With about a quarter of the season done, we’ve seen a good sample of play from all thirty teams. Already there are many surprises — what is going on in San Jose and Los Angeles? Are the Predators and the Islanders really this good? We know that extrapolating from the current standings is a mistake, since winning predicts future winning only very weakly. The best predictor we have so far is even-strength puck possession, so I have trained a simple model using 5v5 score, venue, and schedule adjusted shot attempts, for and against, with which to simulate the remainder of the season.
A hundred thousand simulations after each game gives a decent idea of how teams will end the season, based both on how many points they’ve already picked up and how well they’re playing the most important facet of the game, that is, at even strength.
Atlantic division: (Click to enlarge)
The dotted line represents a make-shift playoff cutoff. By averaging the 8th and 9th place conference point projections, we obtain a number that will accurately represent the number of points required to make the playoffs, barring very unlikely scenarios where one division has six or more teams above the cutoff while the other only has two or fewer. As of yesterday, the Atlantic appears to be the stronger division, taking five teams to the playoffs. Montreal have been fortunate so far and, while they are projected to finish third, they aren’t the league-toppers that they appear as now. More interesting is the strong play of Florida, who have projected as a bubble team all year, which is a nice change from the past few years. Buffalo remain in a class of their own.
The real surprise here is the Islanders, who have won five straight and has many people wondering if they are the real deal. Their 5v5 play backs up their record, though, and then some, they are the current model favorites to win the league, if they keep on owning the puck like they have. As of today, they are the best puck-possession team in the league. Many analytics folk thought that NYI had a whale of an offseason, and so far they’ve been right.
Carolina have had much less reward for effort than one might expect, and the rangers as a bubble team is a falter that only some predicted.
After the islanders, the other “where did they come from?” team this year has been Nashville; their underlying numbers suggest that their success so far this year is no fluke. The central division as a whole, however, is incredibly strong, boasting five of the top-nine possession teams in the league. The only reason they project as “low” as they do is that they have to play so many games against one another. Dallas and Colorado are only a point apart in the standings, but their play has been worlds apart—Dallas very unlucky and Colorado simply very poor. Winnipeg quietly building a very serious season in the shadows.
Initially very jumbled, already there is a little separation in the projections for the Pacific teams. Vancouver and Anaheim are the class of the division and appear locked in a very entertaining fight for top spot. Calgary are riding a massive wave of luck but early points are extraordinarily handy and they project as a bubble team despite very poor possession. The issues in Los Angeles and San Jose are real, but not quite as severe as they
appeared at first. Edmonton is a mirror image of Calgary, being sunk by very bad luck, especially goaltending, but already the losses take a big toll.
They aren’t playing like McDavid candidates (nobody but Buffalo really is) but they already face a huge hill.
In understanding these plots, it’s as important to consider what the projections exclude as well as what they do include. They don’t include explicit goaltending terms, which are notoriously hard to predict. Implicitly, this means that all of the teams are expected to have league-average goaltending.
If your team is especially strong in goal (like the Rangers, say), you should mentally round up their projections; if your team’s starter is one of the worst starters in recent memory (like the Jets, say), you should mentally round them down. If a previously ordinary goalie has a Vezina-type year, as happens every year, the model will not predict that. Shooting percentage, which is even harder to predict than goaltending, is not included. Injuries and trades, of course, are not included. And, more subtly, some teams are playing so well or so poorly at even strength that it’s probably safe to expect that they’ll move back towards normalcy a little, this effect will bend the teams at the edges of the plots back down towards the center, and that’s more fun for everybody.