why possession isn’t everything

It’s true. Possession isn’t everything. The problem with that argument is no one seriously looking at it disagrees. It’s a strawman argument.

But while no one is arguing that puck possession is everything, it does seem to be the only thing that people really focus on when analyzing hockey (with some exceptions obviously).

That’s not a wrong thing to do, possession is a tremendous indicator of future success, but it’s missing part of the pie. The part where shooting and goaltending talent exist. The part of the game that isn’t played 5 on 5. These things matter. It’s not at the level that possession matters in the long run, but it’s a vital part of a team’s success. Factoring those two pieces creates a better measure of future success and arguably a team’s true talent.

That’s what I aimed to do this summer, put the pieces together in a way that makes sense given what we know about the game right now. (I swear it was to further the discussion and not for gambling).

Before I get into the details, this is an infographic I put together based on one created by Chris Boyle for Fenwick Close. It has every team from the last five years and how did they did in the playoffs, or if they even made it. It’s pretty clear that being good by this stat usually means a good team on the ice (click to enlarge) and the distinction is more clear here than it is with Fenwick Close:


The equation isn’t too complex:

POP = (Fen-Close + (0.6(Goals-Close/PDO-Close))) x (PP% + PK%)

The reason to use each individual stat was simple.

Fenwick Close: The best indicator of future goals and a very good proxy for scoring chances/ puck possession. It’s the most important thing to be good at.

EDIT: We now know that Score Adjusted Corsi is a much better predictor then Fenwick Close, however as you’ll see below, subbing it in in place of Fenwick Close has a very minimal effect.

Goals Close/PDO Close: Not as important as possession, which is why it’s weighted less. By measuring how much a team actually outscores their opponent, you get a decent proxy for a team’s shooting and goaltending ability. This is among the hardest things to measure because it’s inflated by luck and small samples, which is why I divided by PDO (doesn’t do much, but I think it helps in smaller samples).

PP+PK: Special team goal differential is an often ignored part of the game. Adding the two efficiencies together is a decent proxy for that, but it correlates even better with PPGF/60 / PKGA/60. Having special teams talent is a real ability, but again, luck of the small sample size plays a role. The reason I multiply it is because the difference in special teams ability will only matter the farther a team is from average.

Being good at all three is what makes a team elite. Look back at the last President’s Trophy winners and Stanley Cup champions and you’ll find that they’re often there because they’re really good at all three things.

There’s teams that obviously excel with just two, but they don’t succeed as often as their three-tool brethren. Especially in the playoffs.

POP was originally meant to be a predictor for the crapshoot that is the playoffs (Playoff Output Projection). In that, it has done reasonably well over the last seven years (as you probably saw from the graphic above). It does slightly better when weighing the last 25 per cent of the season more heavily (although that might just be an effect of the Kings two Cup wins).

Screen Shot 2014-12-08 at 1.41.04 AM

Part of its playoff accuracy could be a string of good fortune, maybe, but it’s been very good over the past seven years, better than any other measure.

But success in the playoffs isn’t the only thing it can predict. POP does a very good job of predicting who will get there in the first place, again better than any other measure. And that’s because it correlates very well with regular season points.

Screen Shot 2014-08-27 at 7.41.22 PMUsing POP, it’s a lot easier to suss out fake success brought on by a heavy dose of luck. The teams that get by on an unsustainable goal scoring ability fool the standings, but they rarely last in the playoffs, and will usually have a POP rating that reflects their ability.

Something else I wanted to look at was how it does during season compared with the final result.

Here’s POP’s correlation with a team’s final points percentage for this year (I didn’t have regulation win percentage on hand, but the correlation would likely be higher).

Screen Shot 2014-08-27 at 7.43.24 PMAfter only 20 games, POP was able to “explain” about 70 per cent of the end of the year standings. Just 20 games in, using only POP, you would be able to make a very reasonable projection for what the end of the year would look like.

EDIT: Some people on twitter dot com pointed out that it would be better to show correlations to future points rather than end of the season points, so I did that here. For context, I did the same test with Score Adjusted Corsi. The caveat here is that it’s only one season worth of data, but it still did better in that season, and I would suspect that trend would continue in other seasons. I also subbed in Score Adjusted Corsi for Fenwick Close. The difference is negligible it seems. It’s seems slightly better in the first half, and slightly worse in the second. 

Screen Shot 2014-12-07 at 5.14.36 PM

At the end of the year, you’d see that the top teams in POP make the playoffs at a very high rate, more so historically than Fenwick Close.

Screen Shot 2014-08-27 at 7.48.37 PM

And all of this is the point of analytics; using every important data point available in order to accurately forecast the future and further the understanding of what’s being analyzed. It’s not about what’s already happened, it’s about using what happened to show what will happen next. Ignoring useful information seems to go against that.

This isn’t meant to be some magic formula that encapsulates the entire sport. What it’s meant to be is an indication of the things that matter the most in hockey: possession, talent and special teams.

Combining those three things into POP creates, in my opinion, the current best indicator of true talent at the team level in hockey, and the best predictor of future success.

For team’s current POP, check back here every Wednesday for updates. For a look at every team’s POP from the past, go here.




13 thoughts on “why possession isn’t everything”

  1. Out of curiousity, why did you weight the Goals Close/PDO Close at 0.6? Was that number chosen at random?

    Really really enjoyed the post, one of the best reads I’ve had in a while

    1. At first it was an arbitrary number decided by “how much more should possession be valued in comparison to talent” and that was the first thought.
      I tested it against playoff series with other weights and that one worked the best.
      I then started looking at correlations between fenwick and goals/PDO with playoff wins. It was noisey so I put them into seperate buckets and the r2 for goals was just about 60% as good as the r2 for fenwick if I recall correctly. So a lot of blind luck haha.

      The other reason is that by weighting it by 0.6 and adding it to fenwick close, the average team would be at 80. Elite teams at 90. Poor teams at 70. Makes it easier to differentiate between teams

  2. given recent work suggesting score-adjusted corsi is a better predictor of future success than fenwick close (due to fenwick close omitting valuable data points), have you revisited POP at all, using score adjusted corsi (or fenwick) in place of fenwick close?

say something

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 )

Google+ photo

You are commenting using your Google+ 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 )

Connecting to %s