Why should my draft model be taken seriously? The last 17 seasons of draft retrodictions explains why.

Here it is, like I promised to some – about 3 days later than I wanted/expected. I will explain parts of the spreadsheet later here, and go on a (hopefully) fun romp down memory lane with my new found knowledge in hand:

 1998-2014NBAdraftRetrodictionResults

But to get there, I compiled all the college ratings (two years weighted when applicable) for the 1031 D1 players who have been drafted &/or played at least 1 minute in the NBA since the 1999 NBA draft. So, compiled all the stats (making needed corrections), calculated team ratings on 1244 past teams (most players & had to calculate their last 2 seasons), and calculated all the player ratings for every guy on those teams. Oh, yeah – had to make sure I had the exact birth date for every guy in my database – since I’ve changed in my age progressions from rounded to exact to improve my results (even if ever so slightly) for my peeps.

This was work I needed to get done – since I know there are those who want to see my work in action – whether it has real value in finding the sleepers and those that are just overrated. I wanted to do all 17 seasons since I had complete college player stats back to 1997 – and I want to get to my next project which involves similarity scores. I needed to calculate the projections of those 1031 guys to make it happen.

My similarity scores (when done) will not be your conventional ones. I won’t be comparing basic college stats between players with same rounded ages. I will be projecting every player based on their college ratings (which adjust for pace, SoS, team quality, etc) to the exact same age (I’m thinking 26 – I’ll have to project backwards for Bernard James since he was older than that as a rookie). From there – I can run the scores with EVERY player in my database based on the ratings broken down into every statistical subset. I am about certain no one has ever run similarity scores nearly as complex as these will be (adjusting for everything, including age) – at least not publicly. I think it’ll be pretty awesome when I get it fully flushed out (at the very least in terms of fun information, if not possibly practical enough to be incorporated into my next draft model) – hopefully before the draft next week.

Now, to the retrodiction – there’s something I need to say up front….

Don’t get too caught up in the misses – look at the whole picture:

There is absolutely no way ANY projection system will get anywhere close to everything right. My model has misses – the most blatant one (to get this out of the way right up front) is that Michael Beasley (drafted #2 behind Derrick Rose in the 2008 draft) projected better than any other player over the last 17 drafts. He was so utterly dominant as a very young man in college – if any draft model didn’t have him projected at or very damn close to the very top – they are lying. I won’t lie – I try to learn from my misses.

In a similar realm – Mike Sweetney (drafted 7th in the 2003 draft from Georgetown) had the 7th highest projected career WAR of the 1031 player list. He was a killer in college – not so much in the NBA.

OK – got that off my chest – the other 8 guys in my overall top 10 in projected career WAR all became NBA stars: Irving, Love, Anthony Davis, Durant, Blake Griffin, Bogut (OK, maybe not “star”, but very good/vital to a great team), Brand, and Steph Curry. So, at least there’s that.

But, what we are looking for here is the entire scope of the rankings. We want the guys that rank very high by the draft model but slip to the 2nd round or later to be the best sleepers – & they generally are. The top 9 career WAR of 2nd round or undrafted, 6 of them were tabbed top 10 prospects (lottery quality) by the model – Boozer, Millsap, Arenas, Brad Miller, Michael Redd, & Korver.

We also want to know what players to avoid. We want to avoid that horrible “bust” if we have the top pick. #1 pick Olowokandi – the model hated him, ranked 25th in his draft – the model liked Mike Bibby infinitely better (heck, it even liked undrafted Brad Miller much, much more). #1 pick Anthony Bennett, ranked 7th. #1 Oden – while ranked #2 by the model in ’07, Durant’s projected career WAR dwarfed Oden’s. #2 Thabeet, ranked 11th by the model in his relatively weak class (best player available, some guy with a crazy huge projection named Steph Curry). #2 Evan Turner – #5 by the model (Cousins, Favors, Hayward, & Monroe all projected a bit better). #3 Adam Morrison – #10 by the model.

But, generally, we are looking at the guys that both the model and NBA scouts love. This is usually a very strong recipe for success – with only small handfuls of misses.

