Interesting Article on Draft Age and How It Correlates with Success

[DefenseWinsChamps]

PickNRoll I think the communication breakdown came when we used the terms early and late differently.

Every stat is arbitrary to a certain degree (think about our friendly conversation about Bender's vertical yesterday), but I still think this adds to the conversation. It should not be the only tool to evaluate players, but it helps.

I don't think I've asked, but if you had to pick someone in this draft, who would you take?

I'd take Murray, then Poeltl, Dunn, Hield, Brown.  Something like that.  Luwawu, Valentine, D. Jackson, and Sabonis in my second tier.  Ron Baker as my 2nd round flier.

How good do you think Murray could be?

[DefenseWinsChamps]

You really don't even need data to make these types of inferences.  It follows from truisms like:

1. players selected early are better than players selected later
2. younger players have a smaller body of work and less is known about them
3. older players have a larger body of work and more is known about them
4. as more is known, variance between expected and actual production will decrease
5. younger players have more room to improve
6. truly exceptional players tend to leave school early
7. older players who remained in college 4 yrs are less likely to be truly exceptional
8. young players who are not selected top 3 are more of a gamble
9. young players who are selected top 3 tend to be exceptional

And so on.  The conclusions in the article can either be derived logically or they're probably wrong.

All you have to do is look at the "other" classification.  For picks 4-20 in all age groups, it ranges from 50% to 90%, completely overwhelming all other results.

Mike

That's right.  He's created a formula that's looking for such extreme deviation -- guys who jumped or fell ~30 spots.  All of the interesting data is in the middle of the bell curve, being ignored.

It's a good starting point. I wouldn't be surprised if the guy tinkers with it more.

[MBunge]

The age correlation is well known. I mentioned a few times that Murray at three follows this model. But I think it is more about getting a better prospect more than age. If a guy is 18 and has all the same measurable and stats as a 21 year old you take the 18 year old because he has more development years. Not difficult. Come late draft guys have more holes and questions about their game. Guys with more years have learned how to play with those holes. While younger guys probably haven't even noticed the holes.

I agree for the most part. Anyone in this draft who this would apply to?

Hield and Dunn are the best examples imo. Hield is better than Murray now, but Murray is a million times better than where Hield was at the same age.

These studies further my hesitancy for drafting Hield or Dunn at 3.

The other thing to keep in mind is the limitations of biology and physics.  If Murray stayed until he was a senior, would he wind up averaging 60 points?  Guys who look good at 19 or 20 may be as close to their ultimate limits as 21 or 22 year olds.  They just reached them more quickly.

Basically, you need to look at the individuals players and evaluate if what they do translates from college to the NBA.

Mike

[PickNRoll]

You really don't even need data to make these types of inferences.  It follows from truisms like:

1. players selected early are better than players selected later
2. younger players have a smaller body of work and less is known about them
3. older players have a larger body of work and more is known about them
4. as more is known, variance between expected and actual production will decrease
5. younger players have more room to improve
6. truly exceptional players tend to leave school early
7. older players who remained in college 4 yrs are less likely to be truly exceptional
8. young players who are not selected top 3 are more of a gamble
9. young players who are selected top 3 tend to be exceptional

And so on.  The conclusions in the article can either be derived logically or they're probably wrong.

Good thoughts. It is nice to see it backed up with data. This allows us to step back and really look at guys without becoming biased positively or negatively about them. Who would you pick in this draft if you couldn't trade?

You're missing the point.  I love data and I value it.  But you have to apply a layer of common sense.

- A model that disagrees with known facts is wrong.
- Logical inference is stronger than statistical inference.  e.g. 20 year olds tend to be taller than 10 year olds.  A model that tells me this is correct, but useless.  I can infer it logically from the fact that people grow as they age.
- A model that produces surprising results deserves extra skepticism.
- models lose fidelity as arbitrary parameters are added

This one in particular is a model without a purpose.  If it's hard to clearly explain what your model is trying to predict, it's a non-starter.  e.g. It seems that it's trying

"to predict the relative likelihood that a player of a given age will tend to be an extreme outlier in terms of NBA production, as measured by win shares, given that a player was drafted either in the top 3 or drafted 4-60th, where outliers are players who either jumped 20 positions, fell 30 positions, or fell 250 points, where points are defined by minutes played, win shares, All-Star appearances, ROY Awards, and MVP awards".

This is not how real modeling works.

[DefenseWinsChamps]

You really don't even need data to make these types of inferences.  It follows from truisms like:

1. players selected early are better than players selected later
2. younger players have a smaller body of work and less is known about them
3. older players have a larger body of work and more is known about them
4. as more is known, variance between expected and actual production will decrease
5. younger players have more room to improve
6. truly exceptional players tend to leave school early
7. older players who remained in college 4 yrs are less likely to be truly exceptional
8. young players who are not selected top 3 are more of a gamble
9. young players who are selected top 3 tend to be exceptional

And so on.  The conclusions in the article can either be derived logically or they're probably wrong.

Good thoughts. It is nice to see it backed up with data. This allows us to step back and really look at guys without becoming biased positively or negatively about them. Who would you pick in this draft if you couldn't trade?

You're missing the point.  I love data and I value it.  But you have to apply a layer of common sense.

