FanPost

Are The Leafs Better Then What Fenwick Close Suggests?

Previously I wrote about Team Level QOC and suggested

  1. Teams face differing levels of QOC because of schedule differences
  2. A small change in QOC has a dramatic impact on the team's Fenwick Close
  3. A low Fenwick Close may result from either poor hockey skill/strategy or from facing more difficult QOC

In this entry, I will look to see what impact correcting for the difference in Team Level QOC has on NHL team's shot differential (Fenwick Close).

Why the big deal about Team Level QOC?

Before we correct or normalize for Team Level QOC, I want to show there is indeed a "problem" a different way. That is, this team level QOC is skewing results for not only teams but also for individuals themselves. For example, let's look at usage charts of the Sabres and who played more then 30 games using Corsi On and Quality of Competition from SomeKindofNinja player usage charts.


What we see is that practically all of the Sabres appear to be in a "shutdown" roll with the exception of Scott who was sheltered. Typically, CorsiRelQOC is used which improves the league wide more comparison between players but note how the Sabres CorsiRelQOC has certainly improved but is still skewed.

Think that through, how is it possible that pracitcally every Sabre player is play again above average competition. This is strange but some obvious answers are the Sabres are playing against more difficult competition. Another more unlikely answer is perhaps other coaches put only their best players against the Sabres.


This problem was the initial reason I looked into Team Level QOC as it was apparent there was some sort of distortion that was not explained or compensated for and even using CorsiRelQOC. The problem is even more nuanced then that because what the team level QOC tell us, is that Corsi itself is impacted by playing more difficult competition (at least at the team level). There are a couple options I see to correct this artificat.

  1. Do nothing and use the data both at the individual and team level as is and note that results have skew that may mislead if taken at face value

  2. Hold the current Fenwick or Corsi constant and improve the QOC league wide so it better reflects the differing level of competition players face at the Team Level league wide.

  3. Or instead what I have done is "normalize" the Team Level QOC (50%) and Adjust the Team (and player) Fenwick so more valid comparison can be made.

Normalizing Fenwick Close

Fenwick Close was normalized for the different QOC that teams face to create an Adjusted Fenwick Close. The normalization was completed by subtracting the difference in QOC from 50% (league average) and then inputting that difference in the best fit regression equation [Team Fenwick = 7.99 x Team QOC + Constant with Rsquare of 60%]. When normalized, every 1% increase in QOC added a 7.99% increase in Fenwick Close % as shown in the Table below. In effect, Adj Fen Close is the Fenwick Close that we would expect a team to have if they all played the same level of QOC. What the normalization has done is transformed the distribution of Fenwick Close which ranged from 43% to 57% to distribution range of 46% to 54% for Adjusted Fenwick Close. Some teams were close to a league average QOC of 50% and their Fenwick Close did not change while other teams had an easier or more difficult schedule and their respective Fenwick changed up to 5%. Below Fenwick Close and the rank of Fenwick Close is also presented for contrast.

