Cavs fans, bear with me here. This summer, my NBA draft-experting led me down a rabbit-hole I could not evade. As the draft approaches, a plethora of athleticism data arrives in late May and early June, and I struggle with the question: “what does it mean”? Dion Waiters is only 6’ – 4”; Jeremy Lamb has long arms; in a laboratory, Harrison Barnes jumps really high. Should I care about any of this? I embarked on a project to track how pre-draft measurements correlate to actual, eventual NBA production. In today’s post, I hope to introduce the process.
Who is this guy? I don't know, but he has a 7-foot wingspan.
I started by compiling the pre-draft measurements for every drafted NCAA player from the 2000 through 2010 drafts. This data was gleaned from the world’s most comprehensive draft website: drafexpress.com. I focused on eight measurements:
- Barefoot Height
- Wingspan
- Reach
- No-Step Vertical Leap
- Maximum Vertical Leap
- Three-Quarter Court Sprint Speed
- Lane Agility (Cone) Drill
- Bench Press reps of 185 pounds.
To my spreadsheet, I assembled every player’s Offensive Win Shares (OWS), from basketball-reference.com, for each of their first four seasons. For players drafted in 2000 through 2007, their “peak” season of OWS from their first four years is also evaluated. The analysis ignores the strike-shortened 2011 – 2012 season, hence no four-year-peak season for 2008 draftees. I purposefully chose separate offensive and defensive metrics.
Also, players were sorted into two groups by age; 21 and under as of February 1st of their rookie year, or Older. Additionally, players were categorized by the five standard basketball positions. Utilizing the positional-labels proves important, as comparing OWS across the entire spectrum of possible heights and athleticism would be meaningless; obviously both tall and short players are successful; clearly little guys are faster than big men, but both succeed.
After sorting into those various categories, I correlated each of the eight measurements with the players’ OWS’s. Each player had a maximum of five OWS values: 1st season, 2nd season, 3rd season, 4th season, and peak. Near-zero correlation meant no discernible relationship between the measurement and NBA offensive performance. Highly positive correlations reflect that players strong in that particular measurement were likely to be successful offensive players. Negative correlations can largely be regarded as near-zero; I won’t advance any theories that a certain group of players is better off being smaller or less athletic. (As a final note, speed and agility correlations were made negative; i.e. smaller sprint times resulting in larger OWS are reflected as positive correlation.)
With that as the basics of the study; the pre-draft measurements provide a fairly minimal array of predictive uses for offensive production. For five positions, two age groups per position, five seasonal OWS values, and eight measurements; there were four-hundred correlations. The graph below reflects their distribution.
As you can see, this is approximately a bell-curve centered near zero. Of the 400; only 228 (57%) are positive correlations, only sixty-five (16.3%) exceed 0.25, and only two exceeded 0.50. As a frame-of-reference, here are graphs reflecting 0.25 and 0.50 correlations:
The 0.25 correlation graph is fairly useless. In the specific case of this graph, Danny Granger was both tallest and overwhelmingly most offensively successful. This alone drove the positive correlation. The next four-tallest players were all offensively worthless.
The 0.5 correlation starts to resemble something meaningful. Four of the five highest-leapers managed successful seasons, and the fifth was Greg Oden. With one exception, the low-fliers all struggled.
Part of the draw towards the low correlations is second-round picks adding noise to the data, as every NBA flame-out was awarded zero win-shares for each season. Looking only at first-round picks, with their guaranteed contracts, provides the distribution below. There are only 320 correlations here, due to sample-size issues. For both guard positions, there were relatively few upperclassmen first round picks, so I left everyone in one age-group.
194 (60.6%) were positive, with sixty-six (20.6%) exceeding 0.25. Most encouragingly, a tantalizing eighteen rose above 0.50.
Well, hopefully I have capably communicated the basics. The only conclusion I hope you drew today is that the pre-draft size and athleticism measurements offer little predictive information relating to NBA offensive performance. Over the course of the season, I plan on providing insights into:
- What are those high-correlation measurements? How useful are they?
- What about defense? A reasonable hypothesis would say that size & athleticism are more critical there.
- Which measurements rarely or never provide strong correlation with offensive or defensive performance and hence, are reasonable to ignore?
- Are there athletic traits that NBA teams are over or under-valuing? Certainly some negative correlations are due to GM’s overpersuing players based on certain athletic profiles that do not reliably prove successful.
- Are there combinations of traits that prove highly reliable towards predicting success of a drafted player? What about failure?
- Did the hand-check rules initiated in 2004 – 2005 make speed & athleticism more important?
I hope this turns out to be an interesting and provocative series. Let me know your thoughts in the comments section.
Awesome idea. I’d add another measurement to the database: hand size. I also think that bench press stats may have to be correlated to wingspan: it is easier for a guy with short arms to bench press.
Nate Smith,
Hand-size has only started being tracked in the last few years. There isn’t enough data available to decide anything meaningful.
