Happy draft day, everyone. You can thank Eric for this one. Even if the answer to his first question from the tweet is “all of them to some extent but not as many as much as they should” and the answer to his second question is “I promise this isn’t my area of expertise but this post should make it easier to look through what’s out there”.

Draft analytics is no ordinary moving target in all sports at this point, especially in the NHL: Figuring out who will be good in the NHL during the draft process is a target moving 100 MPH in one direction while you are moving as fast in the other direction while gusts of wind are pushing back against you and you have to hit the target with a whiffle ball. Teams have to evaluate these prospects at 17 going on 18 with what they have done in their young hockey careers from numerous leagues around the world and get a gauge of what that means for where they should start their post-draft career and from there, what kind of player they will be over the next 10+ years of their hockey-playing lives. The past, present, and future all have to be considered at once for every single player and then compared to their peers in the draft class. 

So how do they do that? Well, we don’t know because teams will never let other teams know their secrets. How do we do that? Let’s take a look at what kind of draft research has been published and is currently out there. And while a lot of NHL teams are likely still committed to traditional scouting techniques for determining their draft rankings, the analytical work is likely the same. The difference, hopefully, is that the teams have better access to data from different junior and international leagues to, again hopefully, assist them.

NHL Point Equivalencies

Point equivalencies were the first place for prospect evaluation because even now, for some leagues, all we have publicly about them are scoring data – Goals, assists, and points. Plus, we know at this point that at minimum, we will have this information for each league that draft prospects will be coming from. So how do we estimate how valuable goal and point totals are from one league to another? Research on comparing production points in different leagues goes way back on this one from Gabriel Desjardins back in the early-2000s. 

To get an understanding of how players translate their draft year performance to their post-draft season, and then to the NHL if that isn’t a direct transition, the point per game totals in those two seasons can be compared. So for example, to get an idea of how a player performs playing in the NCAA one season and then the NHL the next season, divide the PPG in the NCAA season by the PPG in the NHL season. Then take all the players who go from the NCAA to the NHL and average all of those together to get the NHLe value for that college hockey.

From here, comparisons can be made to differentiate how they jump from one league to another can happen based on the ages the player is when they play those two seasons or how it changes from year to year, etc. Then of course since defensemen will usually have different point levels than forwards, the calculation can also be drilled down further to account for that and other factorized variables that can come out of player evaluations. Some analysts will even take the common second league being the NHL and use that to compare a player’s performance in one league to a similar player’s performance in a different league to get a sense of how they would have likely performed in that other league. Proportions can be fun; never forget fractions.

Value by Draft Position

For estimating the value of a draft position, it’s a good thing we quickly learned about regressions earlier this week. The difference is that instead of a traditional linear regression, this will be one of exponential decay (an exponential function that decreases as x increases) to smooth the values.

Now here’s where it’s going to get tricky and, honestly, a tad underwhelming. While the function will always have an input of draft position, whether pick-by-pick or a clustered clump of adjacent picks and output of either some projected level of career performance on length, the thresholds for the output are usually going to change from analyst to analyst, as well as year to year while we acquire a whole new draft class annually.

The easiest place to start is probabilities that the player drafted nth overall will surpass a threshold of games played; let’s say 82 games for the hypothetical. To recreate this, take each player selected at each position in the draft. The frequency of playing 82 games is calculated by taking the number of players at that draft position who played at least 82 games divided by the number of players that have been drafted at that position. The frequency will be the estimated probability of a player at that spot playing 82 games. Then, with the rest of the draft positions calculated, the smoother, or line of best fit, is calculated to give the smoothed estimated probability for that draft position. The smoother is used to account for the fluctuation between consecutive picks. The current results that we have seen at each draft position have high variance just based on the limit in not having an infinite number of NHL drafts to sample from, and so this exercise is done to theoretically find the empirical value of the draft position. 

Here is how it looks visually from a Sportsnet article during the pre-draft process in 2015. This one is good because it shows both the fluctuation from single pick to single pick based on the specific players drafted at each position while also showing three different thresholds of games played. The biggest and least surprising takeaway here based on familiarity with the NHL is how it only takes reaching around the 30th, 20th, and 15th overall pick before there being less than a theoretical 50% chance that a player selected in that draft spot plays 60, 100, and 200 games in the NHL, respectively.

This article also shows a plot of expected points per game coming from forwards drafted in each slot which much quicker dropoff where after the first round, it’s theoretically expected that the forward will average no better than a point every five games by the end of the first round, which to me is best interpreted as the struggle to find consistent NHL forwards after round 1 as opposed to the raw average itself.

“Other”

Here are some other interesting posts about drafting from the last handful of years. 

NHL Draft Pick Value Chart form Don’t Tell Me about Heart: This is another draft value chart, but this post does a good job of showing how the draft value for skaters is a lot more predictable than that of drafting goalies. This also does a good job of showing how draft values work. For an overview, they will start with a baseline of the first overall pick and through the wonderful world of ratios, compare the rest of the selections to that of the first pick. 

NHL Draft Probability Tool from (again) Don’t Tell me about Heart: This one works backward a little bit and instead looks at the players in an upcoming draft (in this case 2016 – time flies) and the probability that they are available at a given pick. This is done with draft rankings that are publicized by media members and other researchers and help drive the point across that the best baseline for draft rankings is the consensus. Another example of having more data usually being a better gauge of expectation.

Draft Theory from Peter Tanner (Moneypuck) via Canucks Army: More explanations of draft value and some of the logic steps of how to evaluate prospects with data-based concepts

What Does It Mean To Draft Perfectly? An Evaluation Of Draft Strategy In The National Hockey League by Namita Nandakumar: A lot of Namita’s hockey work has been scrapped from the internet after she joined the Kraken’s analytics department, but this one was her undergrad project at Penn, so it lives on in the university’s repository. It uses point shares to evaluate the careers of the 2007 draft as a test case and evaluate how each team drafted compared to how they would have with perfect information of how players’ careers would have turned out. It’s lengthy but not too daunting to comprehend statistically (And shows that the Sabres had all but three years in the first 10 years of the 200s where they were below the top 12 in efficiency – The best in the NHL based on those calculations).

Exploiting Variance in the NHL Draft from (again) Namita: This one focuses more on the variability from one pick to another and also goes through the logic that goes into the importance of hitting on draft picks in a salary cap league where talented players on entry-level deals allow for more cap space to attract more talented veterans and build an optimized talented team.

So hopefully everyone enjoys the draft or at least enjoys their weekend. And hopefully, there are plenty of things to talk about next week with lots of new faces on the Sabres roster.

Photo Credit: Bill Wippert/NHLI via Getty Images