Here are the overall facts about this draft model:

Without any of the millions invested in scouting, combine results, draft mocks, high school rankings, individual workouts, interviews, college coaches’ input, etc. – this draft model in a general sense very strongly outperforms actual NBA team drafts. Any players drafted high by a team (but not rated highly by the draft model) SHOULD on average greatly out perform any/all late or undrafted players that the model really likes. There is an inherent bias – the team has paid guaranteed $$ over a number of seasons for the player – they will do whatever they can to help him become successful. But, more often than not, the opposite is true, the later drafted players the model likes often have better careers – without the early guaranteed contracts and team coddling/focus.

To best show this, I split EVERY player of the 1031 into 3 categories:

1. Rated higher by Hoopsnerd model than the actual draft position (“HoopsNerd Favored” in the “FactorWork” sheet of the spreadsheet).

2. Rated higher by NBA teams than the model (draft position higher than draft model rank – the “NBA GMs Favored” group

3. The ties – when the draft position & draft model rankings exactly agree, we put them in the “Rankings tied” group & ignore them.

Now, I created a weight column called “Factor” – for the HoopsNerd model section it is actual Draft rank (draft position  of the D1 players drafted) divided by draft model rank (vice versa for the NBA GMs section). So, Boozer was the 26th D1 player drafted in his draft – but the HoopsNerd draft model had him ranked #1. 26/1=26 – his factor is 26, his career WAR will be given greater weight by a factor of 26 in the calculations. Brad Miller’s factor is 25.5, Paul Millsap 19.0. It’s not all complete sleeper finds – Todd MacCulloch, Mike Harris, Ryan Bowen, & Jae Crowder have factors over 10 also.

Unfortunately for the GMs, their highest factor is Olowokandi at 25.0. They do have Aldridge at 15.0, so that helps them.

So, I take the actual career WAR of every guy in each group (as well as draft rank and model rank) – multiply it by their respective factor – and compile the totals and average the results.

I did this method because it gives more weight to each side’s outliers, while less weight to the players which barely ended up in one side or another (say, 5th in draft model, 6th in draft position). There’s more subjective wiggle room there for the middling guys – the outliers are where we want to focus more.

What we get (which can be seen at the bottom of the “FactorWork” sheet):

The average career WAR (weighted by factor) of the guys drafted higher by the GMs than the draft model rank was 13.3.

The average career WAR (weighted by factor) of the guys the draft model liked more than the NBA GMs (as evidenced by draft position) was 15.1.

Now, that doesn’t seem like that much – but with a database this size that does amount to about 1000 more wins for teams the draft model liked more than the NBA teams.

But – there is one HUGE factor that needs to be mentioned – the better average career WAR of the HoopsNerd model favored guys (15.1 to 13.3 each guy) was achieved with an AVERAGE draft position of over 17 picks LATER.  The average draft position (relative to all D1 players, ignoring high school & internationals) for the HoopsNerd guys who compiled that average career 15.1 NBA WAR was 35.6 (nearing mid 2nd round). The average draft position for the NBA GM guys who compiled that average career 13.3 NBA WAR was 18.2 (mid to later 1st round).

Of course, the players the model liked more than NBA GMs often had to work their way onto teams compared to their counterparts – they often didn’t have the luxury of a guaranteed roster spot. Yet, on average, they STILL had the better careers, despite each having been picked over a half a round later.

How valuable would it be to any franchise if they routinely drafted BETTER future NBA players much later in the draft than other teams (as well as have a much smaller chance of busts on the higher picks)? How much would that extended approach help fill deeper roster spots with players possibly more capable of producing when given a chance – D League draft, summer league free agent rookie signings, etc?

OK, more to come – I want to look at each draft for fun – but for now I have to run…… I’ll get to that soon.

I’m back, let’s have a little fun

Well, to me it’d be fun. It’ll probably be fun for readers delusional like myself.

I decided to go draft to draft – me & my model versus the NBA teams. I will be armed with a vague knowledge of how teams value prospects – just like any team would be going into any draft.  That allows me to “draft” certain players my model loves who are projected by mocks to be 2nd round or later picks later in my fantasy draft – no one in their right mind in 1998 would draft Brad Miller #2 overall in the 1998 draft when it was common knowledge he would be available late 2nd round.

