Another weekend, another couple of ugly losses for the 2021-2022 New York Islanders.

The Buffalo Sabres are on a tear since beating Jack Eichel and the Vegas Golden Knights in February, so the Islanders giving up five goals and only managing 22 shots on net isn’t too embarrassing, right?

And the Carolina Hurricanes are one of the best teams in the NHL, so going into the third period tied, only to lose 5-2 isn’t that bad, right?

Don’t kid yourself.

It’s time for the third edition of Analytics Toolbox, where we will learn about PDO and luck in the NHL. Maybe this bad season was nothing more than an accumulation of bad bounces?

Let’s dive in.

What is PDO?

PDO is the sum of save percentage and shooting percentage. For a team, this means the average save percentage of all their goalies (total saves divided by total shot faced) plus the teams average shooting percentage (total goals scored divided by total shots taken).

League average PDO is always 100 because shooting percentage and save percentage are intrinsically linked. A shot either goes in, or it doesn’t, so the sum of the two percentages on a league wide level will always equal 100. Think about it this way — in a single season, if 5,000 goals are scored on 50,000 shots league-wide, that means league average shooting percentage is 5,000/50,000 = 10%. Accordingly, league average save percentage is (50,000-5,000)/50,000 = 90%. 90 plus 10 equals league average PDO, 100.

Alright great. Simple percentage calculations. Why do I care about this?

Simply put, PDO measures PUCK LUCK. A PDO greater than 100 is interpreted as lucky, whereas a PDO less than 100 can be interpreted as unlucky.

Some of you may not be convinced — you may be thinking that some teams have very talented shooters and goaltenders. Naturally, we would expect these teams would have high PDOs, so we shouldn’t disrespect them and call it luck!

You are correct. For PDO to be a true indicator of luck, it would have to be entirely random. This is not quite the case – PDO is almost entirely random, but a tiny correlation exists between teams PDO’s one year vs. the next. Look at this figure from JFresh.

The blue data series represents a teams PDO in “Year 1” and the red data series represents a teams PDO the next season. The scatter of the red data points around the blue indicates that yes, for the most part, PDO is incredibly random. However, both the “Year 1” and “Year 2” trendlines have a negative slope. This was obviously expected for the “Year 1” data, as it is sorted from Highest PDO to lowest PDO, but if PDO was completely random we would expect the “Year 2” trendline to have a slope of zero. We can ascertain that a small correlation exists between “Year 1” and “Year 2” PDO based on the consistent negative slope we see in both trendlines.

The trend in the PDOs indicates that shooting talent and goaltending talent do exist, but hockey is a game of bounces to the extent where their effect on long-term results is pretty minimal.

I’m going to go on a nerdy vent about confidence intervals below (please feel free to skip it):

JFresh does not report the confidence interval on the slope of the “Year 2” trendline. As a huge geek, this frustrates me immensely. 95% confidence intervals are standard in statistical analysis. Without getting too much into the weeds, a confidence interval gives us a range of possible slopes, rather than one number. A 95% confidence interval gives us “a range that we can be 95% confident the slope falls within.” If the confidence interval for the slope of a regression line contains zero, we cannot be confident that an actual trend exists — there is a reasonable probability that the true slope is zero. Based on the randomness of this data, I would not be surprised if that is the case here. But let’s give JFresh the benefit of the doubt. Perhaps he checked the confidence interval but didn’t want to write a whole paragraph about it and expose himself as a major nerd.

Alright, sorry about that. Before we jump into some Islanders’ talk, let’s briefly discuss player-level PDO. PDO for players utilizes on-ice save percentage an on-ice shooting percentage. We learned about on-ice metrics in Analytics Toolbox #2, so you can probably guess how these are calculated. A player’s on-ice save percentage is the average of their goalies’ save percentages when they are on the ice, and their on-ice shooting percentage is the average of their teams shooting percentage when they are on the ice.

Player PDO interpretation is like team PDO interpretation — greater than 100 is typically thought of as lucky, and less than 100 is unlucky. The “little bit of skill, whole lot of luck” principle applies as well — specifically, some very talented players can have an impact on their on-ice shooting percentage, but the repeatability of this impact is small.

Islanders Time

Can PDO explain the Islanders’ tough season this year? Are they a team that is having a down-luck year? Can we expect them to regress to their form of the last few seasons if management decides to run it back with this group next year? The short answer is no. The figure below shows the Islanders PDO in the Barry Trotz era — as of Saturday.

The 2021-2022 Islanders are right about where you would expect them to be based on the last few seasons — holding an above-average PDO on the back of their goaltending performances. The figure below breaks down the Islanders PDO into shooting percentage and save percentage and compares them vs. the league average.

The Barry Trotz Islanders seem to be a team more focused on good goaltending and/or limiting high quality chances than sniping and/or creating high quality offence. Their team save percentage is significantly above league average the past few seasons, and their team shooting percentage is significantly below.

The problem for the Islanders? They are still getting great goaltending and not scoring on a high percentage of their shots. This year does not appear to be an anomaly.

So, why has this season been such a mess? There are probably many answers to that question, with varying degrees of credibility.

I’ll propose an answer myself. Well, maybe not an answer, so much as a data-driven observation. The figure below shows the difference between the Islanders’ shots against and shots for at 5-on-5 during the Barry Trotz era.

The Islanders are giving up a lot more shots compared to their opportunities this season. The difference between this year and the average of the past three years is approximately 2.4 shots per game played. I’ll spare you the math, but trust me, that this works out to about 14-15 more goals against (or goals not scored, because it is a differential) over the course of the regular season. The Islanders have a penchant for one goal games, so those 14-15 goals could be the difference between a win and a loss. Imagine if that differential shrunk by half or didn’t exist at all? What would the standings look like then?

Based on the above analysis, it looks like parts of the Islanders’ game this season have changed for the worse. I’m not going to claim I know the root cause of the issue, but I do know they are giving up more than they are generating at a rate much worse than the past few seasons. In my opinion, management should not bring back this roster “as is” next season if they hope to regain their form of the last few years.

Regardless of how it’s done and who takes action, the New York Islanders need to be fixed before next season.

John Smark