FanPost

Are Current Internet-Based Hockey Possession Stats Predictors of Success?

First off, I want to say thank you to the writers and other contributors on Canes Country. I look forward to reading the articles and posts on this site daily (actually more often than that)!

I think PackPride17's Storm Tracking article for this week and Cory Lavalette's Game Analysis - numbers to know help show that many of hockey's newly hallowed advanced stats are not necessarily accurate predictors of success (success being more wins). Corsi and Fenwick alone do not necessarily relate to the final outcome of the game or over time. While I am a believer in playing the possession game, good possession stats without the ability to score or keep the other team from scoring doesn't result in wins. There just doesn't appear to me to be enough grounding of these metrics around successful outcomes to lead them to be a major deciding factor in how a team gets put together. May as well measure the average number of times a goalie spits out water between drops per game vs. that goalie's save percentage.

If one looks at team 5 on 5 CF% (Corsi For Percentage) for this season to date (on war-on-ice.com), the top 8 teams in the east are: T.B., NYI, DET, PIT, BOS, CAR, WAS and FLA. Six of those teams are currently holding a playoff spot (not bad). But the fact that the Canes are 6th in the east is one heck of an anomaly. The Canes have scored fewer points than all of the other teams in that list and have given up more goals than all but the Islanders at 5 on 5. Change the situational drop-down to "All" and the Canes move up to 4th best in the east, so I think that possession stats alone don't equate to more wins.

Look at the Edmonton Oilers. This last off-season, they traded for Nikita Nikitin and Teddy Purcell. They signed free agents Keith Aulie, Mark Fayne, and Benoit Pouliot. A quick pull of 5 on 5 stats for each of these players shows that all but one were among their respective 2013/2014 team's Corsi For Percentage leaders at their positions.

So looking at the previous year's possession stats it appears that these would be positive possession players to pick up (say that 5 times fast). Two top 4 D with special teams experience and better possession stats compared to their previous teammates, a 3rd pairing defensive D and a pair of positive possession bottom six forwards. These looked like decent pickups at positions many thought needed to be upgraded, based on their Corsi analytics. In fact, I was hoping the Canes would try to pick up a few of these players primarily because of their previous Corsi/Fenwick analytics, previous game usage and special teams experience.

Some Advanced Stats comparing previous season and this season:
Nikita Nikitin D
2013/2014 - CBJ - ATOI=14.9 - CF%=49.3 - P60=0.8 - ZSO%Rel= +2.0
2014/2015 - EDM - ATOI=15.2 - CF%=48.6 - P60=0.5 - ZSO%Rel= +6.5

Teddy Purcell RW
2013/2014 - T.B.- ATOI=12.8 - CF%=53.6 - P60=1.4 - ZSO%Rel= +5.0
2014/2015 - EDM - ATOI=13.1 - CF%=49.6 - P60=1.0 - ZSO%Rel=+12.0

Keith Aulie D
2013/2014 - T.B.- ATOI=09.4 - CF%=41.6 - P60=0.4 - ZSO%Rel= -8.8
2014/2015 - EDM - ATOI=12.9 - CF%=44.9 - P60=0.2 - ZSO%Rel=-10.7

Mark Fayne D
2013/2014 - N.J.- ATOI=15.6 - CF%=55.3 - P60=0.5 - ZSO%Rel= -7.4
2014/2015 - EDM - ATOI=14.9 - CF%=45.9 - P60=0.4 - ZSO%Rel= -9.7

Benoit Pouliot LW
2013/2014 - NYR - ATOI=11.3 - CF%=55.1 - P60=1.8 - ZSO%Rel= +7.8
2014/2015 - EDM - ATOI=12.2 - CF%=51.0 - P60=1.7 - ZSO%Rel=+14.3

ATOI = Average Time on Ice per game
CF% = Percentage of all on-ice shot attempts for
P60 = Points per 60 minutes
ZSO%Rel = Percentage of all on-ice non-Neutral Zone Faceoffs in the Offensive Zone minus the Percentage of all off-ice non-Neutral Zone Faceoffs in the Offensive Zone

After these changes, is Edmonton a better possession team than before?
YES - Edmonton's team 5 on 5 CF% stats have IMPROVED from 44.3% last season to 48.1% this season (almost 9% better than the previous season). Unfortunately, Edmonton's winning percentage dropped from 35.4% to 27.1% (to date), and their goal differential dropped from -67 over 82 games to -75 over 70 games (also to date). So again I think that possession stats alone don't equate to or predict more wins.

Want a good predictor of regular season success? GF-GA differential seems to do the best job I've seen if looked at with most of a season played or retroactively. It looks to me by way of eyeballing final standings that it nails at least 7 of 8 playoff teams in both conferences every season, with that 8th usually going to a team that was 9th in goal differential. That holds true for at least the last six seasons although 2010-2011 deviated at 6 of 8 in the west. In other seasons it went 8 for 8 which I bet makes this a prediction model that is successful at better than a 90% clip. In other words, Boston could be in real trouble compared to Ottawa. In the west, Winnipeg could be in trouble vs. Los Angeles and Calgary.

But this is a correlation that is obvious. Consistently scoring more goals than what is given up will result in a team winning more than they lose. Oddly, regular season goal differential doesn't seem to predict Stanley Cup success.

What I see as the problem is that there aren't many stats for specific activities that are measured as related to overall team GF-GA differential. Sure there are basic stats like points scored and goals allowed, but not many real win/loss or GF-GA related metrics, at least not publicly shared.

If I didn't have to work for a living and had tons of spare time with nothing better to do, I'd try to research how all current discrete data available about player and team activities in hockey might relate positively or negatively to scoring, keeping the other team from scoring, the difference between the two and the general percentage of wins.

I'd then look to establish an automated living model; in other words set up processes to calculate the expected predictive stats and then keep them running continuously over time to ensure they either stay predictive else they get replaced or updated if they deviate. I personally dislike that some of the fundamental ideas behind some modern internet (/layman) hockey analytics are based on one time studies. Some of these one time studies may have been built around a single season's worth of data from as old as two CBA's ago. In any other endeavor where metrics are used, continual validation is critical. Another item missing from many internet-based hockey metrics are objective baselines. Internal and external baselines are needed in order to establish the real value of metrics and what are better or worse values for a metric.

PDO is a great example, as the baseline used was based on a theoretical idea (this plus that should at least equal 1.000) and does not use a baseline built from actual external measurements. Another interesting example is a recent internet article on the effectiveness of late round drafting. When we hear that 16.7% of late round draftees will play in at least 1 NHL game and 10% or less may see 100 games, many mentally compare those numbers to 100% jumping to that number as a baseline because without anything else to compare to, 100% will usually come to mind. Compared to 100%, the study's percentages look horrible, and some people used this study as a reason to argue that getting 5th, 6th and 7th round draft choices in return for a final year player not part of future plans was a wasted trade. Instead this study really established the baseline, and going forward the success of future draftees should be measured against this study (and be used to update the baseline).

What makes things difficult for Hockey analytics is that Hockey is a flowing, mostly continuous game. Baseball and Football are games of discrete activities: everyone gets into a starting position, something happens and a few seconds later the play is over with everyone standing in or moving to their new starting positions. Additionally, points in Hockey are awarded at a vastly slower rate than Basketball, so Basketball analytics may not directly map over. Not sure what analytics exist the Soccer world, as to me that is the most similar sport to Hockey in terms of flow and scoring rate that has a professional level.