TEAM

Adj Fen Close

Fenwick Close

Rank Fenwick Close

1

STL

53.80

53.91

6

2

NJD

53.08

55.03

3

3

LAK

53.00

57.35

1

4

NYI

52.80

52.01

11

5

CHI

52.78

55.8

2

6

PHX

52.67

50.23

14

7

NYR

52.34

53.88

7

8

PHI

52.29

48.49

19

9

PIT

52.25

49.87

15

10

DET

52.07

53.92

5

11

SJS

51.51

52.41

9

12

DAL

50.97

47.25

23

13

ANA

50.31

48.21

21

14

BOS

50.15

54.35

4

15

OTT

49.81

52.05

10

16

NSH

49.70

45.88

25

17

CBJ

49.30

45.39

26

18

BUF

48.63

43.73

30

19

MTL

48.57

53.63

8

20

TOR

48.34

44.01

29

21

WSH

47.89

47.72

22

22

WPG

47.88

49.71

16

23

CGY

47.82

48.22

20

24

VAN

47.70

51.68

12

25

CAR

47.67

51.08

13

26

FLA

47.30

48.99

17

27

MIN

47.07

48.68

18

28

TBL

46.86

44.99

27

29

EDM

46.24

44.48

28

30

COL

46.12

46.71

24

There are some interesting insights in the Adj Fenwick Close, for example,

  • Some good teams like PEN and STL look better when their Fenwick is adjusted for playing against more difficult competition
  • Some "good" teams like the Canucks and Habs fall down considerably when adjusted for the easy competition they played this year
  • And other teams standings confuse me as much as the traditional Fenwick Close does (NJD, NYI and how BUF is now apparently "better" then the leafs).

A scatter plot (double click to enlarge) of QOC, Fenwick Close and Adj Fenwick Close vs P% in 2012/13 season reveals that

  • QOC is 5% correlated to P% and is not important factor on why teams win
  • Fenwick Close is correlated 24% to P% and is considered to be one better "predictors" of team winning
  • Adj. Fenwick Close is correlated 25% to P% which rivals Fenwick Close correlation

Vspg_medium

QOC with a 5% correlation is not useful for "predicting" which teams will win but as shown previously is important driver in a team's Fenwick Close. Said differently, at team level QOC tells us something important about Fenwick Close but does not tell us much about which team is more likely to win. Fenwick Close and Adj. Fenwick function equally at "predicting" which teams will win though Adj. Fenwick Close appears to be marginally better. Below I include the scatter plot of Adj. Fenwick Close vs P% with team names for 2012/13 season (again double click to enlarge).

Adjfenwick_medium

Is Adj. Fenwick Close Better?

Adjusted Fenwick Close was designed to remove the inequality in QOC schedule differences that team face over the regular season and so now what remains is closer measure of the team skill portion of shot differential. The "improved" Fenwick Close was achieved without sacrificing the correlation factor at least for the 2012/13 season. In that sense, Adj Fenwick Close offers value. However, getting to Adj Fenwick Close is time consuming and tedious and appears to offer little value in aggregate at least as measured by Rsquared. It also appears to fix some discrepancies (differing quality of competition) but does not improve other discrepancies.

I'm surprised that Adj Fenwick Close correlates as well with P% but I expect over splits and multiple seasons that Fenwick Close will be as good if not superior. The reason I believe this is that P% like Fenwick Close doesn't correct for teams that face the different QOC and so Fenwick Close "should" logically better correlate to P%. This of course is unproven conjecture and as noted QOC is unimportant to winning (5% correlation) so it may have little impact one way or the other. Said differently, the random variation allowed by the Rsquared may make the difference between Fenwick and Adj Fenwick inconsequential for predicting P%. That said, assuming the two measures offer similar correlations over splits and seasons, I don't think we can say which measure is "better" but rather both Adj Fenwick Close and Fenwick Close are imperfect measures that estimate a team's skill. The fact that two different rankings produce similar correlations indicates how much variability is allowed for in these rankings. For example with Fenwick Close, the Canucks are 12th overall and correcting for their easier schedule the Canucks are 24th overall and yet the two measures offer similar correlations. Neither of these rankings are "correct" but both are "valid" because these advanced statistics are dominated (nearly two thirds) by randomness, what some call luck or possibly some unmeasured skill that can allow for a wide range of possible rankings. When it comes to correlations I believe it is best to present the Rsquared, show the scatter plot and let the reader make their own decision of how relevant if at all, the relationship is.

What about them Leafs?

The leafs are not as bad as Fenwick Close may lead us to believe (29th out of 30) when adjusting for quality of competition (20th out of 30). There is still some room for the team to improve player skill and/or team coaching strategy but the overhaul needed to become league average is less then a 1% improvement in Adj. shot differential rather then a 6% improvement in Fenwick Close. This does not tell the whole story. At some point, I may show the analysis on how Fenwick Close penalizes the defensive system the leafs employ by another 1.6% which would make them a 50% Fenwick Close team.

