A New Try at Projecting Draft Potential: DRAFTe
Win Shares over Points
By Thibaud Chatel and Cédric Ramqaj
NOTE: This is a version 2 of the model, after revisiting the age adjustment formula. Any results from the first version should have been removed.
Drafting is science. And a bit of luck. Or a lot of luck some people out there will say. Let’s just agree that a lot of things can happen between draft day and becoming an NHL player.
In the past few years, analytics have entered the drafting picture, trying to help scouts quantifying the potential of a player a few years down the road.
It does not replace the scouts. It does not replace the expertise of a professional human on skating, shooting, positioning, compete level, hockey IQ, etc.
It helps put performance into perspective.
It has been documented, mainly in baseball, how organizations combine both information, the stat part and the scouting part, to come up with a unique grade for each prospect.
Multiple models have been built to link the production of a player in a certain league and what it would mean in the NHL. Because two players can be dominating their junior league (Shane Wright) or battling to get a dozen of points in a senior league like Juraj Slafkovský in Liiga. Two very different contexts. So who’s the best player?
All public or semi-private models so far have been using points production, adjusting for age and league. The pioneers Gabriel Desjardins and Rob Vollman were followed by Emmanuel Perry NHLe. CJ Turtoro later published an interesting network approach to NHLe. Patrick Bacon also has a public model available and Byron Bader’s work on prospects has made him a reference in the industry.
So why would we need another model?
Beyond Points: Win Shares
Because we wanted to try a new approach. One that goes beyond points production and includes as many context information as possible.
For the sake of arguing, we would see some limitations in using points based NHLe, without ever forgetting that it is still the main and logical way to project a player. But almost all models are using transition coefficients from players having played in league A and league B during the same season, to avoid contextual variables like age, developing environnement, etc. We experienced in our own research in Europe that using between season transition coefficients (league A in year -1 to league B in year 0) not only brought a much bigger sample of data but we also found those contextual built in variables very telling and part of the story. Basically, a player moving from NHL/AHL to Europe is almost always someone whose NHL dreams are fading away, looking for a top role in a good league with Power Play time. A player moving from ICEHL (Austria) to Switzerland is very likely a top6 guy moving to be a support player there. What we mean is that the notion of “career path” embedded in between season transition coefficients is part of the story and is already included in every reference of your database. In short, the move from one league to another very often fits one same profile of player and it is okay to use that information as we compare apples to apples.
The second reason would be that, as far as we know, the NHLe values developed before were based on any player moving from one league toward to NHL. Back to the previous argument of following a “career path”, we thought logical to create an NHLe metric only based on 18 years old players, comparing that point in time (draft year) and their NHL career.
It is fairly certain that NHL management now has access to analytics and micro-stats for many junior leagues. Sportlogiq covers the CHL and the USHL for example, as do other providers.
One can only imagine that, internally, some organization have developed a WAR like metric to project prospects.
We have been using Win Shares for our work in Switzerland in order to project potential recruits from any league in the world. Win Shares are based on the Point Shares formula detailed on Hockey Reference (basically 3 points is 1 Win). Win Shares include Goals and assists for the offensive input, 5v5 goals differential or the usual +/- for the defensive input.
The good part is that offensive and defensive performance of a player is put into perspective by his Time on Ice and his team’s overall performance. It has been great for us highlighting well-rounded players, defensive defensemen and players surviving in a bad environnement, players that would not look that good by points alone.
But the best aspect of Win Shares is that you can build them for almost any league in the world as long as you have the stats mentioned above, as well as the following information on teams: Games, Wins, OTW, OTL, Goals For and Against, Shots For and Against if possible. It may require an exhausting hunt over the internet to find all the information between Elite Prospects, the league’s websites and sometimes media sites but so far we always got what we needed. Let’s note that we put all leagues into a 3-2-1 points format so 3 Point Shares =1 Win shares for any player anywhere.
So far, we re-created Win Shares for almost 300 seasons of hockey, going back to 2007-08 for all the leagues we could. Let’s give a huge shoutout to pick224.com for the data available there and the scrapping work that enabled us to get 5v5 goals differential and eTOI (estimated Time on Ice) for many junior leagues.
Draft Year Win Shares
We put together a reference database of all players drafted in their draft year (excluding over-agers for now) from 2008 to 2017, with their respective Win Shares performance in the league they were in. Players with at least 10 games in two or more leagues that year are seen as multiple entities, giving one reference for each league they played in. We also excluded goalies completely for now.
Why only look at the draft year? We initially tried to include the Draft -1 year, with different weights, but the Draft season alone was always better.
Age adjustment
Age has a significant impact on evaluation and future success, as Byron Bader has highlighted in the past.
With less sample than Byron, we split one year of birthdates into four quartiles: September 16 to December 15, December 16 to March 15, March 16 to June 15, June 16 to September 15. Then looked at how many players played at least 100 NHL games.
We chose to put the entry barrier at 100 games, symbolizing established player with multiple seasons presence in the big league, or at least someone that really got a strong shot at establishing himself. Also to include as many seasons as possible, when putting the bar at 200 NHL games would have removed a lot of 2017 and 2016 players.
