What's new
By:

Statistical likelihood for stars/busts in draft

I

idiot

Well-Known Member
The analytic draft blogger (often of SB Nation's Canishoopus), Layne Vashro, has just come out with his likelihood predictions for stars, starters, bench, or busts among NCAA players in this year's draft. It uses NBA players' college & NBA stats from the past 30 years to predict this year's draftees NBA production. (I believe Pelton does something similar.)

His model predicting likelihood of stardom, busting, etc., is based on measures such as college statistical production, measurables (height, weight, age, etc.), and combine athleticism. He also has two other models I find interesting: EWP -- expected wins production in most productive NBA season; and HUMBLE -- a model that blends EWP with Ford's and DX's mock drafts (I don't find his player comparison model quite as helpful.)

Since the bust/star model is his latest addition (as it waited for combine info), I'll highlight that model here. The full results can be found here.

According the the model, the likeliest stars coming from this draft are:
Embiid 38%
Smart 33%
Gordon 31%
Payton 27%
Wiggins, Anderson, Vonleh 25%
Ennis 20%
Parker 19%

The greatest likelihoods for either maxing out as a star or starter (thus not as a bust or bench player) are:
Embiid 96%
Anderson 93%
Parker 78%
Vonleh 76%
Gordon, Ennis 71%
Payton 69%
Smart 68%
Randle, McGary 53%
Stokes 52%
Warren 50%
Wiggins, Birch 49%

A few notable (relatively) high bust likelihoods:
Early 67%
Hood 57%
Payne 49%
McDermott 48%
Stauskas 41%
Young 37%
Lavine 31%
Warren 25%
Harris 22%
Randle 18%

As with all historically based statistical models, this shouldn't be taken as any superior guide to truth. His model has had both notable "successes" and "failures" in the past compared to other draft rankings. It's just another piece of data to consider.
 
Last edited:
PS -- Lowest bust likelihoods:
Embiid, Gordon, Smart 2%
Vonleh, Wiggins 3%
Payton 4%
Anderson 6%
Parker 7%
Ennis 8%
 
Last edited:
Id like to see his historical performance

Also I was under the impression that embiid was a higher risk than noted here. Maybe that is all based on his back issues.
 
A start on both the method to his models and issues of historical performance can be found here. He goes further into technical detail in several posts on the APBRmetrics forum, among other places.
 
How does he evaluate those who did not participate in the combine when many of his metrics are based upon combine results?

Sent from the JazzFanz app
 
I wonder how many years a player gotta play to be called "bust" or decided as stiff? Some calls Kanter "bust" on this board too.
 
embiid is way too high. still raw offensively, too lanky, and back problems. there is a great amount of risk when you take embiid.

these names should scare cav fans.

yao ming
andrew bogut
andrew bynum
greg oden
brook lopez


i'll take yao's 6 great years, but still it's only 6 great years.
 
Last edited:
Statistics like these are not based on reality, only on probability. There is a big difference. When you compare drafts, it's like comparing apples to oranges.
 
From a recent twitter of his: he imputed combine scores for those who didn't participate. In that case, it's a matter of trusting his estimates for those players. I assume, though don't know (for example), that he imputed a little higher vertical score to Wiggins than Gordon showed at the combine.

From what I've read, however, the combine estimate errors he may have made are unlikely to change the model's predictions for the players very substantially, unless he just made very bad estimates. In other words, the model gives stronger weighting to those things we actually know (college production, age) than to the combine scores.

But, yes, it's right to assume that there is greater error in the predictions for those who didn't participate in the combine than for those who did. But we don't know which way the error lies because he did make specific estimates, rather than just giving non-participants estimates of average scores.
 
Fabtrey said:
embiid is way too high. still raw offensively, too lanky, and back problems. there is a great amount of risk when you take embiid.
Click to expand...

Yes, Embiid's back problems are totally unaccounted for by the model. We're still on our own (as with everybody else's speculation) in estimating how much risk is added there.
 
Eenie-Meenie said:
Statistics like these are not based on reality, only on probability. There is a big difference. When you compare drafts, it's like comparing apples to oranges.
Click to expand...

Yes, the first sentence is exactly right. Historical statistical modeling can only tell us probabilities. It misses things outside the statistics that we may be able to see with our eyes. (Though by the same token it does provide value that goes beyond what we think we can see with our eyes, by giving us reminders that players with X statistical, measurable characteristics are generally likely to succeed in the NBA according to Y probabilities.)

For the last sentence, I'd say the assumption of this model is more that we're assuming that NCAA (and NBA production) production from year to year means something similar (for a simplified example, that 16 points and 9 rebounds in the NCAA translates into a certain level of NBA production from year to year). Those are questionable assumptions since the quality of the and style of both NCAA and NBA games may differ from year to year. But I think they're fairly reasonable and even necessary if any statistical modeling based on history is going to be made.

So again the point is that Vashro's models shouldn't be taken as superseding our eyes, but that they can perhaps be used to see probabilities that aren't easily taken into account through our eyes.
 
illyasova said:
I wonder how many years a player gotta play to be called "bust" or decided as stiff? Some calls Kanter "bust" on this board too.
Click to expand...

Vashro defines "bust" for his model as no season above 2.5 win shares. Kanter hasn't gone above this mark yet, but I'd say he's fairly likely to, unless his next coach doesn't give him playing time.
 
Last edited:
68.5% of all stats are made up on the spot. 9 out of 10 people who are constipated don't give a crap. These stats mean we need to move up.
 
idiot said:
As with all historically based statistical models, this shouldn't be taken as any superior guide to truth. His model has had both notable "successes" and "failures" in the past compared to other draft rankings. It's just another piece of data to consider.
Click to expand...

I would like to know these first before I start giving any credence to this post.
 
orangello said:
I would like to know these first before I start giving any credence to this post.
Click to expand...

It doesn't provide a perfect answer (and I wish it was easier to find a fuller answer), but the link I provided in post #5 does point to some of the biggest "home runs" and biggest "swings and misses" the model makes. It also provides a ranking, as of a couple years ago, of the highest several predicted EWPs by position over the last several drafts.
 
PS, to orangello's question: if you go to the model spreadsheet I linked to in the OP, and click the RETRO tab, you can find additional useful predictions over the past few years.
 
You must log in or register to reply here.

Share this page
Back
Top