The WAR Lineup Creator
A couple of years ago, the incredible Domenic Galamini created a Google sheet that allowed users to create custom rosters. This tool then calculated projected rates of shots for and shots against and offered a prediction on whether or not this team would make the playoffs at the end of an NHL season.
As has become common for me in my hockey stats life, I was inspired by Dom’s work and decided to create a similar tool. In this piece, I’ll explain the basics behind how I pulled together the data, how to best use the tool, and some ideas for where this project goes next.
Let’s dive in.
The Stats Behind the Scenes
The EvolvingWild twins, Josh and Luke Younggren, have created an excellent stats website called Evolving-Hockey. In addition to providing access to counting stats (goals, assists, et al.) and advanced stats (Corsi For per 60, individual expected goals, for example), Evolving-Hockey also includes WAR data. In the most brief terms, WAR (Wins Above Replacement) is a single-number measure of a player’s overall impact relative to the impact of a “replacement player” — a theoretical next-player-up from the AHL, with zero acquisition cost for the team. You can read all about the stat for yourself in their multi-part series.
No reasonable analyst will argue that understanding any player can or should be reduced to the WAR value without any further context and examination, but this number is extremely handy for beginning conversations about how a player has impacted the game in the ways we can measure.
So, the WAR stat by EvolvingWild is the backbone of the lineup creator tool. But there’s a problem: WAR is a descriptive, not predictive, stat. This means that WAR measures the contributions a player has already made (describing a player’s past production) and isn’t designed as a predictive stat, as constructed.
So, I’ve worked to get around that.
Here’s a snapshot of the source data I’ve derived from EW’s WAR stat, which operates behind the scenes of the WAR Lineup Creator tool:
For anyone who has already used the WAR Lineup Creator, you’ve likely noticed that there are a variety of tabs at the bottom, allowing for multiple, simultaneous users. However, a few tabs are locked — one is the backup file, which I use to replace sheets as they break (I see you, user who loads up the lineup with Justin Abdelkaders, Alex Petrovics, Garrett Sparks clones, and then changes all of the fonts).
One tab is for the source data, pictured above. This allows users to check in on the WAR rate per minute that I’ve assigned to each available player, along with other details like cap hit, team, age, and projected lineup positions (in most cases).
How did I arrive at these figures?
First, I created Marcels for all players who played in the NHL last year. A Marcel is a stat-weighting process, originally conceived by Tom Tango here. The idea is that the most basic stats projections in sport should be based on weighted priors. I’ve used EvolvingWild’s WAR scores from 2016–17, 2017–18, and 2018–19, weighted as 2–3–5 (Tango actually suggests weights of 3–4–5 but I’ve decided to give a little more emphasis to last year’s performances, which is a judgement call on my part). For seasons in which a player did not have data, I awarded the player a league-average WAR rate. Again, this is a judgement call with implications, giving young players without NHL experience (or data) the benefit of the doubt and, to an extent, potentially inflating their projections.
Then, I aged each player by one year, and applied an aging impact based on EvolvingWild’s write-up here. This satisfies the second tenet of Tango’s basics for creating stats projections: applying some aging effects.
Last, I produced a WAR per min projection, based on each player’s past TOIs. By breaking down to the per-minute level, the lineup creator is able to handle changes in TOI% per line or pair, which I’ll discuss further below.
For rookies and prospects who played outside of the NHL last year, I’m standing on the shoulders of the incredible Emmanuel Perry. His prospect data can be found here. I’ve drawn (almost all) players who seem like they might make an NHL roster next year, broken down their projected WAR/82 scores to per minute levels, and included them in the data set. These are not at all weighted with priors and Perry’s WAR and EvolvingWild’s WAR are not derived in the same way, so that’s worth keeping in mind for users of the lineup tool.
For a little context, I’ve compared Goals Above Replacement (GAR, which can be translated into WAR by dividing by ~5.5 or so) values for 2017–18 vs 2018–19 without any of the adjustments I’ve made for the lineup creator (Marcels, age, role). We find here that, for regular forwards, 17–18 GAR and 18–19 GAR has an r² of 0.14. There’s relationship between the two values but adjustments are needed to use past figures to predict future figures.
