Proposal: Let’s Ditch Plus-Minus …and Introducing Ice Tilt!

Photo credit: Flickr / catrinamariee

[Editor’s note: Over the next however long, we’ll be pondering a few ways to brighten up the hockey world. Sometimes silly, sometimes not: here are our proposals for the 2012-2013 season… whenever that may happen.]

Goals, assists, plus-minus. That’s supposed to be how you tell how good a hockey player is. When a skater’s name pops up on the CSN-Washington chyron, they show goals, assists, and plus-minus. ESPN, Yahoo, and NHL.com place plus-minus among their marquee stats for ranking players. But the more we learn about hockey and statistics, the more we know that plus-minus kind of sucks at measuring talent.

Proposal: This season, let’s throw out the plus-minus stat. In this article I’m gonna tell you why I think plus-minus has gotta go, and I’m gonna pitch a stat to replace it.

Plus-minus is a running count of a player’s goal differential. When the player’s team scores, he gets a +1; when the opposition scores, he gets a -1. There are exceptions for shorthanded and power play goals, but that’s the basic idea. Ostensibly, plus-minus is supposed to tell us the net effect a player has on his team’s fortune — as measured in goals. But it doesn’t really do that.

Plus-minus is really a team stat: measuring a player’s ability to some uncertain extent, but moreso measuring the strength of everyone on ice– with a whole lot of random chance factored in too. For example, while Jeff Schultz certainly played well in 2009-10, he was probably not the greatest player in league. But according to plus-minus, he wasleading the NHL with a monster plus-50. We Caps fans know that Jeff (who, I repeat, still played terrifically) benefited tremendously from an explosive Capitals offense (seven 20+ goal scorers), and they were lucky at shooting and stingy in net while he was on the ice.

Manny Malhotra is a victim of plus/minus.

On the other end of the spectrum we have the curious case of Manny Malhotra in Vancouver. In 2011-12 Manny posted a team-worst minus-11, but that’s not because he’s a bad hockey player. It’s actually the opposite: Coach Alain Vigneault trusted Malhotra with an astoundingly large number of faceoffs in the defensive zone (87.6%), allowing the Sedin brothers to take all the tasty ones right near the opponent’s net. In this case, Malhotra’s crummy plus-minus represents neither his talent (as the stat purports) or his team’s performance/luck (as in the case of Jeff Schultz), but rather the peculiar and successful way that player is deployed by his coach.

Aside from the examples of Manny and Jeff, plus/minus is totally hosed by dumb luck. Remember when the Capitals outshot Montreal by miles back in the 09-10 playoffs (WSH had 80 more shots and misses than MTL over 7 games), but they lost anyway? That’s because their shooting percentage was atrocious (or because Halak played like a god, whatever). Washington’s skaters in that series demonstrated all the behaviors that should lead to success in the long term, but they still got clowned by the tiny little sample (and Halak, yes, stop saying his name). Specifically there’s a fella like Alex Semin, who personally fired 44 shots in that series (which usually would be good for 3-4 goals by an average shooter or 6 from a stone-cold sniper like Semin) but instead ended up without a single goal– and a dreary even plus-minus. Shouldn’t our stat better represent the habits that lead to success rather than the vagaries of luck?

From Justin Bourne’s Systems Analyst post

Of course. So it should not have surprised us back in 2009 when Timo Seppa at Hockey Prospectus declared plus-minus as a “nonsense stat.” Or when Ryan Lambert at Puck Daddy lambasted Coach Hunter for benching Mike Knuble based on a bad plus-minus. Or when Driving Play revealed the way-too-high impact of empty netters on the stat. Or when Justin Bourne revealed its flimsiness during line changesAnd so on. And so on.

There are certainly examples of excellent players with excellent plus-minus stats, but in those cases we often see similar representation by underlying numbers (the fearsome #fancystats) as well. If we are going to get rid of plus-minus, maybe we should replace it with one of those.

Our criteria for the stat to replace plus-minus are pretty straightforward:

  1. We want a stat that describes a player’s contribution to the team.
  2. We want to communicate that contribution using a single number.
  3. We want it to be easy to understand by casual fans.
  4. We want it to have a snazzy name.

