2010 NBA Draft: Peak Performances by Pick
Posted by Neil Paine on June 23, 2010
Last year, Justin had a great series about the expected career and rookie-contract value (in Win Shares) of a draft choice by overall pick #, and it's a must-read this week as you prepare for Thursday's event. I don't want -- or need -- to tread the same ground here today; instead, I'm interested in what kind of peak performance you can expect out of a player drafted in each slot. Because as good as longevity is, I feel like a player's peak years say a lot about his overall talent level, and the type of production he's capable of putting up, even if he only did it for a brief time. There's an old saying in baseball: "Once a player displays a skill, he owns it." Peak performance is like that -- a guy like Penny Hardaway may have only been super-elite for 1 season before getting hurt and declining, but 1 season is all it took to show he had that type of innate talent.
Now, the question becomes, how do we measure peak performance? Justin already used WS, so I'm going to look at peak Statistical +/-, defined as the highest 3-year moving weighted average* of the player's career.
(* The average for Year Y was determined by regressing each season to 222 minutes of -2.57 +/-, and weighting the previous seasons like this: Y, 66%; Y-1, 22%; and Y-2, 12%. Players who never made the NBA were said to have peaked at -2.57.)
Here are the average peak performances for each draft slot from 1976-2000:
| Overall Pick | SPM | Overall Pick | SPM |
|---|---|---|---|
| 1 | 3.42 | 31 | -1.54 |
| 2 | 2.01 | 32 | -2.04 |
| 3 | 2.19 | 33 | -1.77 |
| 4 | 1.24 | 34 | -2.40 |
| 5 | 1.33 | 35 | -2.03 |
| 6 | -0.28 | 36 | -1.81 |
| 7 | 0.61 | 37 | -1.91 |
| 8 | 0.68 | 38 | -2.08 |
| 9 | 0.52 | 39 | -1.92 |
| 10 | 0.46 | 40 | -2.12 |
| 11 | 0.59 | 41 | -2.00 |
| 12 | 0.06 | 42 | -2.31 |
| 13 | 0.10 | 43 | -1.57 |
| 14 | -0.40 | 44 | -2.07 |
| 15 | -1.17 | 45 | -1.83 |
| 16 | 0.03 | 46 | -1.63 |
| 17 | -1.12 | 47 | -1.72 |
| 18 | -0.62 | 48 | -1.76 |
| 19 | -1.75 | 49 | -2.16 |
| 20 | -1.10 | 50 | -2.44 |
| 21 | -0.81 | 51 | -2.49 |
| 22 | -1.27 | 52 | -2.12 |
| 23 | -0.77 | 53 | -2.20 |
| 24 | -0.62 | 54 | -2.34 |
| 25 | -1.76 | 55 | -2.40 |
| 26 | -1.59 | 56 | -2.46 |
| 27 | -1.67 | 57 | -1.76 |
| 28 | -2.12 | 58 | -2.10 |
| 29 | -1.32 | 59 | -2.48 |
| 30 | -1.77 | 60 | -2.16 |
If you smooth the data out, this equation emerges...
Peak ~ 4.7 * 0.9Pick# - 1.6
...Which yields this chart:
| Overall Pick | Exp Peak | Overall Pick | Exp Peak |
|---|---|---|---|
| 1 | 2.70 | 31 | -1.36 |
| 2 | 2.29 | 32 | -1.38 |
| 3 | 1.92 | 33 | -1.40 |
| 4 | 1.59 | 34 | -1.42 |
| 5 | 1.29 | 35 | -1.43 |
| 6 | 1.01 | 36 | -1.44 |
| 7 | 0.77 | 37 | -1.46 |
| 8 | 0.54 | 38 | -1.47 |
| 9 | 0.34 | 39 | -1.48 |
| 10 | 0.16 | 40 | -1.49 |
| 11 | -0.01 | 41 | -1.50 |
| 12 | -0.16 | 42 | -1.50 |
| 13 | -0.29 | 43 | -1.51 |
| 14 | -0.41 | 44 | -1.52 |
| 15 | -0.52 | 45 | -1.52 |
| 16 | -0.62 | 46 | -1.53 |
| 17 | -0.71 | 47 | -1.53 |
| 18 | -0.80 | 48 | -1.53 |
| 19 | -0.87 | 49 | -1.54 |
| 20 | -0.94 | 50 | -1.54 |
| 21 | -1.00 | 51 | -1.54 |
| 22 | -1.05 | 52 | -1.55 |
| 23 | -1.10 | 53 | -1.55 |
| 24 | -1.15 | 54 | -1.55 |
| 25 | -1.19 | 55 | -1.55 |
| 26 | -1.22 | 56 | -1.56 |
| 27 | -1.26 | 57 | -1.56 |
| 28 | -1.29 | 58 | -1.56 |
| 29 | -1.32 | 59 | -1.56 |
| 30 | -1.34 | 60 | -1.56 |
The values are compressed, especially at the back end because players who never made the NBA were given the default -2.57 value as their peak, but it still illustrates just how much the expected value diminishes when you move down a pick in the 1st round. This further confirms what DSMok1 found at APBRmetrics, which is that outside the top 10, you're just as likely to find a below-average player as you are to find an average one (hence the expected peak close to 0.00).
