APBRmetrics

Rank Player CPR
1 D'Angelo Russell 10.4
2 Jahlil Okafor 9.1
3 Karl Towns, 7.6
4 Kevon Looney 7.1
5 Bobby Portis, 6.5
6 Myles Turner, 6.4
7 Cameron Payne 6.2
8 Stanley Johnson 5.6
9 Tyus Jones 5.5
10 Justise Winslow 5.2
11 R.J. Hunter 5.1
12 Christian Wood 5.0
13 Terry Rozier 4.7
14 Jordan Mickey 4.3
15 Jarell Martin 3.9
16 Frank Kaminsky 3.4
17 Joseph Young 3.3
18 Kelly Oubre 3.3
19 Rashad Vaughn 3.2
20 Tyler Harvey 3.1
21 Wesley Saunders 3.0
22 Corey Hawkins 2.9
23 Ryan Boatright 2.8
24 Delon Wright 2.7
25 Jerian Grant 2.5
26 Olivier Hanlan 2.4
27 Chasson Randle 2.4
28 Aaron White 2.3
29 Josh Richardson 2.3
30 D.J. Newbill 2.3
31 Devin Booker 2.2
32 Richaun Holmes 2.2
33 Alan Williams 2.2
34 Derrick Marks 2.2
35 Keifer Sykes 2.1
36 Travis Trice 2.1
37 Treveon Graham 2.0
38 Terran Petteway 2.0
39 Anthony Brown 2.0
40 T.J. McConnell 2.0
41 Vince Hunter 2.0
42 Pat Connaughton 1.9
43 Larry Nance 1.9
44 Montrezl Harrell 1.9
45 Michael Qualls 1.9
46 Darrun Hilliard 1.8
47 Justin Anderson 1.8
48 Jonathan Holmes 1.8
49 Rondae Hollis-Jefferson 1.7
50 Dez Wells 1.7
51 Marcus Thornton 1.7
52 Rakeem Christmas 1.6
53 Norman Powell 1.6
54 Aaron Harrison 1.6
55 Quinn Cook 1.6
56 Robert Upshaw 1.5
57 Trey Lyles 1.5
58 Andrew Harrison 1.4
59 Shannon Scott 1.4
60 Sam Dekker 1.3
61 Michael Frazier 1.2
62 J.P. Tokoto 1.2
63 Willie Cauley-Stein 1.2
64 Chris McCullough 1.1
65 Julian Washburn 1.0
66 Juwan Staten 0.9
67 Branden Dawson 0.9
68 Lebryan Nash 0.8
69 Cliff Alexander 0.8
70 Dakari Johnson 0.8
71 Brandon Ashley 0.8
72 TaShawn Thomas 0.7
73 Joshua Smith 0.6
74 Chris Walker 0.2

We are just waiting for Steve's ranking of Seth Tuttle and Jesse's rankings.

First time I've heard of Wesley Saunders. Huge positive differential DX-consensus. Dekker and others vie for most "overrated" by DX or I guess underrated by stat models.

Seth Tuttle has a CPR of 1.7. He slides in 52nd (just ahead of Christmas) on my rankings.

Steve

CPR gets around this by focusing only on each individual’s top 10 performances in each statistic.

if a player plays well against non-conference teams, or mid-level teams, and poorly against top tiered teams, does this not affect their rating?...

CPR has been effective projecting both high picks that busted and late picks that surprised.

i see it lists willie cauley-stein quite low, and also sam dekker. most mock drafts have cauley-stein listed as a high pick, and jordan mickey as a 2nd round pick. yet when you compare their career college stats mickey's are in a few ways better (shot blocking/fewer fouls/rebounding), despite being only 6-8 in height compared to the 7-0 cauley-stein's...

i'm curious if you can show some of the recent drafts CPR listings - i'd be interested in seeing these "...high picks that busted and late picks that surprised..." and any CPR projections that missed...

can you show this without divulging anything proprietary? or might previous years be listed somewhere on your website?...

If there is enough room, it would be useful to see the average absolute differences from DX and the models. Maybe number of highest and lowest rankings of players by each model too. And / or the overall correlations.

If you want to continue to study the rankings, maybe look at avg. rank when 1-2 biggest outliers are thrown out. So far 15 guys have a top 10 ranking but only ten have 2 or more. The worst rankings for a player also receiving a top 10 ranking are for Booker, Cauley-Stein and Maccollogh (projected outside the draft entirely) The same for worst average ranking for such players. 7 of the 15 with a top 10 ranking also have a ranking of 20 or below. Only two outside the consensus top 40 have a top 20 ranking. Onky two outside the consensus top 20 have two top 20 rankings.

A 50/50 blend of DX and Consensus of Models puts KAT on top by a little. In just over 1/3rd of the cases the DX and Consensus rankings are within 5 spots. In 40% of cases the difference is 10 or more. So in about 1/4th of cases are between 5-10 spots different. Only a handful of guys ranked in first round by DX are projected as consensus of models second rounders, the worst discrepancy being for Booker.

bchaikin wrote:CPR gets around this by focusing only on each individual’s top 10 performances in each statistic.

if a player plays well against non-conference teams, or mid-level teams, and poorly against top tiered teams, does this not affect their rating?...