Given the leafs relative position in Adj Fenwick Close and considering the "weak" correlations of Adj Fenwick or Fenwick Close, the leafs are in wide variable range of a competitive bubble and/or playoff team. It would be hard to argue with advanced statistics that the 2012/13 leafs are cup contender as much as suggesting that they are a cellar dwellar or due for collapse as some believe to be the case when using this season's Fenwick Close. Basically, after a lot time consuming number crunching and work I think we can conclude the leafs are competitive team that probably had some fortunate bounces. Surprise.

Much Ado About Nothing

All is not lost with this Adj Fenwick Close make work project as there are some benefits from this exercise:

1. Adj. Fenwick Close is a better measure of "shot differential control skill" then Fenwick Close which reflects both how well a team is able to control shot differential and also the level of competition a team faces.

2. Whereas, Fenwick Close is likely more appropriate for predicting P% in a given season, Adjusted Fenwick Close is likely better for "predicting" playoff and future success (next year) as it is QOC independent and a truer measure of "team shot differential skill" [logic based conjecture and as noted the variability in Rsquare may mean that both will work equally]

3. Corsi at individual level and usage charts (QOC vs zone starts) is only valid for players on the same team (or among players that face with similar Team Level QOC). For example, each player on the Leafs look "worse" as measured by advanced statistics then their true skill because they are facing more difficult competition at the team level. [Recall each 1% increase in Team Level QOC penalizes a team (and its player's) Corsi by 8%]. Therefore, comparing the leaf players Corsi to the a Canucks player who faced an easier schedule would be misleading because the easier schedule gives a "boost" to Canucks player Corsi advanced stats. The takeaway here is that we can compare Corsi and usage for two players on the same team but we need to be careful when comparing Corsi of players on two different teams. [The impact of Team Level QOC on individual Corsi can be corrected. For example, with the leafs each player's Corsi should be boosted by 4.3% to glean a truer measure of their ability to control shot differential assuming equal quality of competition. Likewise, other teams need to be boosted upwards or downwards to make more valid cross team comparisons].

My Takeaways

  • When using advanced stats with weak correlations there is a potential for wide variations in data from reality. [The data showed a jump of 9 spots by leafs and drop of 12 spots by the canucks and other team position changes all of which did not hurt the correlation). There is a definite relationship between shot differentials and winning and the challenge is understanding under what conditions the relationships breakdown and when we should be cautious in using them.
  • Team Level QOC is not important factor to winning because ultimately hockey like life is unfair (some teams have harder schedule and others have easier ones) but it does strongly correlate with Fenwick Close.
  • However, at the team level Fenwick Close can underestimate the true skill of team's ability to control shot differentials particularly when a team faces significantly more difficult competition (i.e. leafs) throughout the season. This is more of a concern when making prognostications about the results in the following season or in the playoffs. And more importantly,
  • Certain advanced stats at the individual level are misleading and overestimate or underestimate player skill when compared to other players particularly for players because to team level QOC differences.

PensionPlanPuppets.com is a fan community that allows members to post their own thoughts and opinions on the Toronto Maple Leafs and hockey in general. These views and thoughts may not be shared by the editor of PensionPlanPuppets.com.

X
Log In Sign Up

forgot?
Log In Sign Up

Forgot password?

We'll email you a reset link.

If you signed up using a 3rd party account like Facebook or Twitter, please login with it instead.

Forgot password?

Try another email?

Almost done,

Join Pension Plan Puppets

You must be a member of Pension Plan Puppets to participate.

We have our own Community Guidelines at Pension Plan Puppets. You should read them.

Join Pension Plan Puppets

You must be a member of Pension Plan Puppets to participate.

We have our own Community Guidelines at Pension Plan Puppets. You should read them.

Spinner.vc97ec6e

Authenticating

Great!

Choose an available username to complete sign up.

In order to provide our users with a better overall experience, we ask for more information from Facebook when using it to login so that we can learn more about our audience and provide you with the best possible experience. We do not store specific user data and the sharing of it is not required to login with Facebook.

tracking_pixel_9355_tracker