Whose special teams are really special in the NHL?

Matt Cane is a contributor to Hockey Prospectus. Follow him on Twitter at @Cane_Matt.

Special teams play has traditionally been a collective blind spot for the hockey analytics world. While this has been the case for several good reasons (not least among which is that 5v5 and special teams results are often strongly correlated) it is an approach that has obvious issues to both the casual observer and veteran analyst. When we ignore special teams performance, particularly when we lack full season results, deviations between observed results and 5v5 performance may not always be solely due to luck. Teams that look too good to be true may actually be perfectly in line with our expectations when we take powerplays and penalty kills into account, while struggling teams with strong possession numbers (New Jersey last year, anyone) may make more sense when we look at their atrocious powerplays. We know intuitively that there are teams with strong (or weak) powerplay and penalty kill units, just as we know that there are teams that are better at drawing penalties, or who more frequently head the box themselves. The trick is to find a way to bring special teams results into our assessments without bringing all of the noise that tends to dominate the data along with it.

Over at Fear the Fin, Patrick D had a great series of articles up examining which statistics mattered most to future special teams success at the team level. Patrick found that when it came to predicting future powerplay success, a team’s Powerplay Fenwick For Per 60 was clearly the best variable we had available to use. On the penalty kill, however, things were a bit more muddled, with PK%, PKGA60, and a combination of possession stats and save percentage all being equally good predictors. This is important to note, as although PK save percentages do vary quite a bit, it that doesn’t mean we should ignore them. We know that goalies talent level differs amongst goaltenders and Patrick showed that (at least on the defensive side of things) it made a noticeable difference.

One of the problems with PK Sv% though is that it is extremely volatile, given the smaller sample sizes involved. We can address that, however, by using 5v5 save percentage, which should provide a good estimate of any given goalie’s ability. Similarly, while we see a lot of variance in PP shooting ability, we should still expect that teams who post higher even strength shooting percentages should be better shooting teams (in general) on the powerplay, and thus if we’re going to try to predict powerplay results we would do well to look at differences in shooting ability.

Building on the work that Patrick has done, I set up two multivariate regressions using observed powerplay and penalty kill Fenwick rates, as well as even strength shooting/save percentages to predict Special Teams Goals For and Against per 60. All data was sourced from war-on-ice.com, and it covers all teams’ full season data from 2007-2013.

On the powerplay, our best estimate of a team’s expected GF60 is given by:

PPGF60 = -2.36 + 0.09*PPFF60 + 0.43*5v5 Fenwick Sh%

While on the penalty kill we get:

PKGA60 = 34.42 + 0.08*PKFA60 – 0.37*5v5 Save %

Both regressions result in reasonable R^2 values, at 0.35 for the powerplay model and 0.28 for the penalty kill model. What’s important to note is that the inclusion of even strength shooting and save percentage data improves our prediction in both cases by a non-negligible amount. If we re-run our models without the shooting/save % variables the R^2 values drop to 0.30 and 0.21 respectively.

With our regression equations in hand, we can use our models to predict each team’s expected special teams scoring rates using this year’s data to date (note that the data below is up to the end of January 11th).

Team xPPGF60 Team xPKGA60
WSH

8.95

VAN

5.77

STL

7.90

WPG

5.78

TOR

7.56

STL

5.82

NYR

7.54

NYR

5.84

S.J

7.43

NSH

5.87

PIT

7.36

N.J

5.95

BOS

7.32

L.A

6.05

OTT

7.20

CGY

6.23

NSH

6.99

DET

6.25

NYI

6.95

OTT

6.26

CHI

6.93

MTL

6.42

ARI

6.93

MIN

6.45

DET

6.90

BOS

6.54

L.A

6.84

CHI

6.60

WPG

6.71

ANA

6.64

EDM

6.70

PHI

6.68

MIN

6.68

FLA

6.82

PHI

6.56

WSH

6.91

COL

6.55

S.J

7.00

ANA

6.50

COL

7.01

VAN

6.49

CAR

7.02

MTL

6.39

PIT

7.05

FLA

6.34

BUF

7.40

CGY

6.24

TOR

7.45

CBJ

6.15

NYI

7.54

DAL

6.02

DAL

7.55

T.B

5.77

EDM

7.60

N.J

5.63

T.B

7.65

CAR

5.61

CBJ

7.97

BUF

5.53

ARI

8.61

Unsurprisingly, the powerplay that’s lead by Alexander Ovechkin has produced the best expected results to date, and by a wide margin. The Blues, whose powerplay strategy Sam Hitchcock recently broke down, have had success and are second behind the Caps, although they sit more than a full goal per 60 minutes behind. Also having a strong year so far are the Toronto Maple Leafs, although their prowess on offense is nearly matched by their troubles when down a man, as the Leafs are only a net +0.1 GF/60 when playing up or down a man.

