Working backwards to figure out the Cup winner

Updated: January 11, 2018 at 2:05 am by Christiaan Conradie

Arguably the most basic question in hockey analytics will probably always remain the same: can we use (a given stat) to predict how well (a team/player/lineup) will perform the future? As such, predicting the Stanley Cup champion is a yearly exercise many of us try to do, often testing out a number of hypotheses by mixing in different variables.

The purpose of this article is not so much to figure out which teams could have been expected to have done well, but which ones should not have.  As such, I calculated each teams’ goal and shot differentials (the raw data can be found here).

While betting on who we want to win is a lot more rewarding should that team win, it is not an effective strategy in the long run (as my wallet can attest from a couple lost bets due to bias). I, among others have been experimenting with possession stats a fair bit, but this (and this) post inspired me to do a post-hoc analysis for the past 3 seasons keeping it simple: using only shot and goal differentials.

This is the 2015-2016 data in a scatterplot:

We can see that teams in the:

• Bottom right had a positive shot differential and a negative goal differential (Quadrant 4)
We know that teams who shoot more tend to score more goals, just like teams who allow fewer shots allow fewer goals. The scatterplot backs this up too; all but 9 teams fall in the “positive shot and goal differential” or “negative shot and goal differential” quadrants. Generally we see that teams who have a higher shot differential should have a higher goal differential. Teams in the top right and bottom left quadrants are where they should be. They may have had some luck one way or the other, but not to much that they would fall into one of the other quadrants.

The problem with this is that we still have to attempt to explain why 9 teams fall in the other 2 quadrants. Could it be luck? Let’s look at the top left quadrant. These are teams who had a positive goal differential but a negative shot differential. If it is true that teams who outshoot their opponents should also outscore their opponents, then teams in this quadrant got lucky. Can we prove that? The 4 teams in that quadrants are the Rangers, Islanders, Panthers and Blackhawks. Let’s look at their shooting percentages:

• New York Rangers – 9.97% (2nd)
• New York Islanders – 9.41% (8th)
• Florida Panthers – 9.84% (5th)
• Chicago Blackhawks – 9.37% (9th)
All top 10. The lucky hypothesis isn’t completely unfounded. Let’s throw in save percentage to see what PDO looks like for these teams:

• New York Rangers – 101.3 (3rd)
• New York Islanders – 101 (5th)
• Florida Panthers – 101.5 (2nd)
• Chicago Blackhawks – 101.1 (4th)
Out of the 5 luckiest teams last season (according to PDO), four of them are in my “you got lucky” category. All 4 of these teams made the playoffs. How lucky did each of these teams get? The average shooting percentage for all teams in the league last season was 8.98%. Of course there is some variation, but in the long run they will all regress to this number. Assuming the league average shooting percentage:

These are the five teams who have the largest difference between predicted and actual goal differential. As we can see all four teams in questions now have a negative goal differential. As for how these teams fared, the Rangers and the Hawks lost in the first round. The Islanders beat the Panthers in the first round and went on to get demolished by the Lightning in the second round. In other words, all these teams lost the first chance they could. Of course we know all of this already, but we could have predicted this with a fair amount of certainty that these teams were lucky (high PDO), and regression waits for no man. Joining these teams on this unfortunate list are the Caps, even though they aren’t in the quadrant in question. They have the largest difference between predicted and actual goal differential with 41! Since they are not in the quadrant in question I won’t say more about them for now. I look look at the other quadrants in Part 2.

Let’s look at the 2014-2015 season:

Note : The Sabers do not appear on this graph. Their shot differential was so atrocious (-936) that I had to cut them off to make the graph legible. They are still included in R2, their omission is purely for aesthetic purposes.

Three teams in quadrant 1 this time, the Flames, Senators and Canadiens. All of them made the playoffs. Their shooting percentages:

• Calgary Flames – 10.52% (2nd)
• Ottawa Senators – 9.14% (14th)
• Montreal Canadiens – 9.17% (9th)
And their PDOs:

• Calgary Flames – 101.6 (4th)
• Ottawa Senators – 101.2 (6th)
• Montreal Canadiens – 101.7 (2nd)
Let’s see what about the differences between actual and predicted goal differentials.The league average shooting percentage for 2014-2015 was 8.9%. Assuming that:

All three Quadrant 1 teams fall in the top-6 and again there is a substantial difference between predicted and actual goal differential. As for how they fared in the playoffs, the Flames beat the Canucks and went on to lost to the Ducks in the 2nd round. The Canadiens beat the Senators and went on to lose to the Lightning in the 2nd round.

Last but not least, the 2013-2014 season:

Five teams in Quadrant 1, the Avalanche, Canadiens, Blue Jackets, Wild and Flyers. All of them made the playoffs. Their shooting percentages :

• Avalanche – 10.12% (2nd)
• Blue Jackets – 9.3% (11th)
• Wild – 9.13% (13th)
• Flyers – 9.36% (9th)
A little more variation here. What about PDO?

• Avalanche – 102.1 (2nd)