Ranking Various Pre-draft Lists and Analytical Models

[Tables of Stats]

here's bader's 2020 list. i haven't really looked at it yet but post your "best" 2020 list and let's compare

The Hockey Prospecting Top 32 (2020)

We introduce the final Hockey Prospecting rankings of the top 32 for the 2020 draft! The rankings factor in the probabilities from the model, but also age, runway left to improve, league-specific p…

hockeyprospecting.com

I intend to dig deeper and look at the expected value of a player taken at that draft ranking in the last 15 years, but even without doing that, it's very clear that Bader's model is dead last. 12 complete busts in 32 picks versus 10 for Button, 7 for McKenzie, 10 for THN, and 7 for the actual draft.

I choose to rank Mukhamadullin and Brisson as busts, as well as any player with less than 20 NHL games played. Khusnutdinov, Bourque, Lapierre, Reichel, Colangelo, and Berard could all be classed as busts, which brings the numbers to 17, 12, 9, 12, and 10, respectively.*

I'll do more analysis later, but I think I've been pretty fair in what gets counted as a bust with both sets of the above players.

*Amirov has to be labelled as a bust, but given his tragic passing, he's the biggest what-if of the draft.

https://docs.google.com/spreadsheets/d/1MND4sUIP7rWR7yirCjX-kyAS6FxCeycscffkn2KoaAI/edit?gid=0#gid=0

 

[Tables of Stats]

I'm currently compiling a list of 1st-round drafted players from 2000 to 2015 so we have a baseline to compare draft results. I intend to show how often that pick selects Forwards, Defensemen, and Goalies, and what the average expected production from each of those positions in the slot has been. I know this sort of thing has been done before, but I wanted to take a crack at it myself.

I have 2 seasons in so far and expect to get at least 5 done today. This project might be finished early next week, if I'm on my game.

 

[Tables of Stats]

The raw data from 2005 to 2011 is entered. I initially thought to do 2005 to 2020, but I realised I'd get far better data cutting things at 10 years post-draft rather than 5.

So far, I have a few ideas for how to slice things up for analysis. For players, those would be goals, points, points per game, peak season, games played, all compared to the average for the player's position and where they were selected. For drafts, that would be total points scored by 1st round picks, points scored by the entire draft cohort, games played for the 1st round picks, and games played for the draft cohort. I know this is going to undervalue defensive players and goalies, but I'm unsure how to work around that as nobody in this space has a good grasp on measuring defensive impact or goaltending. I'm open to suggestions for what people would like to see. This is as much for you as a fans as it is for my edification and entertainment.

I'm going to do my best to note players whose careers were cut short by circumstances unrelated to their skill as hockey players but this can get subjective and I'm very likely to miss players. Please take anything subjective with a pinch of salt.

Lastly, and no promises on this one, I might try my hand at making a model to help find the best draft prospects. It'll be a ton of work, and it will likely require me to get into coding again. I might burn out on the idea before I get anywhere with it, but support from all of you would make it more likely. For the anti-model guys here, don't worry, I'm not pretentious and won't pretend a model beats a comprehensive scouting system that includes watching live games, film, getting physical test data, and layering analytics on top of it. This is just for fun and to see if my interest is enough to start matching the guys who think they can sell their slop for money.

 

[Tables of Stats]

The stats for the first 32 picks from 2000 to 2015 have been entered, along with the numbers of players who played 1, 10, 50, and 200 NHL games. My next goal is to average that data among the 16 years covered, but to do that, I feel like I have to ignore goalies who don't work like any other position and are taken rarely enough that they may prove to defy any attempts to model them. The easiest way would be to add an extra skater for each goalie taken in the first 32 picks, so I'll start there, but if anybody has a better idea, let me know.

The drafts also seem to support the idea that you shouldn't use a first-round pick on a goalie. At a glance, they seem to risk busting more than position players, with a higher percentage of hits found in the later rounds compared to skaters.

https://docs.google.com/spreadsheets/d/1UTjeav-RqH1ZKMkApHPkp6nH79l_PNkzF1lC-XXNJJc/edit?gid=1553796047#gid=1553796047

 

[Tables of Stats]

See the 2000 tab for what I plan to have every year's tab look like. Also, note that I did decide to add an extra number of skaters equal to the number of goalies taken; my feeling is that this will give better results overall.