The numbers in the above table aren’t necessarily all that intuitive to read, however, as most people are used to seeing special teams numbers expressed as success percentages rather than scoring rates. Fortunately, we can transform our scoring rates to percentages by using an expected PP success rate and the following formula:

Expected PP% = PPGF60/(60/(0.8*2+0.2*1))

We assume here that the average success rate is 20%, and therefore the time spent on the powerplay is the full two minutes 80% full of the time, and an average of one minute the other 20% of the time (note that these proxy success rates show an extremely high correlation to actual success rates, ~0.97).

We can also calculate each team’s Expected PK% using a similar formula:

Expected PK% = 1 – GA60/(60/(0.8*2+0.2*1))

So which teams are exceeding what our model would project, and who has been unlucky so far:

Team xPP% PP% Delta PP% xSH% SH% Delta SH%
ANA

19.5%

16.7%

-2.8%

80.1%

80.4%

0.3%

ARI

20.8%

21.8%

1.0%

74.2%

74.6%

0.4%

BOS

22.0%

18.8%

-3.2%

80.4%

80.7%

0.3%

BUF

16.6%

9.2%

-7.4%

77.8%

75.5%

-2.3%

CAR

16.8%

17.8%

1.0%

78.9%

86.7%

7.8%

CBJ

18.5%

24.5%

6.0%

76.1%

79.9%

3.8%

CGY

18.7%

17.8%

-0.9%

81.3%

76.2%

-5.1%

CHI

20.8%

18.5%

-2.3%

80.2%

89.2%

9.0%

COL

19.6%

14.2%

-5.4%

79.0%

85.4%

6.4%

DAL

18.1%

15.3%

-2.8%

77.4%

79.6%

2.2%

DET

20.7%

24.2%

3.5%

81.2%

86.3%

5.1%

EDM

20.1%

13.5%

-6.6%

77.2%

78.6%

1.4%

FLA

19.0%

13.3%

-5.7%

79.5%

81.6%

2.1%

L.A

20.5%

20.3%

-0.2%

81.8%

78.7%

-3.1%

MIN

20.0%

14.3%

-5.7%

80.7%

84.8%

4.1%

MTL

19.2%

13.7%

-5.5%

80.8%

85.3%

4.5%

N.J

16.9%

19.2%

2.3%

82.1%

78.7%

-3.4%

NSH

21.0%

14.4%

-6.6%

82.4%

81.5%

-0.9%

NYI

20.8%

18.6%

-2.2%

77.4%

76.5%

-0.9%

NYR

22.6%

20.7%

-1.9%

82.5%

82.3%

-0.2%

OTT

21.6%

16.9%

-4.7%

81.2%

82.8%

1.6%

PHI

19.7%

21.8%

2.1%

80.0%

74.8%

-5.2%

PIT

22.1%

21.6%

-0.5%

78.8%

88.2%

9.4%

S.J

22.3%

20.9%

-1.4%

79.0%

80.7%

1.7%

STL

23.7%

26.2%

2.5%

82.5%

80.3%

-2.2%

T.B

17.3%

18.2%

0.9%

77.1%

82.2%

5.1%

TOR

22.7%

20.7%

-2.0%

77.7%

82.0%

4.3%

VAN

19.5%

19.7%

0.2%

82.7%

87.9%

5.2%

WPG

20.1%

16.8%

-3.3%

82.6%

82.5%

-0.1%

WSH

26.9%

23.8%

-3.1%

79.3%

79.4%

0.1%

The deltas in the table above are set up so that teams that have a negative delta have underperformed their expected metrics, while teams that have a positive delta should be expected to regress in the second half of the year.

The obvious outperformers in the bunch are the Chicago Blackhawks and Pittsburgh Penguins penalty kills, both of which have posted unbelievably high raw stats, while putting up relatively pedestrian underlying numbers. The Oilers powerplay (and really, everything about the Oilers) have been on the other side of the coin, with their conversion rate while up a man a full 6.6% below where we’d expect it to be. Perhaps more interestingly, the NHL leading Predators are also due for a bump in powerplay success rate, which could make them a particularly dangerous team down the line.

