Cumulative faceoffs & possession: R2=0.408

The possession game is low-risk investment. It's returns are not immediate but over time it grows steadily avoiding booms and busts. Possession is a series of methods, tactics and strategies practiced team-wide, but above all things the possession game is a character and from that character comes a destiny.

"Sow a thought and you reap an action;

sow an act and you reap a habit;

sow a habit and you reap a character;

sow a character and you reap a destiny."

- Ralph Waldo Emerson

At this point in the relatively young history of Corsi, rooted in our minds is its thought and appreciated is its destiny, but less commonly understood are its actions and habits.

Faceoffs of "questionable importance"

I was somewhat shocked to learn last week that faceoffs are of questionable statistical importance. The topic was raised in relation to Sunil's article on the Oilers' latest struggles with faceoffs. He wrote:

"Faceoffs are going to be one of the many hints that the team has flaws and should probably be part of a more holistic assessment of the roster." (source)

This makes a lot of sense. The Oilers have a lot of improvements to make and winning a more faceoffs should naturally be part of that improvement. What made less sense was the idea (not presented by Sunil but present in the community) that underlying statistics do not support their importance. This was well-expressed by Copper & Blue commenter loxyisme who wrote in response to Sunil's article:

"Despite stats saying otherwise... I still think there is some importance to winning faceoffs, I mean you want possession and that’s a way to get it. The thing I’ve noticed while watching the Oilers is that even if you win the faceoff (especially in the defensive end), you still have to do something with the puck." (source)

Intrigued by the notion that statistics indicate that faceoffs aren't important I dug into the referenced materials and consulted the hive-mind of Copper n' Blue contributors. To be precise, the referenced sources don't say that faceoffs are unimportant. They say that :

1. Offensive zone faceoffs affect Corsi for up to 20 seconds (author: Hawerchuk;  Arctic Ice Hockey).
2. Offensive zone faceoffs affect Corsi differently in 5v5 and powerplay situations (ibid).
3. Based on an even-strength shooting percentage of 5.97%, 100 extra offensive zone faceoff wins result in 2.45 extra goals (ibid).
4. Based on a powerplay shooting percentage of 8.98% 100 extra face-off wins result in 3.66 extra goals (ibid).
5. "...no teams take sufficient enough power-play faceoffs to equal a win." (author: garret9; Arctic Ice Hockey).
6. A team’s faceoff percentage explains about twenty percent of its Corsi percentage (author: Garret Hohl; hockey-graphs.com).

The above basically indicate that faceoffs are of questionable statistical importance when it comes to winning games and scoring goals, but contribute to about twenty-percent of team Corsi percentage (CF%).

In order to examine these questions myself I started by replicating Garret Hohl's study, which for my purposes here is termed "divided-by-seasons".

All data was obtained from war-on-ice. Divided-by-seasons means calculations were conducted within-seasons across all seasons of available data (2005 to Saturday, November 14th 2015). Divided-by-seasons raw data may be view in this table. All numbers are even-strength and include both regular season and playoff games.

"Cumulative" data (essentially the same numbers pooled across teams) may be viewed in this table. All charts below are interactive, meaning mousing-over objects shows relevant data.

Divided-by-seasons treatment

The above presents raw team faceoff percentage (FO%) and team Corsi-For percentage (CF%) data from 2005 to present. Data is sorted by FO% in descending order (n=330). There is a trend visible where as FO% decreases, so does CF% (although we might describe this trend as "noisy").

The above histogram is of the raw team FO% data plotted in the first chart the data follows a normal distribution (n=330).

The above shows normalized and standardized team FO% and team CF%. "0" indicates average values while all other lines represent standard deviations. R-squared is 0.184, which is close to the value obtained by Hohl (he looked at data from 2007-present). Dots in the upper-right quadrant are above-average in terms of CF% and FO%. Dots in the lower-left are below average in terms of both variables (n=330).

The above bubble-chart shows normalized and standardized team FO% and team CF% data. Bubble size corresponds to normalized Fenwick-For percentage (FF%). "0" indicates average values while all other lines represent standard deviations. Orange bubbles correspond to performances from the Edmonton Oilers. Since 2005, the Oilers have had one season with an above average CF% and FO% (2005-'06). The subsequent season their FO% was above average, however, their CF% was below-average (n=330).

Cumulative treatment

The above presents team faceoff percentage (FO%) and team Corsi-For percentage (CF%) cumulative data from 2005 to present. Data is sorted by FO% in descending order. Mousing-over the lines displays relevant values. Generally, there is a trend visible where as FO% decreases, so does CF%  although we might describe this trend as "noisy" (n=30).

Cumulative team FO% approximates a normal distribution (n=30).

The above scatter-plot presents normalized and standardized team FO% and CF% data. "0" indicates average values while all other lines represent standard deviations. R-squared is 0.396. Dots in the upper-right quadrant are above-average in terms of CF% and FO%. Dots in the upper-right quadrant are above-average in terms of CF% and FO%. Dots in the lower-left are below average in terms of both variables (n=30).

The above bubble-chart shows normalized and standardized FO% and CF% data. Bubble size corresponds to normalized Fenwick-For percentage (FF%). "0" indicates average values while all other lines represent standard deviations (n=30).

Teams with significantly above-average cumulative CF% and FO% include the Detroit Red Wings, San Jose Sharks, Chicago Blackhawks, and Boston Bruins. The three worst teams Corsi-wise (Edmonton Oilers, Buffalo Sabres, and Colorado Avalanche which are two standard deviations below average) also have poor FO%. Three of these teams have been more than one standard deviation below average cumulatively since 2005.

Discussion

The cumulative treatment reveals a stronger relationship between FO% and CF%. One reason that this treatment is successful in revealing a stronger relationship is that a group of three teams has been cumulatively successful over the past decade or so, while another group of three teams has been cumulatively poor.

What this means is that winning more faceoffs isn't necessarily going to produce more possession, however, teams that have strong possession styles over many seasons, such as Babcock's Detroit Red Wings and Todd McLellan's San Jose Sharks, tend to have significantly higher than average FO%. Teams that are weak possession teams over many seasons tend to have significantly lower than average FO%.

Part of the character of possession style play involves the winning of faceoffs. As presented by Hawerchuck, it may be that for average possession teams winning more faceoffs does not produce substantially more goals or victories. But for very strong or very weak possession teams losing a faceoff may have disproportionately large effects (although such evidence is not presented here).

The above chart displays R-squared values calculated by linear regression for each of the individual variables presented with FO% as "x". We see a clean and neat relationship between several possession metrics and FO% in the cumulative treatment (blue). R-squared values for the divided-by-seasons treatment (orange) are much weaker.

Typically a divided-by-seasons treatment is chosen in analytics because performance in a given season is the topic of interest. Hypotheses typically involve predicting performance in the standings while victories from previous seasons do not carry-over to future seasons. However, a cumulative treatment might be used to measure the effects of a particular coach or style of play on a team over a longer period.

Conclusion

Faceoffs are important. Winning faceoffs is a habit of the best possession teams, even though  faceoff wins don't necessarily lead to many more goals or victories within-season. Loosing faceoffs has been a habit of the Edmonton Oilers teams of the last decade. The last time the Edmonton Oilers were an above-average Corsi team, they had a significantly high FO%; this was the same season they made it to the Stanley Cup finals. If the Oilers are ever to reap the character and destiny of a Corsi winner the habit of winning faceoffs is likely to be a part of that process.