NHL Equivalency and Prospect Projection Models: Introduction (Part 1)
Making Use of the Best Information Available
At the 2013 NHL Entry Draft, the San Jose Sharks traded the 20th and 58th overall selections in said draft to the Detroit Red Wings in exchange for the 18th. San Jose would go on to select Mirco Mueller with the 18th, while Detroit would select Anthony Mantha with the 20th and Tyler Bertuzzi with the 58th.
I don’t have psychic vision, so I had no idea that Mantha and Bertuzzi would both become very good NHL players, or even that both would be available with the picks San Jose surrendered. But as a Sharks fan, I still hated this entire ordeal from the moment it happened. The trade itself was a blatant overpayment, as the 58th pick is far more valuable than the gap between 18th and 20th. But I couldn’t even focus on the trade because I was so unhappy with the selection we made.
Mirco Mueller was fresh off a draft year which saw him score 31 points in 63 games with the Everett Silvertips of the Western Hockey League, good for a points per game (P/GP) of 0.49. I had never actually seen him play, but I knew that greatness seldom came of defensemen who scored less than half a point per game in the Canadian Hockey League (CHL) in their draft year, and I didn’t think it was wise to spend a 1st round pick in the hopes that he would buck the trend.
Eight years later, I’m mostly vindicated. To Mueller’s credit, he’s not a complete bust; he’s played in 185 NHL games and provided just over 1 Win Above Replacement (WAR) according to my model. But he’s also 26 years old, he also spent the entirety of the 2020–2021 season in Sweden, and he’s also signed up to spend the entirety of 2021–2022 in Switzerland. His NHL career is almost certainly done, and he hasn’t even played 200 games. This pick wasn’t the worst of all time, but the final verdict is clear: Selecting Mirco Mueller with the 18th overall pick was not a success.
Those familiar with me may be surprised to hear that I used to care all that much about how many points a defenseman scored in any league. They may be even more surprised to know that I still care. If I’ve consistently advised against using points to evaluate NHL defensemen, and even gone on record saying that I don’t consider Tyson Barrie — the NHL’s leading scorer among them in 2021 — a good defensemen, then why do I care about points scored by defensemen outside the NHL? The answer is two-fold:
- I care about the best thing we have. The NHL provides the public with a wide array of information that can be used to build metrics that are far more valuable than points, both for the purpose of describing past performance, and the purpose of predicting future performance — especially for defensemen. Most other leagues don’t.
- I hold an underlying assumption that in order for defensemen (and forwards) to succeed at either end of the ice in the NHL, they need some mix of offensive instincts and puck skills that’s enough for them to score at a certain rate against lesser competition.
Reason one is straight forward and tough to argue against. In my latest article, I not only documented updates made to my Expected Goals and WAR models, but also performed a few tests to validate whether WAR is superior to points for the purpose of evaluating NHL skaters from a descriptive and predictive standpoint.
If you haven’t read it already, I suggest you do so. But I also understand if you don’t want to, in which case the results below are still enough to show the superiority of WAR to skater points as a descriptive metric:
On the predictive side of things, these test results also display the superiority of WAR:
These tests also validated was that points are not completely useless, or even that much worse than WAR. In the absence of WAR or a similar metric, points are far more useful than nothing. They’re actually even more useful at the NHL level than I thought they were when I initially went into this project.
Reason two is a bit more dicey. It’s an assumption I made after reading some fairly rudimentary analysis back in 2013, and one I’ve since used confirmation bias from Mirco Mueller’s career to hold on to. The analysis which spawned this assumption is also not unlike the majority of draft research I’ve read in that it uses games played at the NHL level as the measure of success. Games played are far from a robust or accurate measurement of a player’s contribution to their team, and I’d argue that selection bias makes them even worse than points.
Either way, both metrics have their flaws, and two months ago I decided the time was long overdue for me to challenge my pre-conceived notions and determine just how well scoring outside the NHL can predict a player’s true contributions at the NHL level. I did so with a two-step research project:
- Build an NHL Equivalency model in order to determine the value of a point in over 100 different leagues that feed the NHL (either directly or indirectly).
- Build a predictive model that uses a player’s NHL Equivalency and various other factors such as height, weight, and age to determine the likelihood that a player will become an above replacement level player and/or a star in the NHL according to my WAR model.
In part 2 of this series, I document the process of building the NHL Equivalency Model.