Ultimately, though, these rate stats only tell part of the picture, as they don’t take into account how frequently teams draw or take penalties. The last thing we can look at then, is how much each team’s man advantage luck (or lackthereof) has been worth in the standings to date. If we take the expected special teams scoring rates we calculated above and compare them to their actual results to date, we can convert the number of goals and expected goals into wins (assuming approximately 6 goals/win) to estimate the effect of special teams play on the standings.                                            d

Team PPGF SHGA Special Teams Wins xPPGF xSHGA Expected Special Teams Wins Wins from “Luck”
DET

38

20

3.0

28.6

24.5

0.7

2.3

PIT

29

18

1.8

26.3

28.9

-0.4

2.3

CBJ

34

30

0.7

23.1

31.9

-1.5

2.1

CAR

23

15

1.3

19.7

23.1

-0.6

1.9

T.B

29

24

0.8

24.0

28.1

-0.7

1.5

CHI

29

13

2.7

29.1

22.2

1.1

1.5

VAN

24

16

1.3

20.6

21.4

-0.1

1.5

TOR

31

25

1.0

29.6

27.9

0.3

0.7

COL

19

22

-0.5

23.0

29.3

-1.0

0.5

STL

38

28

1.7

29.2

22.4

1.1

0.5

ARI

29

36

-1.2

24.2

31.9

-1.3

0.1

S.J

29

23

1.0

28.2

21.2

1.2

-0.2

NYR

25

22

0.5

24.0

19.5

0.7

-0.2

MTL

16

21

-0.8

20.5

23.9

-0.6

-0.3

DAL

22

29

-1.2

22.7

27.7

-0.8

-0.3

WSH

29

28

0.2

27.6

24.6

0.5

-0.3

N.J

24

33

-1.5

19.1

25.1

-1.0

-0.5

NYI

29

28

0.2

27.3

22.7

0.8

-0.6

PHI

31

36

-0.8

24.2

25.3

-0.2

-0.7

BOS

19

26

-1.2

20.9

23.9

-0.5

-0.7

ANA

24

28

-0.7

25.1

25.0

0.0

-0.7

MIN

20

19

0.2

27.6

22.5

0.9

-0.7

WPG

25

31

-1.0

27.0

28.6

-0.3

-0.7

L.A

30

29

0.2

27.5

21.9

0.9

-0.8

OTT

25

26

-0.2

29.8

25.7

0.7

-0.8

FLA

16

25

-1.5

20.8

24.5

-0.6

-0.9

EDM

17

27

-1.7

23.6

26.2

-0.4

-1.2

CGY

26

25

0.2

25.4

16.9

1.4

-1.2

NSH

19

22

-0.5

26.5

19.4

1.2

-1.7

BUF

11

35

-4.0

18.9

28.2

-1.5

-2.5

Both the Detroit Red Wings and Pittsburgh Penguins have benefited from quite a bit of special teams luck in the first half of the year, with each team owing roughly 2.3 of their wins to date to good fortune when playing without 5 skaters aside. Those 4.6 points are not something to scoff at either, as taking them out of Detroit’s record drops the Wings down below Florida and out of a playoff position, while the Pens would fall from 2nd in the Metropolitan all the way down to the first wildcard spot.

Somewhat more surprisingly, the 29th place Carolina Hurricanes have actually been quite lucky to date, posting a special teams goal differential of +8 when we would have predicted them to be closer to -4. Neither Buffalo nor Edmonton have had that good fortune to date, with the Oilers down 1.2 wins below where we’d expect and the Sabres 2.5 a full wins under our prediction (although some would argue that this misfortune is awfully fortunate given their position in the McDavid sweepstakes).

In addition to Nashville, the playoff contending Florida Panthers (and isn’t that odd to write) should see a bump in the second half, as their underlying numbers, while nothing to write home about, are significantly better than the results that we’ve observed so far. And watch out for the L.A. Kings – last year’s champs are finally finding their possession game, and if their results on the penalty kill fall more in line with our expectations over the next few months they should be poised for another run.

While the model we’ve used here is obviously very simplistic, it does serve as a good basis to highlight those teams who have had the bounces go their way while playing away from even strength so far. And while special teams may be difficult to predict, they’re not impossible, and to ignore the portion of the game where 24% of all goals are scored leaves an obvious hole in our analyses. Although we’re unlikely to make predictions as good the one’s we can make for even strength play, the data we have available today does allow us to make a better guess than simply assuming all teams are equal. Each marginal gain we make in our analyses gives us a more in-depth understanding of what’s driving each team’s record, and allows us to better drill down on the factors that drive success at the NHL level.

2 thoughts on “Whose special teams are really special in the NHL?

  1. Pingback: Hockey Prospectus: Whose special teams are really special this year? | puck++

  2. Pingback: UHN │Lunchtime Links – January 14th, 2015 | Ultimate Hockey Network

Leave a Reply

Your email address will not be published. Required fields are marked *

*