For defensemen, the r² is only 0.08. Again, a little bit of relationship but not lots. Again, this is the case for using priors, age, and some minor role adjustments in order to use WAR in a forward-looking way.
The takeaway? There’s plenty of room for error in trying to project a player’s WAR impact in the future. Usage, team styles, cohesion with linemates, luck, and other factors all play roles.
How Best to Use
First, this tool is designed as a fun, stats-based conversation-starter for visualizing how your team’s roster might look for the 2019–20 season. The lineup creator has lots of flexibility to allow you to change lines/pairs, alter TOI%, make trades and signings, and banish your coach’s favourite fourth-line, mid-30s, no-upside grinder to the minors in favour of that can’t-miss second-round draft pick from four years ago or recent 1-year, $700K UFA-signee who definitely will score 50, 60, maybe 80 points in the NHL if someone would just give Nick Shore a damn chance already.
The point is, this tool is designed to be interactive and entertaining. Recently at The Athletic, a number of writers have used the tool to show projected Opening Night rosters with an estimate of standings points for next year. In another fun spin, some writers have used the tool to create All-Ontario or All-Minnesota teams, which is a really interesting way to create some summer content with a visual hook. It’s great to see the tool used in this way as a way to estimate team strengths and weaknesses.
It’s important to note that the TOI% is editable. The default TOI suggestions are set in the locked backup sheet and I really recommend sticking to those unless you have some clear, compelling reason to do otherwise. Setting your top forward line at 40% TOI or setting your top defense pair at more than 45% TOI will usually pump up the projected stats total but that’s only because the lineup creator doesn’t understand that you’re running your players into the ground and that your best skaters will be injured by November.
One last note on best use: There are a few players who, despite some optimism about their potential for next season, have low (or even negative!) WAR projections. There are two ways to understand this:
a) hockey is hard to predict, Sean Tierney personally hates your favourite team, and your player will definitely, no doubt, 100% outperform the projection. Add 10 standings points to the team’s final results and purchase a comfy lawn chair for the Stanley Cup parade next June.
b) temper expectations. Rookies in their first seasons, veterans who are 30 years old or older, and players joining the NHL from other leagues are associated with lots of risk, in large part because we don’t have NHL data to work with when projecting their future.
In a very interesting way, the New Jersey Devils have tapped into each of these categories.
Let’s have a quick look:
NJD’s top line is projected to be excellent, which makes sense. Taylor Hall, Nico Hischier, and Kyle Palmieri have very positive NHL tracks records of various lengths and, if healthy, the lineup tool believes this line will be one of the NHL’s best top-trios.
But things get interesting, funky, and weird from there.
Nikita Gusev lit up the KHL last year, scoring 82 points in 62 games, his fourth-straight season managing more than a point-per-game with SKA St. Petersburg. But his projections for next year are…somewhat tepid.
Gusev is projected to produce a WAR of ~1.2 over the course of a full NHL season. Depending on the TOI he’s given in the lineup creator, this value can be a little higher or lower. But the point is that, now 27 y.o. and coming to the end of his peak plateau, and coming to North America from a league with a 0.75 NHL equivalency factor, there are some good reasons to be skeptical about Gusev becoming an immediate, high-end NHLer.
He might! But it’s not certain.
Jack Hughes (recent #1 overall draftee), top-prospect Jesper Boqvist, and PK Subban (now 30 and coming off a down campaign), are each interesting cases of WAR projection variability embedded in the tool — on the one hand, young and without NHL priors, on the other hand 30 and on the wrong side of the age curve.
Could each of these players exceed their projections? For sure.
tl;dr
Use the WAR Lineup Creator for entertainment and for making stats-informed projections with rosters based on some of the best data that’s available. But do so with the understanding that there’s lots of variability at play and that hockey is beautifully random, and players can, at any time, over- or underperform these projections.
Next Steps…
If you follow me on twitter, you’ve likely noticed that I’m creating team projections for next season. This tool has a tableau play-out, which you can view here. My intention is to continue adding features, including expected goal rate projections and some goalie stats projections too. If there’s something you see that you think would be a great addition, I’m always interested in feedback.
Enjoy!
You can follow me here: https://twitter.com/ChartingHockey
I’m truly grateful that people are willing to support the work I do. If you’d like to pitch in to support me, you can do that here: https://www.patreon.com/chartinghockey.
