Draft Projection- The Projection Project

Recently I saw The Projection Project’s post on their draft projection data , and I wanted to see the differences between their projection and the Central Scouting Service’s. There’s also some additional information about the TPP model that sheds some light on the logic behind draft projections.

There’s a more in depth explainer on the TPP site , but in short, this model uses data on past prospects to compile cohorts of players which are grouped using height and  NHLE points (a model that translates points from junior/college/european leagues into NHL points).

The first dashboard shows how the TPP and the Central Scouting models agree on which prospects should be highly rated, but then they diverge after the fairly obvious characters. TPP vs. CSS What’s interesting is that the two models rank centers fairly similar, while they diverge significantly on wingers and defensemen.

The second chart focuses on which prospects are favored by which model.

TPP vs. CSS 2 The function here is simply (TPP Model Rank # – CSS Model Rank).  Players with negative numbers (like Mitchell Stephens) are more favored by the TPP model, while players with positive numbers (Kevin Stenlund) are favored by the CSS model.

This chart just shows the height and weight of each player on a scatter plot. I don’t think it tells us anything particularly, but it’s fun to look at.

HeightWeight 1

This chart shows the average height and weight for each position. The height doesn’t vary across the positions, but we can see that defensemen are clearly the heaviest position.  There’s probably some prior selection bias happening here, going back to which player was bigger in a youth league.

HeightWeight 2

This is my favorite chart of the bunch.  It’s clear that as the size of a player’s cohort increases, the probability of that player becoming an NHL player decreases. Cohort Size

This may be counter intuitive, since usually the larger sample size is better. However, players like McDavid and Eichel are rare (generational?) talents, so there are very few comparable players in their cohorts.  The players with large cohorts are likely to be run-of-the-mill talents with nothing to separate them from their peers.

MoneyPuck, who’s PCS draft model I hope to address soon, sees the same phenomenon in that model.

The last chart shows what types of past prospects a player’s cohort is composed of.

Cohort Composition You can see that the players with the largest probability of turning into an NHL player are on the bottom right of the chart (the size of the bubble is the indicator there). That is because their cohorts are made up of more players that turned into NHL players.


I had some fun looking through the TPP data, and I hope you learned something about how projecting NHL talent can be done. Thanks for reading.

2014-15 Playoffs- Running Totals

I recently read the @JapersRink season recap for Alex Ovechkin. The quote that stood out to me:

Interesting Stat : Despite not having played a game in nearly a month, Ovechkin still leads all skaters in shot attempts this playoff season…

While this is not technically true anymore, I wanted to see how that looked, and if there are any other interesting things to be found in the playoff data. There are a couple preset measures you can choose from, including Corsi, TOI, and Goals.  I’ll probably add some more as they come to me.

Data from War-On-Ice