So, in essence, it’ll be like a blind auction. Whichever side (GMs or HNboard) had a guy ranked higher gets the guy. Same rank (“Draws”) go to neither. Each side will get the same number of picks – I’ll have an “Undrafted” section which will equal the number of undrafted guys in the data set for that year – there’s no sense in either side getting a guy neither liked.

It’ll make sense as we go…..

1998 Draft

NBA GM Guys CareerWAR DraftRk HNrank HNboard   HoopsNerd Guys CareerWAR DraftRk HNrank HNboard
Antawn Jamison 96.4 4 11 9 Paul Pierce 194.3 9 8 6
Cuttino Mobley 50.5 33 50 50 Vince Carter 164.1 5 5 3
Jason Williams 46.4 7 13 11 Mike Bibby 81.7 2 1 1
Rafer Alston 35.2 32 47 47 Brad Miller 69.4 51 2 18
Raef LaFrentz 34.6 3 6 4 Bonzi Wells 38.4 10 3 2
Ruben Patterson 34.1 26 37 37 Nazr Mohammed 27.8 24 9 7
Ricky Davis 32.5 18 28 28 Earl Boykins 24.0 53 32 32
Keon Clark 16.3 12 20 20 Anthony Carter 11.2 54 40 40
Tyronn Lue 12.5 20 30 30 Greg Buckner 9.2 45 19 19
Robert Traylor 11.6 6 14 12 Brian Skinner 8.4 19 18 16
Michael Dickerson 9.3 13 52 52 Michael Doleac 6.2 11 12 10
Pat Garrity 9.2 16 21 21 Jelani McCoy 3.2 27 15 13
Michael Olowokandi 8.0 1 25 25 Ryan Bowen 2.3 47 4 17
Jahidi White 7.7 35 36 36 Ansu Sesay 0.6 25 24 24
Shammond Williams 6.3 28 35 35 Derrick Dial 0.5 44 29 29
Jerome James 1.9 29 63 63 Charles Jones 0.1 58 22 23
Felipe Lopez 1.7 21 59 59 Sean Colson 0.1 65 39 39
Bryce Drew 1.1 15 27 27 Corey Benjamin 0.0 23 7 5
Sean Marks 0.8 36 65 65 DeMarco Johnson 0.0 31 17 15
Andrae Patterson 0.8 38 64 64 Billy Thomas 0.0 59 31 31
Roshown McLeod 0.4 17 23 22 Miles Simon 0.0 34 33 33
Sam Jacobson 0.1 22 58 58 Steve Goodrich 0.0 63 34 34
Toby Bailey 0.0 37 42 42 J.R. Henderson 0.0 48 38 38
Casey Shaw 0.0 30 51 51 Andrew Betts 0 42 26 26
417.2 20.8 641.6 35.4
17.4           26.7  
Draws CareerWAR DraftRk HNrank HNboard   Undrafted CareerWAR DraftRk HNrank HNboard
Larry Hughes 43.4 8 10 8 Mike James 20.4 52 57 57
Matt Harpring 35.9 14 16 14 Tyrone Nesby 5.9 55 46 46
Sarunas Jasikevicius 3.2 56 43 43
Zendon Hamilton 1.7 57 60 60
Maceo Baston 1.1 50 66 66
Tremaine Fowlkes 0.3 46 48 48
Tyson Wheeler 0.0 39 41 41
Kelly McCarty 0.0 67 55 55
Torraye Braggs 0.0 49 45 45
Cory Carr 0.0 41 44 44
Ryan Stack 0.0 40 61 61
Jeffrey Sheppard 0.0 62 49 49
Gerald Brown 0.0 61 56 56
Mark Jones 0.0 64 54 54
Randell Jackson 0.0 60 62 62
Makhtar N’Diaye 0.0 66 67 67
Corey Brewer 0 43 53 53

I moved two guys down on my draft board (Miller & Bowen) into the latish first round that I would have known then were projected at best late 2nd round picks. There were two draws (Hughes & Harpring) in which my board & gms had ranked the same – I set them aside. There were actually 50 college guys drafted in 1998, so subtracting the two draws we’ll have a cap of 24 players to each side, the rest were “undrafted”.