CPR has been effective projecting both high picks that busted and late picks that surprised.

i see it lists willie cauley-stein quite low, and also sam dekker. most mock drafts have cauley-stein listed as a high pick, and jordan mickey as a 2nd round pick. yet when you compare their career college stats mickey's are in a few ways better (shot blocking/fewer fouls/rebounding), despite being only 6-8 in height compared to the 7-0 cauley-stein's...

i'm curious if you can show some of the recent drafts CPR listings - i'd be interested in seeing these "...high picks that busted and late picks that surprised..." and any CPR projections that missed...

can you show this without divulging anything proprietary? or might previous years be listed somewhere on your website?...

Thanks for the questions. The complete methodology for CPR and a long discussion of the motivation can be found in the book Basketball Analytics: Spatial Tracking.

The first question is an interesting one. I find it very hard to determine the level of competition in a particular game for a particular individual. A "bad" team can still double one player from the opposition every time down the court. In those situations, the player should pass out of the double and may not record many points.

CPR also focuses on complete game performances. I find that in blow-outs, the stars play less. So, it's actually the closely contested matchups where the stars play the most and tend to generate those top performances.

But, I am continuously investigating ways to incorporate level of competition into the model. At this point, I think I need more information (better data) to do it well.

I find Cauley-Stein to be a fascinating case. To my eye, he has the potential to be a great defensive presence in the NBA and quite capable finishing at the hoop in transition and in pick-and-rolls. I would prioritize defense in a center over offense for today's NBA. So, Cauley-Stein fits the mold. The problem for Cauley-Stein may be that what he does well is not well-recorded in the box score.

On the other hand, for a junior, his numbers are underwhelming. Cauley-Stein never had more than 4 blocks in a game this season. By comparison, Whiteside (as a freshman, who didn't always get good minutes) had 20 games with more than 4 blocks, 3 games with at least 10 I believe. Blocks aren't a perfect measure of interior defense, but when limited to box score metrics, they are the best we have to go on.

The stats can't tell us everything about the player. This is especially true when dealing with only box-score numbers. Cauley-Stein's low ranking reflects that he hasn't put up "star" box score numbers (for a junior) this season. It doesn't mean he isn't a good prospect.

steveshea wrote:
Thanks for the questions. The complete methodology for CPR and a long discussion of the motivation can be found in the book Basketball Analytics: Spatial Tracking.

The first question is an interesting one. I find it very hard to determine the level of competition in a particular game for a particular individual. A "bad" team can still double one player from the opposition every time down the court. In those situations, the player should pass out of the double and may not record many points.

CPR also focuses on complete game performances. I find that in blow-outs, the stars play less. So, it's actually the closely contested matchups where the stars play the most and tend to generate those top performances.

But, I am continuously investigating ways to incorporate level of competition into the model. At this point, I think I need more information (better data) to do it well.

I think pace is an issue also. For Joseph Young to be rated almost as highly as Frank Kaminsky at the same age is a pretty big red flag. For him to be rated higher than Delon Wright seems almost criminal.

I'm curious - using your methodology - how the Iona stars would rate. Or maybe the BYU stars. There are guys whose numbers would look exceptional is you took their best 10 games in each stat & ignored pace & quality of comp. Is having the data set limited to 75 players helping keep your outliers somewhat to a minimum?

Young's case looks pretty decent to me. His shooting peripherals are deadly with 93% FT this year, takes 7 3s game and while "only" 36% from 3, was over 40% last two seasons. Looks like a fast athlete who can drive to the basket highlight videos. May be a place for him as bench scorer. Think he should higher than late 50s where he's currently mocked

Data is now complete

Statman wrote:
I think pace is an issue also. For Joseph Young to be rated almost as highly as Frank Kaminsky at the same age is a pretty big red flag. For him to be rated higher than Delon Wright seems almost criminal.

I'm curious - using your methodology - how the Iona stars would rate. Or maybe the BYU stars. There are guys whose numbers would look exceptional is you took their best 10 games in each stat & ignored pace & quality of comp. Is having the data set limited to 75 players helping keep your outliers somewhat to a minimum?

Yes, since CPR does not adjust for pace or quality of competition, I would recommend anyone using the metric be cognisant of the teams' strength of schedule and pace. I would also suggest that Kentucky players are underrated since their unique situation of being incredibly deep on talent limited opportunities for players to post "transcendent" statistical performances.

I haven't seen CPR grossly overrate a "small" school/"small" division prospect to the level of a 7 or 8 rating. Jimmer Fredette got to a score of 5.8, just edging Kawhi Leonard and Klay Thompson in his class. If a BYU or Iona player got to a rating of 4.5, I wouldn't interpret that as necessarily 1st round talent. If one got in the 8-10 range, that would warrant some attention.

Great work guys.

