The NBA Draft through modeling statistics


Nothing in sports represents quite the crapshoot that a draft is. Scouts spend all year studying prospects, yet every year teams blow high picks on busts and unearth steals in the later rounds.

Statistical analysis can theoretically aid that accuracy rate, yet at the end of the day, no matter how good your model is, some players will outperform how good they are supposed to be and vice versa.

Yet that does not mean we should ignore models all together, like the one Kevin Pelton wrote about on ESPN.com in which he ranks prospects by projected future WARP:

Age isn’t the most important factor in projecting NBA success — how players have performed in the past is still more important — but because we’re comparing prospects at different stages of the development process, we can really only understand that performance in the context of age. That’s the fundamental truth on which my draft projections are built.

I start by translating a player’s college statistics to his NBA equivalents. That produces a per-minute rating, player win% (equivalent to PER), that projects how we can expect rookies to perform in the NBA next season. By adding age, I come up with a projection of how many Wins Above Replacement Player (WARP) prospects will produce over their first five years — the amount of time teams control a first-round pick between the four-year rookie contract and one year as a restricted free agent.

As with any statistical projection, the results are far from perfect. There’s too much uncertainty about how any individual will develop to tell the difference between prospects whose projections are decimal points apart. But larger differences can be meaningful indicators of over- or undervalued players.


The biggest surprise to me in Pelton rankings is how low Victor Oladipo and Ben McLemore are, as the two potential Suns picks fell all the way to Nos. 17 and 18, respectively. Oladipo was hurt by the emphasis on previous years since he has improved so steadily each season and McLemore for his low usage rate and high turnover rate.

Unsurprisingly Nerlens Noel and Otto Porter top the list, but then Colorado forward Andre Roberson shoots out of nowhere to be No. 3 on the WARP projection list with a projected rating of 2.6. As a Pac-12 guy, I’ve always respected Roberson’s game, but this ranking seems a bit much. Writes Pelton of Chad Ford’s No. 39 player:

Roberson fits a second-round stereotype — an undersized power forward with big-time athleticism. He struggled last season trying to play more on the perimeter, but has excelled defensively and on the glass against bigger players. Consider Roberson a poor man’s Kenneth Faried.

Potential Suns pick C.J. McCollum ranks sixth thanks to having “the highest translated usage rate of any player in the top 30.”

The Suns’ old analytics team built a proprietary draft model, and new GM Ryan McDonough is no secret to the concept either, so the Suns’ selections will likely be informed by such a model.

However, as Pelton’s most recent piece in which he showcases major hits and misses of his system over the past few years shows, we have yet to get to the point where making draft picks can be boiled down to a science.

Tags: Nba Draft Statistics

  • DBreezy

    Without seeing a teams’ individual model, it’s hard to make a sweeping indictment but I tend to think that statistical modeling of college/NBDL/foreign players seems sloppy at best. As Suns fans we all know how poorly Dragic was famously rated by Hollinger, but if you go back and look at his by the numbers stats on DX you pretty much see the same thing.

    We all tend to notice things like the low ratings of guys like BMac, VO, and Shabazz but to me it’s even more about the misses. It’s seems like there are so many of them relative to actual scouts draft boards that it kinda becomes noise outside of rebounding stats.

  • NOitall

    Here is the problem with advanced statistics…

    First a disclaimer: I do believe advance stats are very helpful in assessing what HAS happened. They can be useful to predict a future outcome based on past results in order to develop strategy. And to some degree, they are useful in assessing an individuals production.

    However, that is as far as I would go with the current slate of stats available to us. I am not sure you can accurately project a player’s future production unless all the variables stay the same [meaning you don't change that players teammates, playing time, 5 man unit mix, etc...

    The big problem with advanced statistics is that the individuals that create these stats are mathematicians, not basketball minds. That is their primary skillset going in. Sure, they may love basketball and have some semblance of knowledge based on the amount of time they have spent working in the field. Yet none of them come from a mindset of the game and truly understanding the fundamentals.

    For instance, part of the big glaring hole in advanced stats is the lack of any statistical measure that can quantify a players contribution to the defensive end of the floor. PER does not take into account the defensive end of the floor save for two basic stats: blocks and steals. You might say "well, that is a defensive stat that can help determine a players contribution", yet that is the crux of an argument which proves that these stat guys don't understand the game.

    Blocks and steals [totals] are absolutely NO indication of a players defensive proficiency. Last season, JaVale McGee led all players in blocks per 36, and Ricky Rubio led in steals per 36. Does that indicate they are the best defenders in the NBA? Are they even in the top 10? No. Why, because JaVale McGee may block a lot of shots, but he misses a lot of block shots and jumps out of position more often than he successfully blocks a shot. So his stat of BPG is misleading, just as Rubio’s ability to steal the ball. I can think of 10 other big men that are better defenders than McGee and probably 20 other guards ahead of Rubio.

    So the baseline assumption that a block shot or a steal is somehow indicative of good defense is a faulty premise to begin with. Add to that the fact that those are basically the only stats used in the PER formula for the defensive end of the floor and you have one fundamentally flawed stat. Is it useful in some sort of comparison to others when all else is equal? Sure, but all else is not equal. Each team is different. Each lineup is different. Playing alongside different players changes that players game such that the outcome in a stat will change. Too many variables to be useful…

    As far as using these statistical measures to translate a players production in college [a completely different environment than the NBA - far more players, far dramatic disparity between teams and players], this exercise is completely useless as a predictor of NBA success.