I'm also likely to do medians rather than raw averages, but that will come later.

 

[Tables of Stats]

If anybody has started reading this yet, I hope you like the little memes I've added to liven things up.

 

[Tables of Stats]

Everything is entered for 2000 to 2015, and some basic data for each draft's 1st round is in the right table. The middle table shows the number of players who played NHL games from the draft. I may list them by percentage, but given how close the numbers are between drafts (~55 players from each draft play 200 games, and ~32 make it to 500+ games), it might not be worth it. Take the stats from more recent drafts with a grain of salt; a draft class isn't locked in until the star players start retiring.

 

[Tables of Stats]

Looking at the 2000 to 2005 drafts, I've made an equivalency value for defencemen compared to forwards. It predictably favours two-way and offensive defencemen, but for now it's the best I've got. You apply this by multiplying the points a forward - or the average player taken at that pick, if the sample size is too small to isolate forwards - by this value.

2000 - 2005 Defence Offset GP - 0.97
2000 - 2005 Defence Offset Goals - 0.26
2000 - 2005 Defence Offset Points - 0.47
2000 - 2005 Defence Offset P/GP- 0.49

The sample size isn't large enough in 6 seasons to isolate F and D at each draft position. I suspect that it still won't be even with the full 16 draft sample size I'm working on, but I think you'd hit par if you were above average for 3 out of 4 markers (GP, G, P, and P/GP) after applying the defensive offset to that average. I'd also say that your par is set by the best remaining value below your current draft slot, so drafting in a "cursed" draft position doesn't absolve you from making a good pick.

 

[Vector]

I don’t have anything to add but I am following along.

 

[Tables of Stats]

I don’t have anything to add but I am following along.

That's more than enough. I'll even take suggestions for better memes if that's what people have to offer.

 

[Tables of Stats]

I worked smarter instead of harder, and now the averages for '00 - '05, '06 - '10, '11 - '15, and the full 16-season averages are ready to go. I was manually working around goalies. That was dumb. Making new pages, deleting the goalies and everything else I didn't need made the problem easy to automate with formulas I can drag and drop.

The next stage will be taking the draft-by-draft tables and colour coding them for steal, good value, average, below average, and bust. I'm thinking my threshold will be 20%+ above average, 19 to 5% above average, 4% above to 4% below average, 5 to 19% below average, and 20%+ below average. Goalies will likely be done as their own pool, all averaged together as there aren't enough taken to do anything with them otherwise.

I'll also go through each year and add all the skaters and goalies who would have been worth using a first-round pick on who went outside of the top ~32 picks. This should give us a trend of how many quality players were missed each draft so we can see how much, if any, drafting has improved over the years.

 

[Tables of Stats]

For those who've looked at this so far, any comments? Could be small stuff like layouts or a typo or bigger picture stuff.

 

[Tables of Stats]

https://docs.google.com/spreadsheets/d/1gnKl7otCLk8wZan3jkfGiSQWcV-8jFDp0lmu9uto02A/edit?gid=0#gid=0

Going back to the 2020 draft, I've graded each player:

1.2+ P/GP = Green = 6 Points
1.19 - 1.05 P/GP = Blue = 3 Points
1.04 - 0.95 P/GP = Purple = 1 Point
0.94 - 0.81 P/GP = Orange = -1 Point
-0.8 P/GP = Red = -3 Points

Defencemen have their par lowered to 0.558 of the above values, and players with under 41 games played are automatically busts. These ratings will change as the draft develops; some players are on the cusp of being several ratings higher. This heavily biases towards offence for value; later iterations will look at games played, and raw goal and assist totals relative to other drafts at the same number of seasons post-draft.

Draft pick scoring values are multiplied by 1.5 for the first overall slot, dropping to 1 at pick 16 before falling to 0.47 at pick 32 to represent the relative impact of those picks.

Hockey Prospecting: -3.032
Craig Button - 32.349
Bob McKenzie - 48.999
The Hockey News - 35.006
NHL Draft - 44.959

Bob McKenzie is 1st, the 2020 NHL Draft is 2nd, The Hockey News is 3rd, Craig Button is 4th, and Hockey Prospecting (the only negative value draft) is DFL.

 

[Tables of Stats]

@ManVanFan @credulous I'd be curious to see your thoughts on this project, seeing as you sparked me to do it. It's a fun diversion.