I won’t say much on these – I have 16 more drafts to go. The HoopsNerd group totaled 224 more career WAR.  The average player in that group had over 9 more career WAR than the average GM group player – while being drafted over 14 college players later (35.4 average draft rank to 20.8).

1999 Draft

NBA GM Guys CareerWAR DraftRk HNrank HNboard   HoopsNerd Guys CareerWAR DraftRk HNrank HNboard
Jason Terry 107.2 9 18 18 Shawn Marion 145.2 8 5 4
Baron Davis 95.5 3 7 6 Lamar Odom 86.4 4 4 3
Steve Francis 71.3 2 13 13 Richard Hamilton 71.1 6 6 5
James Posey 33.1 15 16 16 Metta World Peace 70.2 13 12 11
Jeff Foster 30.4 18 21 21 Corey Maggette 53.1 11 11 10
Lee Nailon 6.7 35 54 54 Wally Szczerbiak 40.3 5 2 2
Dion Glover 5.1 17 32 32 Kenny Thomas 27.9 19 14 14
Francisco Elson 3.6 33 59 59 Todd MacCulloch 9.4 39 3 12
Calvin Booth 3.3 29 52 52 Jumaine Jones 9.1 22 15 15
Vonteego Cummings 1.7 21 44 44 Milt Palacio 2.2 50 34 34
Trajan Langdon 1.1 10 33 33 Evan Eschmeyer 0.7 28 10 9
Michael Ruffin 0.5 26 30 30 Rodney Buford 0.6 44 20 20
Cal Bowdler 0.5 14 42 42 Mike Batiste 0.4 52 35 35
Laron Profit 0.3 31 36 36 Jermaine Jackson 0.3 51 27 27
Obinna Ekezie 0.2 30 55 55 Harold Jamison 0.2 53 22 24
Quincy Lewis 0.0 16 19 19 William Avery 0.0 12 9 8
Tim James 0.0 20 25 25 Derek Hood 0.0 58 28 28
John Celestand 0.0 24 38 38 Tim Young 0.0 47 31 31
Chris Herren 0.0 27 60 60 Rico Hill 0 25 17 17
A.J. Bramlett 0.0 32 47 47 Venson Hamilton 0 42 23 22
Tyrone Washington 0 36 49 49 Louis Bullock 0 34 26 26
J.R. Koch 0 38 53 53 Roberto Bergersen 0 43 29 29
360.5 22.1 517.1 30.3
16.4 23.5
Draws CareerWAR DraftRk HNrank HNboard   Undrafted CareerWAR DraftRk HNrank HNboard
Elton Brand 129.9 1 1 1 Raja Bell 25.4 49 40 40
Andre Miller 115.2 7 8 7 Jamel Thomas 0.0 55 58 58
Scott Padgett 8.0 23 24 23 Geno Carlisle 0.0 57 41 41
Ryan Robertson 0.0 37 37 37 Lari Ketner 0.0 41 50 50
Wayne Turner 0.0 56 43 43
Maurice Carter 0.0 54 45 45
Guy Rucker 0.0 59 46 46
Jason Miskiri 0.0 60 51 51
Kris Clack 0 46 39 39
Melvin Levett 0 45 48 48
Galen Young 0 40 57 57
Eddie Lucas 0 48 56 56

 

Moved Todd MacCulloch back some on my board to mid 1st area – since it would have been common knowledge he was going somewhere mid 2nd round. Moved lightly regarded (eventually undrafted in ’99) Harold Jamison back a couple spots to help get a better chance at landing Venson Hamilton – who never played in the NBA. Wouldn’t have known that at the time, sure didn’t help me here. Also just missed having Raja Bell and his 25.4 career WAR on my side, but he was ranked just a little too low on my board and went undrafted.

Still, the HoopsNerd guys had over 156 more total career WAR than the GM guys. That comes out to an average of 23.5 career WAR per player to the GMs 16.4 career WAR player average. That was done despite the HoopsNerd guys being drafted over 8 college players later (30.3 draft rank to 22.1) on average than the NBA GM group.