It’d be nice if our magic stat eliminates the effects of shooting luck and deployment, but that would make it kind of like the holy grail/unobtanium/triforce/Katana fleet of advanced stats; I’ll settle for just doing a better job than plus-minus for now.

Maybe the most obvious candidate for a replacement stat would be a possession stat like Fenwick or Corsi. These two plus-minus-type stats count even-strength shots and misses (Corsi includes blocked shots too). Because shooting percentage doesn’t distort possession stats, Jeff Schultz wouldn’t see his numbers quite so blown up, but guys like Manny Malhotra would still look like a bum since he spends so much time in his own team’s zone. We might just have to live with that.

Like goal differential, Fenwick and Corsi are also expressed as a plus-minus, which essentially makes them counting stats– so a pretty good player that plays a lot may look better than a great one who hasn’t played as much (See: Martin Brodeur). For that reason I’d prefer if we expressed it as a ratio: the user’s Fenwick score divided by the total number of Fenwick events while that player is on the ice. So instead of just saying “Brooks Laich has a plus-38 Fenwick score this season,” we would say “Brooks Laich drives 50.5% of his team’s puck possession.” (Note: Because of the level of competition, this percentage would floats darn near 50% with extremes of around 40 and 60%.)

And here’s the part where we really sex it up: Instead of calling it something obscure like Fenwick or explaining how possession works, we can just call it Ice Tilt: “Brooks Laich is tilting the ice 54.6% in the right direction.” When he’s on the ice, the puck just seems to gravitate towards the other guy’s net– as if the ice were literally tilted that way. And it’s easy to comprehend: over 50 is happy and under 50 is sad.

Back in 09-10, Jeff Schultz wouldn’t have led the league in Ice Tilt (other hockey geeks have long used similar stats with names like FenF%), he would have ranked 151st with a 52.2% (stats from David Johnson’s hockeyanalysis.com). Doesn’t that seem a bit more in perspective to you?

Of course, Malhotra still gets screwed using Ice Tilt, ranking 19 out of 20 on the Canucks with 41.5%, so this is in no way a perfect solution. Deployment and quality of competition are still going to wreak havoc on Ice Tilt.

But despite its problems, Ice Tilt would be a great way to put plus-minus to bed. And it would be pretty cool if CSN would snapshot a player like this:

Alex Ovechkin (37 G, 47 A, 55.4% Ice Tilt)

If you’re curious, here’s how the 2011-2012 Capitals (minimum of 300 minutes) looked using Ice Tilt.

Player Ice Tilt (%)
Mathieu Perreault 55.4
Alex Semin 53.8
Jason Chimera 53.3
Nick Backstrom 52.7
Joel Ward 52.5
Jeff Halpern 52.3
Roman Hamrlik 51.2
John Carlson 50.9
Brooks Laich 50.5
Dmitry Orlov 50.4
Karl Alzner 50.0
Dennis Wideman 48.4
Matt Hendricks 48.0
Troy Brouwer 47.6
Mike Knuble 47.4
Alex Ovechkin 47.1
Jeff Schultz 46.8
Marcus Johansson 45.1

There are still bunches of shortfalls with using possession stats alone (the aforementioned Malhotra effect in particular), but I think it’s plainly superior to plus-minus. It’s pretty easy to understand and less distorted by luck. Plus, Ice Tilt is a pretty sexy name, and it immediately paints a picture of what it means. It’s still going to wobble from game-to-game, but I’m just throwing out ideas here.

The last few years have seen smart research by smart people on better ways to measure player talent and contribution. Scoring chances (laboriously tallied by folks like WaPo’s Neil Greenberg) and zone entries (which may be the big statistical breakthrough of the next year) inch us towards the ultimate goal: an objective, undistorted, and straightforward way to communicate player talent. We’re not there yet, but I think an incremental step like Ice Tilt would be good progress.

So anyway, that’s my thought– possession percentage instead of goal differential. If you’ve got a better idea or just want to call me names, use the comments below. If you’ve got your own idea for another proposal, let’s hear that too. We’ve got a lot more coming!