One number I ignored that may be relevant is “reach + vert”.
Cool! Looks like you are angling for a job with a front office!
For some time people have shown that “athletic” stats like rebounding/steals/blocks seem to imply an NBA-capable level of athleticism. Basically, that the game won’t be too fast at the next level if you are in the upper echelon of those stats at the college level. Of course, that doesn’t mean you can actually play, but the idea is that those are the things that won’t even change or get better – where you can develop techniques. I’m not sure that line of thinking is totally true. Clearly, players learn how to play NBA defense after a few seasons, but very few players increase their career FT shooting drastically. http://weaksideawareness.wordpress.com/2012/01/18/nba-players-with-mid-career-improvement-in-ft/
I think it is obvious that rules changes have created a system where the offense always starts from the perimeter. In the past, point guards were responsible for pick n rolls or entry passes, as well as spot up shooting when the defense collapses. Look at today’s point guards. That skillset is no longer preferred. Now, point guards need to be able to break down their defender, get into the paint with a quick first step, be able to finish around the hoop (seriously look at the best PGs of the 90s and look at them today – the most glaring difference has GOT to be ability to finish at the rim) and be able to shoot off the dribble. The last vestiges of the old school point guard are Jason Kidd and Andre Miller. There has been a change in post positions as well. Lateral quickness and outside shooting touch are valued because it’s good to be able to get the heck out of the paint when your lightning quick perimeter player has his head down and is barreling towards the hoop. The whole concept of the stretch 4 is part of the drive n kick paradigm vs the inside outside game of yore. Feed the big man, let him back down his man and kick it out if there’s a double team.
I think this system is incentivized by the hand check rules and the way contact is called in the paint. Any contact between a defender and a driving offensive player is a foul unless the defender does that slide into the paint with perfectly set feet and collapse like a cardboard box in the wind. Any sort of contest at all = foul shots for the offensive player. Any contact on the perimeter that causes the perimeter play to lose the ball is usually whistled a foul. If you go back and watch old games, this did not exist. So, while the way defense is called has certainly helped perimeter defenders I think it is MORE difficult for traditional post players to score today. So many of the moves that post players used to score have been taken away or severly hampered. The first thing to note is that any attempt to back down a player forcibly carries way to much risk of an offensive foul do to a flop. So PFs and Cs are afraid to be aggressive. Also, any attempt at all to do a spin move carries the risk of the “wraparound” move that is called 99% of the time.
Finally, then number one reason perimeter players are thriving is that there are so few bruising Cs that require a game plan or personnel to match up. Really, unless you are playing Dwight Howard are good mood Andrew Bynum, you might as well trot Joel Anthony out there. No one else is really enough of a load that you will beat if you don’t get into a big-man arms race.
Kevin, I’d be interested to see how collegiate awards predict any success. If a player was ACC defensive player of the year, does that have predictive power about NBA defensive rating? Also, I’d be interested in the false positive indicators. The things that GMs are overvaluing that is leading to busts.
Tsunami,
If I am angling for a front office job; I am doing it wrong.
As Nate said, you have to account for the fact that these traits aren’t entirely independent, which unfortunately makes the analysis much more difficult. That means that if you want to check, say, how important the sprint is for a point guard, you ideally want a set of point guards who are very similar in all other measurements, but very different in sprint times, which is probably (unfortunately) a very small dataset.
For instance, say that sprint time and height are both advantages for point guards, but sprint time is much more of an advantage. Because taller PG’s generally have slower sprint times, you might measure an overall negative correlation between height and performance. But we know that the taller PG’s aren’t worse because they’re taller, they’re worse because they’re slower. If a 6’4″ PG is just as fast as a 6’0″ PG, he’s certainly the better prospect. So in our attempt to correlate height with success, we have to adjust for the fact that taller PG’s are generally slower than shorter PG’s, or we will end up with the erroneous conclusion that height is a disadvantage.
Nate’s case is another example of this. Bench press reps are likely negatively correlated with wingspan. If wingspan is a much bigger advantage than bench press reps, as it probably is, it would be easy to reach the false conclusion that fewer bench press reps mean a better player.
I’m not a statistics major, so i really don’t know how to go about making these adjustments. The crux of the problem is that while you need to know how sprint speed correlates with success to assess how height correlates with success, you also need to know how height correlates with success or accurately assess how sprint speed correlates with success!
