Predicting Which Players Will Succeed on the Powerplay

Embed from Getty Images

Alexander Semin did not have a good season last year. After producing decent numbers in his first two seasons in Carolina, with 35 goals and 51 assists in 109 games, Semin struggled in 2014-2015, putting up only 19 points over 57 games and seeing his shooting percentage drop below 10% for only the second time in his 10 year career. With three years remaining on a contract paying $7MM per season, the Hurricanes decided to cut their losses, buying out the Russian winger prior to the start of the UFA period in July.

While at first glance Semin’s release seems like a reasonable response for a former top scorer who appeared to have lost the magic touch, if we look at little closer at Semin’s numbers a different story beings to emerge. Semin logged only 1.5 minutes of powerplay time per game in 2014-2015, down more than 2 minutes from his 2013-2014 total, and well below the 4+ minutes he would see at the start of his career in Washington. While other factors certainly played a role in his fall from grace (a 97.5 PDO at 5-on-5 doesn’t help), there’s no denying that the coaching staff’s decision to keep Semin off the ice when the ‘Canes were up a man cost him (and likely the team) points.

Although Semin is an extreme case, the general story of a player losing points as his powerplay time decreases is not uncommon amongst NHLers, and illustrates that opportunity often matters just as much as ability when it comes to a player’s results. Each team’s powerplay minutes are limited, and valuable to both the team and player, given the higher scoring environment that exists when a team is up a skater. Overall, teams scored roughly 25% of their goals on the powerplay last year, despite the fact that less than 20% of total ice time was played with a team on the man-advantage.

These rare minutes aren’t spread over a team’s whole roster either, therefore it’s critical that clubs are able to identify the players most likely to make significant contributions on the powerplay. This, however, isn’t easy to do. Because of the limited ice-time, coaches are often faced with a chicken and egg scenario. It’s difficult to tell whether a player is good on the powerplay without giving them minutes, but no one wants to give a player special teams minutes until they’ve proven themselves. With so few opportunities to try new things, how can a coach get a reliable read on who they should be playing when they’re up a skater?

One solution is to look at how well players perform at even strength. In theory, we’d expect that the same skills that lead to success at 5-on-5 should generally be useful with an extra skater. But to test out our theory, we need to choose a statistic to use as a proxy for success on the powerplay. Ideally, we’d use goal differential, but when we’re looking at individual players we run into issues with the small sample size accumulated by each player over the course of a season. In any given season there are only about 120 or so players that hit the 200 minute mark on the man advantage, which is already a low amount of ice time to base our analysis on.

Instead, we’ll look at which even strength stats best predict a player’s powerplay Corsi Differential Per 60 (PPCD60), which should give us a reasonable sense of which metrics will have an impact on a player’s long term (i.e. more than one season) goal differential. We’ll limit our sample to players who played more than 200 powerplay minutes in a given year and look at forwards and defencemen separately, since there may be different variables that are indicative of PP success for each position.

After running our analysis, we find that the two even-strength variables that provide the best predictions of a player’s PPCD60 are a player’s Individual Corsi For Per 60 (iCF60, how many shot attempts he personally takes per 60 minutes) and his Teammates’ Corsi For Per 60 (tmCF60, how many shot attempts his linemates take while he’s on the ice)1,2. For forwards, our regression equation is:

PPCD60 = 1.07 * iCF60 + 1.23 * tmCF60 + 21.4

While for defencemen it is:

PPCD60 = 0.79 * iCF60 + 1.07 * tmCF60 + 30.3

For both forwards and blueliners the model does a reasonably good job of predicting a player’s Corsi Differential, with an R^2 for forwards of 0.27, and the R^2 for defencemen coming in slightly lower at 0.18. While both values indicate that there’s still some variance within our results, much of this is likely due to the fact that we’re still dealing with limited TOI for most of our players. What’s more important is that even though we’re working with small sample sizes, we can still make a fairly good estimate of a player’s powerplay success using only their even strength results.

Using our models we can identify players on each team who may be deserving of more powerplay time, or, in contrast, those who may not have earned the man-advantage minutes that they’ve played so far. The table below shows each team’s predicted “Best Alternative”, the player who is currently playing less than 30% of a team’s powerplay TOI, but who the model believes would improve their Corsi Differential by the most if they were to replace a current powerplay regular. It also displays the player that the model thinks should be the first to go based on their historical stats to date this season3, and the change in CD60 that could be expected from making the switch.