    Cody Zeller played certain competition. He also played alongside players of certain ability. Kelly Olynyk played different competition and had different teammates. In the NBA, they would have played all of the same players by the end of the season. In college, they may have never played anyone the other played. That variable alone means you cannot compare the production of either player to the other. It is meaningless. It has as much meaning as if one player played in JUCO and the other played in the NBA.

    It all goes back to being able to understand a player’s abilities and whether those abilities translate in three ways: first, can they translate overall to NBA level, and second, will they translate into the system the team employs, and third, will they translate within the skillsets of the teammates that player will be playing with.

    The reason the Spurs are so good at finding players is that they aren’t enamored with length, athleticism, stats or potential. They are good at understanding how a player’s skill, IQ, and effort level fit within those three criteria. Stats do not even matter when assessing those criteria. What matters is fundamentally understanding the game, and being able to watch a player and determine if that player fits. And finally it comes down to how you use that player.

    Stat guys don’t have that ability. Coaches do. Again, I am absolutely a believer is using statistics to help make sense of what has happened to help you adjust what you might do. However, I don’t believe that stats help you inherently understand the game on their own and these stat guys simply do not have that understanding. Pelton, Hollinger and Oliver and the like are all brilliant in what they do, no doubt. But I would take Popovich and his clipboard over the amalgam of stat guys and their supercomputer any time and every time. In fact, I would take Pat Riley, Phil Jackson, Jerry Sloan, George Karl, Larry Brown and a few others [Dean Smith, Coach K, etc] over any stat guy. Those guys don’t need stats to explain what they already know.

  • Red

    so you go on and pick player X who is top of the list and then he tears his ACL in pre-season practice ;)

    Still, interesting though – are there indications of previous seasons where this model has provided a suggestion of so called “steals” (late 2nd round picks that prove to be decent players) eg. Chandler Parsons or Isaiah Thomas in 2011.

    or negatively seen: Top draft picks/lists that turned out to be rather poor? (Eg. Olowokandi or Jan Veseli)

  • Scott

    If Roberson has IQ, I see him as a taller, longer, more athletic Dudley. He compares himself to Shawn Marion, and I can see the resemblance.

    He doesn’t have star-quality, but he appears to be a solid role player.

    He needs to work on his handle and his shooting, and develop his post-college game the way Dudley did.

  • Scott

    (Upon closer inspection, I note that Roberson is only 1/2″ taller than Dudley. His wingspan is 4″ longer.)

  • Scott

    Looks like Gentry will assist with the Clips.

  • Foreveris2long

    Noitall, Good stuff. I love statistics to compliment what I witnessed. While sometimes they tell the whole story, I too would not rely heavily on them in deciding who to draft. You are dead on about PER not having a lot of defensive relevancy. I mean if a guy shot 20% from the 3 point line in college and a pro team needs someone to stretch the floor, that stat might be indicative that the scoutee might not be a good long range shooter. However guys like Pop can draft a baller without paying significant attention to college stats.I hope McD does not rely too heavily on college stats. I think Blanks relied on stats to a fault in drafting Marshall and look where that got us. Great stuff.

  • Foreveris2long

    Would any of you do the deal with Minnesota who is supposedly offering Derrick Williams, their 39 and 26 picks for the Suns #5 pick? I think the same deal is being offered to the Bobcats. While I am still hopeful Derrick Williams , who I like, has a breakout year, I say no because at # 5 we should get Noel, Len, Oladipo, Burke McCollum or McLemore. If it was #9 and 20 I might bite. 26 IMO is too far down the food chain.

  • Foreveris2long

    Oops I meant to say their #9 and 26 picks plus Derrick Williams.

  • Scott

    I never was a big fan of Derrick Williams, FWIW. And I projected him to be another Beasley-like tweener, though he doesn’t have the added burden of the mental-emotional blocks Beasley has.

  • NOitall

    Think about this potential deal in this way… Foregoing the opportunity to pick at 5 only to pick at 9 is obviously giving up more choices to choose from. In most drafts, that is a big drop, as it is widely held [and through analysis] that value AFTER the 5th pick drops considerably [meaning the ability to get someone that ultimately ends up as a quality starter in the league]. However, that is most drafts, but not this one.

    It is clear that, although there may indeed be quality players, even future stars, that NOBODY can definitively project who that my be even on a majority consensus for this draft. That means, yes you should pick up a future star, but the certainty of who that is peaks at about 5%.

    So in this draft, what are you giving up by moving down 4 spots? I am not sure you are giving up anything. Reason is, the guy you like might very well be available at 9 as he would at 5.

    So, in this proposed deal, getting 9 & 26 for the 5, may or may not be a good deal. However, getting a guy that is 21 years old, doesn’t have the mental baggage that Beasely does, and has actually shown an improvement from year one to two, is a good pick up. Why would you abandon a guy like that simply because he struggled early when he was 19 and 20 years old. He may be a tweener, but equally he may turn into a solid player in this league and there is NO evidence to the contrary [unlike Beasley who had piles of evidence in his 2nd year that he was going nowhere].

    I would do that deal in a heartbeat. I would then parlay one or both of my later 1st round picks to another team and require them the take back beasley….

  • http://va.mu/fkT6 dog training union city tn

    Amazingg things here. I’m very glad to seee your article.
    Thanks sso much and I am looking ahewd tto contact you.
    Will you please drop me a e-mail?

    My blog post: dog training union city tn