At first, we based ourselves on the percentage of players that became NHLers, but we realized later that we might have been led into the wrong.
Why so many Winter born players are drafted? If you look at Canadian or American census, there is no natural difference in births between months of the year. So, picking up over twice as many Winter born players compared to Fall born players (look at Byron’s numbers) is the result of a bias from NHL organizations. Why? There might be multiple reasons, from Winter born players always being older than Spring and Summer born players and ahead of them in their development, to the thinking that Fall born players are, to the contrary, already too old and having a supposed natural advantage (NHL organization applying a bias to correct a supposed bias…).
BUT, we believe that it led to an artificial increase of the volume of players drafted from Winter months, to the point where the 27% of draftees making the NHL is probably too low because the original numbers of drafted players was too high. By comparing that 27% to the average 32%, the disadvantage of Winter born players in our first version of the model was severe. And, to the opposite, the advantage of Fall born players (41% making the NHL) was very high.
We came up with a solution to remove the bias of drafting too many or too few players based on their birthdate by looking at the career performance in the NHL. By focusing on NHL careers, we removed the “noise” of all the players drafted but who didn’t make the big league. Looking at the Average Win Shares in the NHL, Fall born players are still looking good, while Summer born players move on top. Winter and Spring born players are still below but the difference between each quartile is much smaller that way, as the age adjustment will be.
In the end, every drafted player in the database was studied as the result of:
Win Shares per game in his league * age adjustment = Draft Year Win Shares
Linking Draft Year Win Shares and Future Success
Maybe the key idea behind this project is to put a prospect’s performance into context. When previous NHLe were “only” applying an aggregated multiplicator (based on all forwards from the OHL for example) to a performance, we wanted to dig deeper than that.
We looked at all drafted players in the database and compared them to each others for each position inside each league: a forward from the OHL was compared to all forwards from the OHL. Players were ranked per percentile according to their Draft Year Win Shares.
For the forwards from the OHL, Connor McDavid had the best Draft Year Win Shares with 0.17 WS per game. He is followed by Steven Stamkos (0.132) and John Tavares (0.128).
Then we looked at how many players reached 100 NHL games and what has been their career average NHL Win Shares. We grouped players by group of 10 percentiles (0 to 9, 10 to 19, 20 to 29, etc.).
What we saw was a clear link between being among the best of your class in your draft year, the probability of becoming a successful NHL player and how fast you can be in the big league.
In all leagues, the top 10 percentile seems to be on another level, on every metrics. Then it decreases till often the 50th percentile, where things can become a bit more blurry.
What it says to us is that a 2022 prospect whose Draft Year Win Shares puts him in the 80-89th percentile group among forwards from the OHL, would have a 50% chance of becoming an NHLer after about 3.2 years, for a career impact of 1.6 Win Shares per season. A top 10 percentile defenseman from the SHL would have a 100% chance of playing in the NHL right away…
The last step was to compute our own NHLe.
Win Shares NHLe
At first, we wanted to look at each group of 10 percentiles’ NHLe, which is calculated for every player as: NHL career average Win Shares / Draft Year Win Shares. For Connor McDavid, he moved from his 0.17 Draft Year Win Shares to a career average 0.082 in the NHL after this season, for a personal NHLe of 0.48.
But we quickly realized that our samples were too small for almost every league, causing too much noise, so we decided to group all players of one league together.
As expected, senior leagues have a much bigger NHLe, as 18 years old play limited minutes there though they might be the next NHL star. Just the fact they are playing in a senior league tells a lot about their potential. To that regard, the NCAA is almost like a senior league too.
The OHL is still the best junior league, just ahead of the Russian MHL. We will need to talk about Quebec at some point but the last 15 years have not been good.
DRAFTe
Here we go. Past our reference period of 2008-2017 (that will grow every year if we continue), any draft eligible prospect is assigned a potential metric, that we called DRAFTe. It is a fairly simple formula based on what we explained above.
Once all DRAFTe calculated, all prospects are now put on the same level and can be compared, and ranked.
Concluding thoughts
Is a Win Shares based projection better than a Points based projection? We did the same exercise of looking at pure points production, per historical percentile, with a new NHLe per league and the results are somewhat similar, sometimes better, sometimes worse.
First, points production are a big part of Win Shares, so both things are not completely different. Then, points are better to predict… NHL points. Being in the top 10 percentile of points producers on your Draft Year is a better predicator than being in the top 10 percentile for Win Shares. Is it a thing? Or is it because the NHL has a bias toward players that can score? There we touch a very long topic of projecting players from a somewhat neutral environnement (their draft season) before they enter a very subjective environment: their NHL franchise.
You might project the potential of a player but his path will depend on where he will play his development years, how good the development department of the NHL franchise is, what kind of players they prefer, is there room for the player in the organization at that time according to the compete level of the team, the salary cap situation, the number of veterans, etc.
But back to before the draft. It is interesting to see that Win Shares is a bit more linear in terms of projecting success. So we will stick with it for the near future and see what happens.
And yes, the 2022 projections are available online on our Tableau page. Click on the image below.
Liked that post?
And consider supporting us with a coffee !
You can also follow me on Twitter pages here and NL ice Data here