For the record, Ice Tilt as I’ve shown it here is David Johnson’s FenF% during 5v5, close score, adjusted for zone starts by removing events in the first 10 seconds after faceoff.

  • HockeyAnalysis

    How does your ice tilt improve things all that much? Let’s address your complains about +/-.

    1. “Plus-minus is really a team stat” So is ice tilt.
    2. “In 2011-12 Manny posted a team-worst minus-11, but that’s not because he’s a bad hockey player.” Malhotra had a team worst 41.5% ice tilt.
    3. “Driving Play revealed the way-too-high impact of empty netters ” This is simply due to a poor definition of +/- which should be updated. The way it deals with short handed goals should be fixed too.
    4. ” Justin Bourne revealed its flimsiness during line changes.” To some extent this would affect Ice Tilt too. Maybe to a lesser extent.

    Ice Tilt would still have many of the same problems associated with it as plus-minus does. Specifically the quality of teammates, quality of competition, and zone start usage that goals are (in fact zone starts affect fenwick more than goals). Ice Tilt removes some luck factors over smaller sample sizes but driving shooting percentage is a skill and it does not take that into account so over larger sample sizes is not as good (in my opinion anyway).

    You note that your Ice Tilt metric fixes the Jeff Schultz problem, but it also placed Alex Ovechkin 47th, just ahead of Niklas Hagman and Shawn Thornton and just after Matt Cooke and Ryan Kesler. Yeah, the guy who scored 50 goals (for the 4th time in 5 years) ranked just ahead of career 30 goal Shawn Thornton. That just doesn’t seem right.

    We all yearn for a single, easy to understand metric that fairly and accurately evaluates every player but it doesn’t exist and probably won’t ever exist. Hockey is just too complex for such a stat.

  • http://www.russianmachineneverbreaks.com/ Peter Hassett

    All good points, but I think they’re all (or almost all) also addressed in the post.

    “this is in no way a perfect solution. Deployment and quality of competition are still going to wreak havoc on Ice Tilt.”

    “Manny Malhotra would still look like a bum since he spends so much time in his own team’s zone. We might just have to live with that.”

    “It’d be nice if our magic stat eliminates the effects of shooting luck and deployment, but that would make it kind of like the holy grail/unobtanium/triforce/Katana fleet of advanced stats; I’ll settle for just doing a better job than plus-minus for now.”

    “the ultimate goal: an objective, undistorted, and straightforward way to communicate player talent. We’re not there yet, but I think an incremental step like Ice Tilt would be good progress.”

    I’m not saying it’s a perfect idea (I say at least 4 times why it is not), but I am saying it’s a BETTER idea. You can agree with that, right?

  • http://www.russianmachineneverbreaks.com/ Peter Hassett

    “driving shooting percentage is a skill and it does not take that into account so over larger sample sizes is not as good (in my opinion anyway).”

    Great point, that one– although does plus/minus have any fidelity in reporting that?

  • HockeyAnalysis

    The point is, +/- is a flawed stat when considered on its own without context (ice time, quality of teammates, competition, etc.) as you point out but that is a problem with almost all stats including Ice Tilt. Is Ice Tilt better than +/-? Maybe, but probably in the same sense that getting punched in the arm is better than getting kicked in the nuts. Getting punched in the arm is less painful, but still not an enjoyable experience. Ice Tilt may be less bad (over smaller sample sizes, loses its benefit over larger sample sizes), but on its own is still not a useful evaluator of talent.

  • HockeyAnalysis

    A major component of plus/minus is goals scored and a major component of goals scored is shooting percentage so yes, plus/minus takes into account the ability to drive shooting percentages. Just need the sample size large enough to minimize the influence of luck.

  • Bean

    A lot of statistics are flawed. You also really can’t compare a forward to a defenseman. You have to look at the overall picture. I don’t think anyone, anywhere, thought Jeff Schultz was the best player or defenseman in the league, solely on the stat he led it in plus/minus.

    IMO, removing plus/minus would be liek removing Assists. Some assists are totally earned because the player made the play that allowed the other person to score. Other assists are more, well you touched the puck last and the other person did all the work and scored.

  • Otter

    Let’s just call it as it is. Schultz sucks and is a waste of bench space and $2.75m.