has begun! 

Several of my recent blog posts have looked at standings projections, but we're going to look at just a few more.  First, is the final projection by Dom Luszczyszyn of the Athletic (@domluszczyszyn). It will probably surprise non-analytics fans to see Edmonton so low, and for analytics fans to see Detroit and San Jose so low, and Tampa Bay so high. Beyond that, I don't think that there are too many head-scratchers here, so this should serve as a pretty good reference point about what expectations were at the start of the season, roughly. 

On a game-by-game basis, Emmanuel Perry of Corsica Hockey (@mannyelk) is tracking the projections of all the various models out there, and wow there are a lot of them. Keep your eyes on that: http://corsica.hockey/predictions/

One of the models is the MoneyPuck model, which I have been following for years. Here are the latest projections from that site

And, Colin Cudmore (@CudmoreColin) figured out how the odds of how many Canadian teams will make the playoffs, based on Micah Blake McCurdy's projection model (which we covered in last week's blog post). Nothing terribly surprising here.

As for me, I took a closer look at players who were on a PTO. By my count, only eight of them signed a contract, give or take maybe one. Of them, it sounds like people are expecting the greatest impact from Cody Franson. Me? I think it will be Daniel Winnik. I wrote up my thoughts on Winnik, Franson, and the others for NHL.com (link).

I also took a look at Jagr for NHL.com, and what we can expect from him this season (link). I figure he could crack the 40 point mark again, because his 5-on-5 scoring rate is just incredible. It might slump due to age, and due to being placed on the third line with Versteeg and Bennett instead of the top line in Florida with Barkov and Huberdeau, but we'll see. There's also room for him to get more points on the power play. Most fans seem to agree with this optimistic assessment of his scoring potential this season.

I also think it's crazy that Jagr signed for only $1.0 million. Consider all the solid veterans who signed for $1.0 million this summer, including Jokinen, Cammalleri, Sharp, Hemsky, Desharnais, Jagr, and so many others. Given that rather mediocre, shot-blocking #4 defensemen really cashed in to lucrative long-term deals, you really have to wonder what's going on in NHL front offices. Maybe it's just a supply and demand thing.

Speaking of the summer, I have also marveled at how the Arizona Coyotes have transformed their top four from one of the league's worst, to one of the best. In 2015-16, Ekman-Larrson was joined by Mike Stone, Connor Murphy, and Zbynek Michalek. Two years later, and OEL is joined by Alex Goligoski, Niklas Hjalmarsson, and Jason Demers. Good gravy, that's a big improvement.

It reminds me of the Tampa Bay Lightning from years ago, which also used trades and free agency to build a really good blue line. Nashville built a great blue line too, but did it through the draft, before making the big Weber-for-Subban trade. The commonality in all three situations is that they each had a franchise defenseman already. But, the point is that blue lines can be rebuilt, and there are blueprints for doing so. There's no reason to have a terrible blue line forever. As for the next great blue lines, perhaps we can keep our eyes on Carolina and/or Philadelphia.

I also started looking through the historical data on NHL.com. In case you missed it, they added a LOT of historical data to their database. I'm sure I'll be spending a lot of time going through all of this, but first here's a look at the best goalies of the pre-expansion era. Is that pretty much how you would have had it?

Top SV%, 1955-1967 (min 100 GP)
.921 Bower
.919 Hall
.917 Plante
.912 Worsley, Hodge
.909 Rollins
.907 Sawchuk

Finally, here's a neat tool from Ice Thetics (@icethetics). It's a logo map so you can see where all the teams are located. This one is just the NHL, but you can add all the other teams and see where they're really concentrated.



Wingers, PTOs, and Projecting the Standings

posted Sep 27, 2017, 2:27 PM by Robert Vollman

The BIG news this week is that the NHL has officially digitized a lot of its records. Now, more of the stats on NHL.com go further back in time. For instance, I just worked on something related to penalty shots, and saw that the data went back to the 1934-35 season. So, this is obviously a great day for hockey stat lovers.

Another piece of news is that Corsica Hockey will be joining the Nations Network. Just when you thought everybody was going to work with the Athletic, the Nations Network lands another top resource.

Now, let's get on with today's post, which includes a look at the best LW/RW, players on PTO, and for which teams stats and conventional wisdom have the most wildly different expectations.

Mr. Burns Team of Ringers

Remember that episode where Mr. Burns rigs a company softball tournament by hiring a bunch of ringers? Well, this week we had fun thinking about who he might choose for a company hockey tournament. (Image from Jeff Veillette)

First, we tackled the right wing. Kane was most popular choice, but it was relatively close. I can't say that I disagree, given his incredible scoring totals over the years, but I wouldn't object to Kucherov or Tarasenko, either.

The choice of left wing was a little more clear, Jamie Benn of the Dallas Stars. Unlike all of these wingers (except Brad Marchand), Benn is used as more of a two-way player, and yet he can still score. Still, I'm surprised that the support was so resounding - I figure Ovechkin would be the most popular choice.

The rest of his team is relatively obvious. What do you think, can Mr. Burns beat Shelbyville with Benn-(Crosby or McDavid)-Kane up front?

Players on PTO

One of the other big discussion points of the week were players on PTOs, so I put together a player usage chart of those who were still active on September 25. As you can see there's no one who is likely to really move the needle. In my view, Parenteau and Purcell make nice additions to the bottom six, and Franson is a great third-pairing defenseman, but they are all secondary players, at best.