Finally, here are the probabilities that you get a player peaking at each SPM level with a given pick:
| Overall Pick # | >2 | 0-2 | -2-0 | <-2 | Overall Pick # | >2 | 0-2 | -2-0 | <-2 |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 56% | 36% | 4% | 4% | 31 | 8% | 20% | 0% | 72% |
| 2 | 44% | 36% | 16% | 4% | 32 | 4% | 8% | 12% | 76% |
| 3 | 48% | 28% | 16% | 8% | 33 | 0% | 20% | 12% | 68% |
| 4 | 40% | 28% | 28% | 4% | 34 | 0% | 4% | 12% | 84% |
| 5 | 32% | 24% | 28% | 16% | 35 | 0% | 16% | 12% | 72% |
| 6 | 12% | 32% | 20% | 36% | 36 | 8% | 4% | 20% | 68% |
| 7 | 24% | 32% | 32% | 12% | 37 | 4% | 12% | 20% | 64% |
| 8 | 32% | 36% | 16% | 16% | 38 | 0% | 16% | 4% | 80% |
| 9 | 12% | 48% | 24% | 16% | 39 | 0% | 8% | 20% | 72% |
| 10 | 28% | 24% | 40% | 8% | 40 | 0% | 4% | 28% | 68% |
| 11 | 28% | 24% | 36% | 12% | 41 | 0% | 12% | 16% | 72% |
| 12 | 28% | 28% | 16% | 28% | 42 | 0% | 4% | 8% | 88% |
| 13 | 28% | 16% | 28% | 28% | 43 | 8% | 4% | 24% | 64% |
| 14 | 24% | 8% | 32% | 36% | 44 | 0% | 12% | 4% | 84% |
| 15 | 12% | 16% | 16% | 56% | 45 | 8% | 16% | 0% | 76% |
| 16 | 24% | 20% | 24% | 32% | 46 | 8% | 8% | 20% | 64% |
| 17 | 12% | 20% | 12% | 56% | 47 | 0% | 16% | 20% | 64% |
| 18 | 12% | 28% | 20% | 40% | 48 | 4% | 8% | 16% | 72% |
| 19 | 8% | 8% | 24% | 60% | 49 | 0% | 12% | 8% | 80% |
| 20 | 8% | 24% | 16% | 52% | 50 | 0% | 4% | 16% | 80% |
| 21 | 8% | 28% | 16% | 48% | 51 | 0% | 0% | 8% | 92% |
| 22 | 4% | 20% | 28% | 48% | 52 | 0% | 4% | 20% | 76% |
| 23 | 8% | 24% | 36% | 32% | 53 | 4% | 4% | 8% | 84% |
| 24 | 16% | 20% | 8% | 56% | 54 | 0% | 0% | 20% | 80% |
| 25 | 8% | 12% | 12% | 68% | 55 | 0% | 5% | 5% | 89% |
| 26 | 16% | 0% | 16% | 68% | 56 | 0% | 0% | 5% | 95% |
| 27 | 4% | 4% | 32% | 60% | 57 | 5% | 0% | 16% | 79% |
| 28 | 0% | 20% | 8% | 72% | 58 | 0% | 11% | 11% | 78% |
| 29 | 8% | 8% | 40% | 44% | 59 | 0% | 0% | 8% | 92% |
| 30 | 4% | 12% | 8% | 76% | 60 | 8% | 0% | 8% | 85% |
Notice the Curse of the #6 Pick: for every Larry Bird (+9 peak), you see an inordinate number of Sharone Wrights, Ron Mercers, and Stacey Kings (-3). Meanwhile, a surprising number of good players have come from #16 (John Stockton, Ron Artest, the underrated Rickey Green), and #24 (Andrei Kirilenko, Terry Porter, Arvydas Sabonis, Sam Cassell, Latrell Sprewell).
