Luck In The NHL Playoffs, And How We Can Measure It


Image: Youtube

Sports and luck go hand-in-hand, especially in hockey. The nature of the sport—relatively low-scoring, fast-paced, involving high degree of parity—maximizes the effect luck can have on results. The NHL playoffs involve far more luck than most experts care to admit.

This is demonstrably true in individual games, which are often won on exceptional luck. Caps fans will remember Lars Eller’s 2OT winner in Game 3 against the Columbus Blue Jackets last year that bounced off his skate, then off a defenseman, then his skate again before going in. Evgeny Kuznetsov’s series winner against the Pittsburgh Penguins came seconds after Pens forward Tom Kuhnhackl saw his one-timer beat Braden Holtby but hit the left post, an inch away from sending the series to a seventh game. The effect of lucky goals or events is somewhat mitigated by seven game series, but luck still plays a large role in determining who advances.

The good news is that, to a certain extent, luck is actually measurable in hockey through an advanced stat known as PDO. PDO (not an acronym) is the sum total of a team’s shooting percentage and save percentage over the course of a period of time; in other words, it’s the percentage of shots that meet the desired outcome for a certain team. PDO theoretically averages at 100, with a mark over 102 considered exceptionally lucky and under 98 exceptionally unlucky.

Admittedly, at first glance, that doesn’t sound like a measure of luck. But the hockey analytics community has long since accepted PDO as luck-based for several reasons.

First and foremost is the number of goals that go in due to sheer luck. Goals like Eller’s in Game 3 last year are anomalies, but they’re surprisingly common ones. This means that one must analyze a massive sample size to minimize the significance of goals like that on save and shooting percentages. No reasonable viewer could pin the blame for Eller’s goal on Blue Jackets goalie Sergei Bobrovsky, but that goal alone reduced his playoff save percentage by roughly 0.1% from .900 to .899. That doesn’t sound like a lot, but it adds up when you take into account how many goals are reliant on luck.

Along those lines, NHL goalie parity emphasizes the effects of lucky goals on PDO. Save percentages have relatively little variation, even between the best goalies and the worst goalies, over time. This is especially true when measured at only 5v5, as PDO typically is (special teams disproportionately damages goalies’ save percentages and inflates players’ shooting percentages). Among NHL goalies who played 1500+ minutes this year, there was only a four percent difference in 5v5 save percentage between the best and worst marks. That’s why a 0.1% drop from one uncontrollable goal makes a notable difference.

Shooting percentage is even more luck-based. So many shots go in (or don’t) based on luck that massive sample sizes are required to achieve any statistically significant mark. Thus, any single game shooting percentage—in fact, any single season shooting percentage—is going to be affected largely by luck.

Why does all this matter? Well, PDO is more predictive of single game outcomes than any other advanced stat. Only one of the Caps’ 24 playoff games last year went against what PDO would suggest (PDO>100 would be an expected win, PDO<100 a loss). PDO is three for three this year. This does NOT imply causation and is only descriptive, but it does show how much we can tell about a game just from how lucky teams get.

To be sure, there’s an element of “creating your own luck” involved here that’s important to consider. Eller’s goal isn’t possible without Brett Connolly sending the puck toward the net, as NHL players do dozens of time per game in hopes of getting that lucky bounce. But that’s why we analyze stats like shots, Corsi for (shot attempts), expected goals, and scoring chance percentages—all of which are predictive, but not as much so as PDO. So it appears that though a team can help create its luck, it can’t control how much luck the hockey gods dole out, especially on any given night.

Nowhere is the effect of PDO more evident than in the Caps’ current series against the Carolina Hurricanes. The Hurricanes, in theory, should create more luck than nearly any other team in the league given their superiority in possession and shooting analytics. Based on those numbers, the Caps performed similarly poorly in their Game 1 4–2 win and Game 3 5–0 loss, which is obviously counterintuitive given the different scorelines. The only difference? A 100+ PDO in the win and an abysmal 91.4 in the loss.

I recognize that it’s not much fun to just say that the NHL playoffs are all luck, and in fact that’s not what I’m saying. But I think that there’s more luck involved than we care to admit. There’s a balance between rewarding better teams and unpredictability that is hard to strike, and at times hockey leans too far into unpredictability.

-Tiger Bjornlund

View this post on Instagram

#Back2Back #ALLCAPS

A post shared by NoVa Caps (@nova_caps) on

About Jon Sorensen

Jon has been a Caps fan since day one, attending his first game at the Capital Centre in 1974. His passion for the Caps has grown over the decades, which has included time as a season ticket holder, social media and community organizer, and most recently led to the founding of NoVa Caps in 2014. Jon earned a Bachelor's of Science in Engineering at Old Dominion University, and is a Systems Engineer during intermissions, which has been instrumental in supporting his Capitals habit.
This entry was posted in Data and Analytics, News and tagged , , , , , . Bookmark the permalink.

Leave a Reply

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