Different Expectations

Micah Blake McCurdy posted the results of his highly esteemed team projection model this week, named Edgar. Here they are:


How does this compare to conventional wisdom? To answer that, we can compare this to the subjective results of a survey that Dom Luszczyszyn published not long ago

The biggest differences? In absolute terms, it's Edmonton, Tampa Bay, and Detroit. The former two are more beloved subjectively, and the latter by the model. To a lesser extent, the same could be written about Nashville and Anaheim subjectively, and about Colorado, Vegas, and Vancouver statistically.

However, some of those differences are because a statistical model smashes the teams closer together, making the good teams look more average, and making the bad teams look a bit better.

So, in terms of where teams will finish in the standings, the biggest differences are:
  • Detroit, who is commonly believed to finish last in the Atlantic, is statistically projected to finish 4th.
  • Edmonton and Nashville, who are each expected to win its divisions, are statistically projected to finish 5th (and behind Detroit!)
  • San Jose is seen to be a very heavy favourite in the West statistically, but finish 4th in the Pacific subjectively
  • Likewise, Dallas and St. Louis are expected to finish 1-and-2 in the Central by the model, but 4-and-5 by subjective opinion.
  • Subjectively, Tampa Bay is seen as the best bet to win the President's Trophy other than Pittsburgh, but statistically they're more middle-of-the-pack
Obviously, we get attached to the subjective opinions, because they are flooded with recency bias. We believe that last year's standings were exactly what they should have been, even if we re-ran the season many times. In reality, luck had a huge impact on the standings, and therefore on our subjective opinions of each team. That's why there are ALWAYS massive changes in the standings relative to subjective expectations. Indeed, at least one of the four division winners usually misses the playoffs the next year - will that be Washington, Montreal, Chicago, or Anaheim? Plus, based on subjective opinion, did anyone expect Toronto (30th overall in 2015-16) or Columbus (27th) to do so well in 2016-17?

Keep your mind open. Maybe Detroit is an ok team, and maybe Tampa Bay, Edmonton, and Nashville aren't nearly as good as we think. Or, maybe the results of the model are wrong. At the very least, it's interesting information.


Thursday Grab Bag

posted Sep 21, 2017, 11:21 AM by Robert Vollman

In the NFL, there's a Fourth Down Bot that calculates whether a team should punt, kick a field goal, or go for it on fourth down. I'd love to see something like that in hockey. Like, perhaps, a Pull the Goalie Bot, which calculates at what time the goalie should be pulled, based on the score, the scoring rates of the teams involved, the manpower situation, and the location of the most recent event (i.e. you don't want to pull the goalie in the defensive zone). 

The Florida Panthers

Jason Demers was traded by the Florida Panthers to the Arizona Coyotes, for Jamie McGinn. That gives the Coyotes the sensational top four of Ekman-Larsson, Hjalmarsson, Goligoski, and Demers, and depletes Florida's to read Ekblad, Yandle, Petrovic, Matheson. There's some offensive punch there, but that's not exactly a lineup that Crosby will struggle to score against. Meanwhile, McGinn is a depth winger with an annual cap hit of $3.33 million. Yowch.

In Hockey Abstract 2017, I wrote the following about Florida's reputation for being a pro-analytics team.

This move is yet another example that building an analytics team is different than actually using them. If you judge a team's adoption of analytics by how many of their moves meet with agreement from the analytics community, Florida is average, at best (and dropping). Instead, teams like Nashville, Carolina, Pittsburgh, and Arizona are the ones who appear to be including the opinions of the number crunchers in their decision-making.

Wasted Cap Space

Of course, if there's a criticism for teams like Nashville, Arizona, and Carolina, it's the enormous amount of wasted cap space they have. Colin Cudmore just updated his list of the wasted cap space throughout the league, and there are three teams in equally bad situations - Colorado, Columbus, and Toronto.


The Dallas Stars

I saw an interesting post from Brent Severyn on Twitter. Apparently, Ken Hitchcock is coaching the biggest team he's ever had. 

Not only did Dallas add a great coach and plenty of size, but they added Ben Bishop in nets, Marc Methot on the blue line, and Martin Hanzal and Alexander Radulov up front. Yes, they paid a possibly excessive premium to get those players, but watch out for the Stars this year.

Twitter Polls

I had a few polls this week, let's take a look at the results. Let's start with more on Erik Karlsson. Despite likely missing only about a dozen games, his injury is seen as the most significant blue line injury right now. I'll ask about forwards this week.

If Karlsson misses a bunch of games and doesn't win the Norris, then that opens up the door for someone new -- assuming that it doesn't go to a usual suspect like Brent Burns, P.K. Subban, Drew Doughty, or Duncan Keith. If it does go to someone new, who might it be? Victor Hedman, apparently.


That might be a case of recency bias. Hedman finished third in the voting last year, while the others received their Norris consideration in years previous. I was also scolded for leaving off Shea Weber -- my bad! I honestly thought he had already won. He finished second a couple of times, but he has never won. I might have to re-run this poll with Weber instead of Suter.

Last poll of the week, it was about players aged 37 and up. Who is likely to win the old-timers scoring race? Joe Thornton was the choice, despite being outscored 68 to 50 last year by Henrik Zetterberg.


Supporting the Community

A new Patreon campaign was launched, for Natural Stat Trick. It's in our own naked self-interest to get as many financial resources as possible in the hands of those with sites like these. If you are in a position to contribute, here are the four current Patreon campaigns, to my knowledge.

In closing, here's an updated list of important dates, provided by our friends at Cap Friendly. Most importantly, the Waivers period begins tomorrow.



1-10 of 36