The big takeaway, though, is that in any given draft there are only about a dozen players that will actually be any good someday; the rest of the draft is simply filled with below-average guys and players who never even made it to the league. Obviously, the ability to identify those 12 or so above-average players and not whiff on them when you have a high pick is ultimately the difference between a contender and a cellar-dweller. But as these charts illustrate, doing that is a lot easier said than done.
June 23rd, 2010 at 1:35 pm
Excellent work, Neil!
June 23rd, 2010 at 1:46 pm
Thanks, although it's pretty much just a confirmation of your earlier findings. :)
June 23rd, 2010 at 3:13 pm
Why did you choose -2.57 SPM? I typically use -3 as replacement level.
June 23rd, 2010 at 3:28 pm
That was the value that I found I needed to regress to in the formula that best predicted the following season's SPM. You regress everyone's SPM to the mean by adding 222 minutes of -2.57 SPM to their actual stats, then average the previous 3 seasons' regressed SPMs according to the weights I listed, and make an age adjustment to get a projection for year Y. Then, if you want to know their "true skill" for year Y-1, you just subtract out the age adjustment. Peak "true skill" is what I used here, which is why guys who had no stats simply got regressed to -2.57 with no other data included in their true skill calculation.
June 23rd, 2010 at 5:57 pm
This blog is more and more relying on SPM data, which Neil seems to have going back to 1960 or earlier, yet which for some reason is not available anywhere on the site. It is starting to make reading the blog very annoying, because we (the readers, the public) don't have access to any of the data being used to support the analysis in the blog.
June 23rd, 2010 at 6:06 pm
I try to alternate SPM with other stats like Win Shares, it just so happens that Justin already did a draft analysis using WS. Would I like to see SPM on the site? Sure. That's up to Justin, though. It's ultimately his call, I'm just an employee.
June 23rd, 2010 at 6:17 pm
Neil -- I understand it's not your decision, but it's frustrating as a fan of the site and your blog not to have access to SPM, which you clearly view as a far better metric than WS.
June 23rd, 2010 at 6:31 pm
That's not necessarily true, I think Win Shares have value and they have their place. It would be more accurate to say that I clearly view SPM as a far better metric than PER, since they're both per-minute and SPM has the advantage that a) its weights were derived empirically, not from guesswork; and b) it at least tries to take defense into account.
Btw, that wasn't meant as a slam to Mr. Hollinger, whose work I admire very much and who was actually responsible in part for my interest in advanced basketball stats.
June 23rd, 2010 at 7:20 pm
interesting about after pick 10. Would love to see something if certain teams are better than others with later round picks.
82games.com touched on this last year (but curiously never finished), but also used more traditional stats analysis for it.
Or just something like "which teams beat the average, which don't" taking into account all draft positions for a team.
June 23rd, 2010 at 8:23 pm
Can we get a poll / petition going to convince Mr. Kubatko to add SPM? Come on, you guys give us hours of entertainment and reference materials for free. He owes us.
June 23rd, 2010 at 8:39 pm
I believe part of the delay is that Justin wants to fine-tune the regression himself (he did teach statistics at Ohio State once upon a time, after all)... He hasn't had any time to do it, though, because he's working on the College Football Reference site this summer (that's right -- college football reference!!!).
June 24th, 2010 at 9:39 am
Hey guys - you can download an SPM spreadsheet here:
http://www.mediafire.com/?oqulydhmzwh
It uses the latest formula (http://www.basketball-reference.com/blog/?page_id=4122).
June 24th, 2010 at 11:15 am
Neil -- Thanks a lot for the spreadsheet. Much appreciated.