2000 Draft

NBA GM Guys CareerWAR DraftRk HNrank HNboard   HoopsNerd Guys CareerWAR DraftRk HNrank HNboard
Jamal Crawford 60.8 7 27 27 Michael Redd 61.6 34 10 10
Mike Miller 48.8 4 5 5 Quentin Richardson 26.7 16 4 4
Jamaal Magloire 23.3 17 43 43 Morris Peterson 23.2 19 8 8
Desmond Mason 21.4 15 42 42 Eddie House 18.1 29 22 22
Joel Przybilla 13.9 8 16 16 Speedy Claxton 15.1 18 3 3
Etan Thomas 10.5 11 17 17 Eduardo Najera 11.4 30 28 28
Keyon Dooling 9.8 9 23 23 Brian Cardinal 7.8 35 15 15
Chris Mihm 9.3 6 7 7 Jason Hart 4.8 38 31 31
Marcus Fizer 5.4 3 6 6 Donnell Harvey 2.6 20 9 9
DerMarr Johnson 4.1 5 11 11 Mark Madsen 0.5 23 18 18
Jake Voskuhl 3.2 26 51 51 Khalid El-Amin 0.4 27 12 12
Courtney Alexander 2.5 12 46 46 Eddie Gill 0.3 47 20 20
Jerome Moiso 1.8 10 40 40 Jabari Smith 0.3 36 30 30
Jason Collier 0.8 14 45 45 Pepe Sanchez 0.3 50 19 19
Mamadou N’Diaye 0.8 21 53 53 Paul McPherson 0.3 49 34 34
A.J. Guyton 0.2 25 35 35 Dan McClintock 0.0 40 13 13
Mateen Cleaves 0.0 13 25 25 Chris Porter 0.0 41 14 14
Dan Langhi 0.0 24 26 26 Pete Mickeal 0 44 21 21
Erick Barkley 0.0 22 33 33 Scoonie Penn 0 43 24 24
Lavor Postell 0.0 31 39 39 Jaquay Walls 0 42 29 29
Mike Smith 0.0 28 50 50 Mark Karcher 0 37 32 32
216.5 14.8 173.5 34.2
10.3 8.3
Draws CareerWAR DraftRk HNrank HNboard   Undrafted CareerWAR DraftRk HNrank HNboard
Kenyon Martin 53.9 1 1 1 Ime Udoka 4.0 46 48 48
Stromile Swift 18.2 2 2 2 Malik Allen 2.6 45 37 37
Richie Frahm 1.1 48 41 41
Kaniel Dickens 0.0 39 44 44
Hanno Mottola 0.0 32 52 52
Alex Scales 0.0 53 38 38
Desmond Ferguson 0.0 51 47 47
Terrance Roberson 0.0 52 49 49
Chris Carrawell 0 33 36 36

Well, it was probably going to happen. While the draft model didn’t “win” this draft – much of that is because of how it played out – after the two top player “draws” were taken out, the next 13 drafted players ended up on the GM side. In a draft with very little talent  in general (only 7 players projected above 30 career WAR, which is horrible – boy did that play out in reality), only ending up with 3 players who were drafted in the first round was too much to overcome.

With so little depth of talent – I didn’t even have the flexibility of changing my board because of moving some super late “sleeper” (late 2nd round or undrafted) my model liked way more than the GMs.

Although, then there’s Mike Miller. One of the few players the model liked, if he had just projected 6.5 more career WAR (from 63.6 to 70.1), he would have leapfrogged Speedy Claxton & Quintin Richardson on my board – and he & his 48.8 career WAR (in a WAR depleted draft) would have switched sides – giving HoopsNerd an easy win.

But it didn’t happen. It was a perfect storm of circumstance going against the model – the NBA GMs ended up with 43 more career WAR, averaging 2 more career WAR per player. But, the GMs landed all the top draft picks in a draft with little overall talent – the GMs on average picked almost 20 spots earlier (14.8 to 34.2) per player to get that extra 2 WAR per player. In terms of cost-benefit analysis – that isn’t good, having to draft almost 20 spots earlier to get a slightly better player.

2001 NBA Draft coming soon…

Dan

2 thoughts on “Why should my draft model be taken seriously? The last 17 seasons of draft retrodictions explains why.

Leave a Reply

Your email address will not be published. Required fields are marked *