Team Best Worst Excepted PPCD60 Delta
OTT Shane.Prince Milan.Michalek 22.89
BOS Brad.Marchand Ryan.Spooner 13.78
CGY Michael.Frolik Jiri.Hudler 11.55
NYR Dylan.McIlrath Ryan.McDonagh 10.30
EDM Matt.Hendricks Mark.Letestu 8.90
STL Magnus.Paajarvi Troy.Brouwer 8.12
PHI Radko.Gudas Mark.Streit 8.07
TOR Shawn.Matthias Peter.Holland 7.51
VAN Sven.Baertschi Brandon.Sutter 7.49
T.B Vladislav.Namestnikov Jonathan.Drouin 7.42
BUF Mark.Pysyk Cody.Franson 7.41
NSH Barret.Jackman Ryan.Ellis 7.19
S.J Joonas.Donskoi Patrick.Marleau 7.10
ARI Stefan.Elliott Michael.Stone 6.94
N.J Eric.Gelinas David.Schlemko 6.88
CHI Bryan.Bickell Teuvo.Teravainen 5.76
MIN Jason.Zucker Thomas.Vanek 5.66
NYI Josh.Bailey Brock.Nelson 5.54
DET Jakub.Kindl Niklas.Kronwall 4.98
WSH Nate.Schmidt Matt.Niskanen 4.76
CAR Andrej.Nestrasil Jeff.Skinner 4.46
WPG Jacob.Trouba Tyler.Myers 4.19
ANA Josh.Manson Cam.Fowler 4.15
CBJ Cody.Goloubef David.Savard 3.40
L.A Dustin.Brown Anze.Kopitar 3.16
COL Nick.Holden Francois.Beauchemin 2.78
DAL Mattias.Janmark-Nylen Patrick.Sharp 1.95
MTL Lars.Eller Dale.Weise 1.94
FLA Alex.Petrovic Dmitry.Kulikov 1.52
PIT Beau.Bennett David.Perron 1.04

There are going to be players in the table that won’t necessarily make the most sense due to sample size issues (I know that I wouldn’t be rushing to replace Teuvo Teravainen with Bryan Bickell, for one), but there are also a lot of switches, particularly towards the top of the table, that are easy to envision. Giving Shane Prince, one of the few bright lights in Ottawa’s possession game, more powerplay time over an aging Milan Michalek wouldn’t shock anyone, but the expected difference in their powerplay shot generation is staggering – Michalek has been awful so far, while Prince remains one of the Sens’ few decent secondary offensive drivers. Similarly, while Ryan Spooner has struggled, Brad Marchand has remained one of the more underrated possession drivers in the league, so it’s no surprise that the model believes he’d be a better fit on the Bruin’s first PP unit than the former Peterborough Pete.

While this method isn’t meant to offer an exact system for setting powerplay lineups, it does provide a guide for where potential replacements can be found when a special teams unit isn’t performing at the level that it needs to. Much like most other statistics, these numbers need to be taken in the context of numerous other factors, but can offer an objective comparison for a scout or coach’s in-game observations. This method is just one of many tools that could assist in decision making about who to send out, and when.

As for Alexander Semin, while he struggled to stay in the starting lineup for Montreal, his performance in limited minutes wasn’t all that bad – he currently projects to be a better PP fit than all but one of the Habs’ regular forwards. Given the struggles on the powerplay that Montreal has faced since Semin was waived, it appears as if the Canadiens could use him back in the lineup.


1 Other predictors tested include High Danger Scoring Chances (Individual/Teammate), Scoring Chances (Individual/Teammate), Goals, Assists, Primary Points, Corsi For RelTM

2 All coefficients are significant at the 0.05 level.

3 Data up to January 8, 2016.

One thought on “Predicting Which Players Will Succeed on the Powerplay

  1. An R^2 of 0.27 and 0.18 indicate that there is A LOT of variance in the data (you mentioned the low TOI). With that much standard deviation not being accounted for it’s very hard to base much off the data in regards to consistency. Very interesting post though, definitely something to look at going forward.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s