The only option I know, which may or may not work correctly here, is to simply run it through once, ignoring everything I’ve said, and get correlation values for every statistic. Then, repeat the process, but when attempting to find the correlation for height, for instance, adjust the players’ OWS values based on how gifted they are in other areas. For instance, say you have the following data:
Player—Sprint—Height—OWS
A———-3.2——6’0″—–3
B———-3.2——6’0″—–4
C———-3.2——6’4″—–5
D———-3.4——6’4″—–2
E———-3.4——6’4″—–0
The best fit line from your initial run on sprint should indicate that PG’s with a 3.2 sprint should have about 4 OWS, while PG’s with a 3.4 sprint should have about 1 OWS. For height, it should indicate that 6’0″ players have about 3.5 OWS while 6’4″ players have about 2.3 OWS. But for the second run of the height correlation, if you adjust OWS based on sprint time, you get
Player—Sprint—Height—OWS
A———-3.2——6’0″—–(3-4)=-1
B———-3.2——6’0″—–(4-4)=0
C———-3.2——6’4″—–(5-4)=1
D———-3.4——6’4″—–(2-1)=1
E———-3.4——6’4″—–(0-1)=-1
Now, the best fit line should have a positive slope, indicating that the 4 inch gain is worth about [(1+1-1)/3] – [(-1+0)/2] = 5/6 = .85 OWS after adjusting for height, correctly indicating that being taller is an advantage if speed is held constant.
Best of luck, this is going to be a hell of a project. I can’t wait to see the results!
Interestingly, it seems like if you add in some performance stats (points, rebounds, etc.) you’ll get something close to what Hollinger’s done with his draft rater. Except you’re doing it for free, and making your methods publicly available. Yay!
Interesting take on the game, Tsunami. Have you been a blown away as I have early in the season by the assist totals that big men are racking up? I’ve seen some big games. It seems that Big men like the Gasol Brothers, Diaw, Noah, AV are bumping up assists more than before. I attribute it to these dribble handoff and princeton offense where big man hit cutting wings or do pivot screens. Point guards have really become distributors and finishers much more than before. However, there are plenty of “old school” point guards that could play today: KJ, Mark Price, Stockton, Isaiah… I don’t think the skill sets have changed as much as people say they have. I think a lot of guys who would’ve been considered short shooting guards in the past are becoming points, and there’s nothing wrong with that.
Nate – yeah definitely the best PGs from 20 years ago would be just fine with todays rules. The difference is, someone like John Wall may have not even been drafted in the 1st round 20 years ago.
The idea that what you wanted most out of your point guard was an explosive first step and an ability to finish at the rim was icing on the cake, not the meat and potatoes. KJ was a great example of a PG that could get to the rim and finish and draw fouls. But KJ also shot extremely well from the line and the field. John Wall is average at the line and very below average from the field. And people have always known this about him. Still, he was picked #1 and seen as one of the few people with the potential to be a superstar. Pretty much every scouting report about KI had to give lip service to the fact that he wasn’t an elite athlete in any sense and it was always implied if not outright stated that the lack of elite athleticism would keep him from being a superstar.
Well Nate – I baited myself into compiling some stats. This link is a spreadsheet containing some stats from the top 10 point guards in 2011-2012 and 1991-1992. https://docs.google.com/open?id=0Bxo35h7IHewyT20tVE9NeTFsNXM
Unfortunately, it is difficult to compare shot selection. My hypothesis would be that today’s PGs take and make a higher % of shots from the painted area than PGs from 20 years ago. (and that the ratio of off the dribble 3s to spot up 3s is higher) However, this is very difficult to prove with stats. I tried using Free Throw Attempts per FGA as an indicator since I would assume that mid-range jumpers are going to earn less fouls than forays to the rim. However, using this ratio to indicate ‘elite finishers’ does not even work for today’s players (for whom we have shot distribution breakdowns). For example: based on subjective observation I would say that Ty Lawson is an elite finisher at the rim. The stats seem to back that up to as 35% of Lawson’s shot attempts are inside and he converts them at a blistering 63%. However, in this list of 20 players, his FTA/M would rank him 16th out of 20. Meanwhile Deron Williams is very pedestrian on this list in terms of inside shot% (7th out of 10) and his conversion rate was only 57% (8th out of 10) and yet he was 7th out of 20 in FTA/M. So it’s certainly not perfect, but there are some small conclusions you can draw from this data. Here are mine, feel free to point yours out here or I think you can comment on the spreadsheet.
1.) Chris Paul is an insane outlier in terms of PER when you look at the top 20 point guards from the 2012 and 1992 seasons combined.
2.) Kevin Johnson was a SUPEROUTLIER in terms of Free Throw Attempts per Minute, almost doubling the average of the other 19 point guards on this list. WOW. Dude GOT TO THE RACK.
3.) The top Point Guards from 2012 were drafted almost 9 spots higher than those of 1992. (note, Michael Adams and Jeremy Lin were excluded from my averaging)
4.) Today’s top point guards score a bit more and yesterday’s point guards distributed more.
5.) Russell Westbrook is a chucker and John Stockton had to be padding his assists.
Kevin, ya better put on those wine-colored glasses! Dion is going off! Haha!
This is a really cool idea, you always wonder with all the media focus on combines and workouts how well they actually translate to real performance. Have you tried any regressions at all, using the athletics measurements as actual predictors of win shares? That may be a way to address some of the issues with the traits not being completely independent of each other. I’ll be looking forward to this series, I have lots of experience with stats and love seeing them used in basketball. Good work!