Not taking part, though I've noticed Cliff Alexanders DX Rating is inaccurate - Should be 46 not 64 (Is based on most recent version) TJ McConnell listed at 46 in current tables, 63 on current DX.

It's probably worth checking these numbers now or, getting Givony/someone at DX to check against most up to date rankings when it gets published, as there may be an update. I imagine a lot of people will be interested in both differences between consensus stat and DX rank, and a combined prediction value.

Here are my initial ratings:
Rank Name "Longevity"
1 Justise Winslow 2.02
2 D'Angelo Russell 1.83
3 Frank Kaminsky 1.79
4 Stanley Johnson 1.66
5 Jahlil Okafor 1.63
6 Karl-Anthony Towns 1.62
7 Delon Wright 1.57
8 Willie Cauley-Stein 1.51
9 Rondae Hollis-Jefferson 1.43
10 Tyus Jones 1.32
11 Jerian Grant 1.28
12 Kevon Looney 1.18
13 Sam Dekker 1.12
14 Myles Turner 0.99
15 Cameron Payne 0.96
16 Bobby Portis 0.92
17 Robert Upshaw 0.80
18 Chris McCullough 0.79
19 Montrezl Harrell 0.76
20 Kelly Oubre 0.76
21 Christian Wood 0.75
22 R.J. Hunter 0.65
23 Trey Lyles 0.61
24 Devin Booker 0.60
25 Jarell Martin 0.55
26 Justin Anderson 0.51
27 T.J. McConnell 0.29
28 J.P. Tokoto 0.28
29 Aaron White 0.22
30 Richaun Holmes 0.18
31 Michael Frazier 0.18
32 Anthony Brown 0.08
33 Dakari Johnson 0.03
34 Jordan Mickey -0.04
35 Andrew Harrison -0.05
36 Michael Qualls -0.09
37 Pat Connaughton -0.11
38 Terry Rozier -0.11
39 Rakeem Christmas -0.13
40 Tyler Harvey -0.14
41 Josh Richardson -0.19
42 Alan Williams -0.22
43 Vince Hunter -0.25
44 Norman Powell -0.26
45 Quinn Cook -0.29
46 Darrun Hilliard -0.30
47 Chris Walker -0.30
48 Olivier Hanlan -0.34
49 Julian Washburn -0.40
50 Seth Tuttle -0.42
51 Larry Nance -0.43
52 Aaron Harrison -0.53
53 Keifer Sykes -0.53
54 Rashad Vaughn -0.54
55 Chasson Randle -0.58
56 Branden Dawson -0.61
57 Brandon Ashley -0.64
58 Dezmine Wells -0.76
59 Treveon Graham -0.79
60 TaShawn Thomas -0.83
61 Wesley Saunders -0.84
62 Cliff Alexander -0.87
63 D.J. Newbill -0.88
64 Corey Hawkins -0.89
65 Terran Petteway -0.93
66 Juwan Staten -0.97
67 Derrick Marks -1.01
68 Joshua Smith -1.05
69 Shannon Scott -1.06
70 Ryan Boatright -1.11
71 Jonathan Holmes -1.15
72 Joseph Young -1.17
73 Travis Trice -1.22
74 Marcus Thornton -1.33
75 Le'Bryan Nash -1.49

I will continue to tinker until the draft so these may change over time but this is what I have for now.

dtjmcauliffe wrote:Great work guys.

Not taking part, though I've noticed Cliff Alexanders DX Rating is inaccurate - Should be 46 not 64 (Is based on most recent version) TJ McConnell listed at 46 in current tables, 63 on current DX.

It's probably worth checking these numbers now or, getting Givony/someone at DX to check against most up to date rankings when it gets published, as there may be an update. I imagine a lot of people will be interested in both differences between consensus stat and DX rank, and a combined prediction value.

The DE rankings are the top 100 prospect list excluding all of the players not on our joint list (largely just the internationals)

Crow wrote:If there is enough room, it would be useful to see the average absolute differences from DX and the models. Maybe number of highest and lowest rankings of players by each model too. And / or the overall correlations.

If you want to continue to study the rankings, maybe look at avg. rank when 1-2 biggest outliers are thrown out. So far 15 guys have a top 10 ranking but only ten have 2 or more. The worst rankings for a player also receiving a top 10 ranking are for Booker, Cauley-Stein and Maccollogh (projected outside the draft entirely) The same for worst average ranking for such players. 7 of the 15 with a top 10 ranking also have a ranking of 20 or below. Only two outside the consensus top 40 have a top 20 ranking. Onky two outside the consensus top 20 have two top 20 rankings.

A 50/50 blend of DX and Consensus of Models puts KAT on top by a little. In just over 1/3rd of the cases the DX and Consensus rankings are within 5 spots. In 40% of cases the difference is 10 or more. So in about 1/4th of cases are between 5-10 spots different. Only a handful of guys ranked in first round by DX are projected as consensus of models second rounders, the worst discrepancy being for Booker.

Frankly, I'm just not knowledgable enough about statistics to know if there is an ideal ranking method for this. Now that the data is available, I hope it can be exploited and analyzed by everyone to provide for additional insights.