Over the last few months I have seen quite a few discussions on whether or not the Islanders is a mirage. Some say good possession numbers equal a good team and those same people often say PDO is a measurement of luck. I honestly didn't know what to think. This is why I decided to take a deeper look.

I may not have been an advanced stats expert going in to this analysis, but I am pretty good with numbers, and I really find the conclusions interesting, which is why I have decided to post it.

Be aware that all graphs have (,) instead of (.). That's not a mistake, that's just the Danish way to do it.

I am not going to look at individual stats. I simply want to look at what defines a good team in terms of statistics. In order to get the best possible result, I have decided to look at every game played in the NHL going back to the season 2007/2008 and up until January 9th 2019 (13,730 games). That's a very large dataset, so the trends you will see should be well documented.

I'm looking at all strengths (not just 5v5) as I want to see what defines a great team in all situations.

So let us start by looking at records in this time frame:

Not much to say about that. It's been some rough years as an Islanders fan. They are ranked no. 27 out of 32 teams (I have counted Atlanta and Winnipeg as two separate teams). It should be noted that in all the graphs below I have not included Vegas, as the dataset is too small. There's only 3 teams under .500 and the league average point% is .558.

Now we will dig a little bit deeper and look at each team's corsi numbers (ratio between shot attempts taken and shot attempts allowed) and PDO (Shooting% + Save%). There has been taken 828669 shots and the average Shooting% is 9.108 meaning the average Save% is 90.892. The goalkeepers average Save% is higher than this, since empty net goals are accounted for in the teams Save% and PDO.

The teams are in order of CF% and again we find the Islanders in the bottom part.

Now it starts to get interesting, because now we can test if possession numbers actually equal winning. So I've made a graph with point% vertical and CF% horizontal. If there's a direct correlation you should see a straight line (perfect correlation means R2 = 1).

1

The good corsi teams tend to be winning more, but there's definitely more to it than that. How is it then with PDO? Does that correlate with winning?

2

Clearly it's a good thing to have a high PDO. It will win you matches, but it's not exactly a perfect match either. The fact that good goaltending and good goalscoring wins you hockey games is probably not a revolutionairy discovery though.

So it's obviously a good thing to be high in both PDO and CF%. That is a difficult task though, since the two in many cases work against each other. If you take a lot of shots from the outside, then you will have good corsi numbers but a bad PDO. If you only take quality chances, then it will be the other way round.

The next graph shows PDO as a function of corsi, so you will want to be in the upper right corner. I've marked the great teams (above .600) with orange.

3

Up until now we have treated all shot attempts as equal, but what happens if we look at the shots in terms of danger. Natural Stat Trick divides shot attempts in to 3 categories -€” high danger, medium danger and low danger. On average it requires 7.10 high danger shot attempts to score one goal, it requires 21.35 medium danger shot attempts to score a goal and it requires 50.80 low danger shot attempts to score one goal.

Knowing this we can calculate the expected goal differential (xGDF) based on the amount of high-, medium- and low danger shot attempts for and against. It turns out, that even if we do this, there's no great correlation between the expected goal differential and the teams quality (P%).

4

Just in case you were wondering, there should be a near perfect correlation between goal differential and winning. The next graph shows just that. It's point% as a function of actual goal differential per game. The top teams outscore their opponent by more than 0.4 goals per game.

5

So even if you account for shot danger, shot attempts (corsi) is not a good way to evaluate the quality of a team. There's clearly more to it than just outshooting your opponent. So what happens if we calculate the expected goal differential based on shot danger as well as PDO.

6

Now you see a good correlation. So getting a lot of chances is not enough. You also have to get good scoring and goaltending.

This is perhaps not groundbreaking news, but it's still interesting since many people look to CF% or HDCF% for answers, when in fact those metrics only tell you half the story.

The problem with PDO is that it's difficult to predict. You can calculate an expected PDO based on career numbers, but it's not as easy as it may sound. Instead I will take a look at some numbers from Corsica.

Let us start by looking at Corsica's expected goal differential (xG+/- on the homepage). That stat is based on number of shot attempts and the quality (angle, distance, rebounds and so on) of those shot attempts. It's not based on the quality of the shooter or the quality of the goaltender.

So let us see how well Corsica's xGDF correlates to winning hockey games.

7

As we have shown earlier, you can't ignore the quality of the shooters and the quality of the goaltenders. This means that you can't compare Corsica's expected goal differential to actual goal differential. If someone does that, they are simply wrong unless goaltending and shooting is exactly average.

So what happens if we combine Corsica's xGDF with Corsica's WAR(shooting) and WAR(goaltending).

8

This is based on a conversion rate of 4.5 between WAR and GAR, which is standard. However, if we use 5.3 the result is better (graph below). This indicates that the WAR numbers are correct relative to each other, but should probably be a bit higher.

9

The last thing I will look at in this post is WAR(total). Instead of looking at xGDF combined with WAR(shooting) and WAR(goaltending) we will simply compare the WAR(total) of each team with team quality.

10

This means that WAR is actually a pretty good indicator of team Quality. It should also be noted that the average WAR(total) is not 0, it's 0.113 per game or 9.27 per season.

Conclusion:

- CF% and HDCF% are not good indicators of a team's quality. They tell half the story, but you also have to factor in quality of shooting and quality of goaltending.

- PDO does not equal luck. Looking at 11+ years of data, some teams have a good PDO and some teams have a bad PDO. If the sample size is small, then luck can play a factor.

- WAR is a pretty good indicator of team quality.

That's it for this post. I hope at least a few of you found it interesting, even though it's quite nerdy. In the next post I will look at Islanders' WAR numbers this season and compare them to expected WAR numbers based on career metrics. Spoiler alert: Islanders are outperforming their career numbers, but not by as much as one could fear.

All data is based on stats from Natural Stat Trick and Corsica.