- A model that disagrees with known facts is wrong.
- Logical inference is stronger than statistical inference.  e.g. 20 year olds tend to be taller than 10 year olds.  A model that tells me this is correct, but useless.  I can infer it logically from the fact that people grow as they age.
- A model that produces surprising results deserves extra skepticism.
- models lose fidelity as arbitrary parameters are added

This one in particular is a model without a purpose.  If it's hard to clearly explain what your model is trying to predict, it's a non-starter.  e.g. It seems that it's trying

"to predict the relative likelihood that a player of a given age will tend to be an extreme outlier in terms of NBA production, as measured by win shares, given that a player was drafted either in the top 3 or drafted 4-60th, where outliers are players who either jumped 20 positions, fell 30 positions, or fell 250 points, where points are defined by minutes played, win shares, All-Star appearances, ROY Awards, and MVP awards".

This is not how real modeling works.

I don't disagree. I just like numbers, even if they aren't perfect. I swear you would argue with me if I said the sky is blue.

Do you have any predictive models you use or like?

[PickNRoll]

PickNRoll I think the communication breakdown came when we used the terms early and late differently.

Every stat is arbitrary to a certain degree (think about our friendly conversation about Bender's vertical yesterday), but I still think this adds to the conversation. It should not be the only tool to evaluate players, but it helps.

I don't think I've asked, but if you had to pick someone in this draft, who would you take?

I'd take Murray, then Poeltl, Dunn, Hield, Brown.  Something like that.  Luwawu, Valentine, D. Jackson, and Sabonis in my second tier.  Ron Baker as my 2nd round flier.

How good do you think Murray could be?

Allan Houston as a baseline.

[DefenseWinsChamps]

PickNRoll I think the communication breakdown came when we used the terms early and late differently.

Every stat is arbitrary to a certain degree (think about our friendly conversation about Bender's vertical yesterday), but I still think this adds to the conversation. It should not be the only tool to evaluate players, but it helps.

I don't think I've asked, but if you had to pick someone in this draft, who would you take?

I'd take Murray, then Poeltl, Dunn, Hield, Brown.  Something like that.  Luwawu, Valentine, D. Jackson, and Sabonis in my second tier.  Ron Baker as my 2nd round flier.

How good do you think Murray could be?

Allan Houston as a baseline.

Like floor? Wow.

Do you think he can transition to point?

[PickNRoll]

You really don't even need data to make these types of inferences.  It follows from truisms like:

1. players selected early are better than players selected later
2. younger players have a smaller body of work and less is known about them
3. older players have a larger body of work and more is known about them
4. as more is known, variance between expected and actual production will decrease
5. younger players have more room to improve
6. truly exceptional players tend to leave school early
7. older players who remained in college 4 yrs are less likely to be truly exceptional
8. young players who are not selected top 3 are more of a gamble
9. young players who are selected top 3 tend to be exceptional

And so on.  The conclusions in the article can either be derived logically or they're probably wrong.

Good thoughts. It is nice to see it backed up with data. This allows us to step back and really look at guys without becoming biased positively or negatively about them. Who would you pick in this draft if you couldn't trade?

You're missing the point.  I love data and I value it.  But you have to apply a layer of common sense.

- A model that disagrees with known facts is wrong.
- Logical inference is stronger than statistical inference.  e.g. 20 year olds tend to be taller than 10 year olds.  A model that tells me this is correct, but useless.  I can infer it logically from the fact that people grow as they age.
- A model that produces surprising results deserves extra skepticism.
- models lose fidelity as arbitrary parameters are added

This one in particular is a model without a purpose.  If it's hard to clearly explain what your model is trying to predict, it's a non-starter.  e.g. It seems that it's trying

"to predict the relative likelihood that a player of a given age will tend to be an extreme outlier in terms of NBA production, as measured by win shares, given that a player was drafted either in the top 3 or drafted 4-60th, where outliers are players who either jumped 20 positions, fell 30 positions, or fell 250 points, where points are defined by minutes played, win shares, All-Star appearances, ROY Awards, and MVP awards".

This is not how real modeling works.

I don't disagree. I just like numbers, even if they aren't perfect. I swear you would argue with me if I said the sky is blue.

Do you have any predictive models you use or like?

I like numbers too, and the sky IS blue.  Sometimes.    I was a science major, math minor, so I love data when it's good.

Here's an example from 82games.com:
http://www.82games.com/nbadraftpicks.htm

This is a much cleaner model.  It answers a very different question, but the question is very clear.  What is the EV of a given draft slot?  Clear, concise, unambiguous.

The formula is banal: points, rebounds, assists.  It's simple and it's not perfect but it's close enough because it relies on a simple truth: players who don't produce points, rebounds, or assists are rarely productive.  The simplicity offers another benefit -- the shortcomings of the model are apparent.  This model will fail for productive players who don't produce pts, rebounds or assists -- a defensive specialist, e.g.  We acknowledge that our model has a blind spot.

The earlier model not only has many blind spots, but it's impossible to identify them.  There are too many layers of misdirection.

[PickNRoll]

PickNRoll I think the communication breakdown came when we used the terms early and late differently.

Every stat is arbitrary to a certain degree (think about our friendly conversation about Bender's vertical yesterday), but I still think this adds to the conversation. It should not be the only tool to evaluate players, but it helps.

I don't think I've asked, but if you had to pick someone in this draft, who would you take?

I'd take Murray, then Poeltl, Dunn, Hield, Brown.  Something like that.  Luwawu, Valentine, D. Jackson, and Sabonis in my second tier.  Ron Baker as my 2nd round flier.

How good do you think Murray could be?

Allan Houston as a baseline.

Like floor? Wow.

Do you think he can transition to point?

Nope.  Shooting guard, but he can do at least as much ballhandling as Klay Thompson.  He's a scorer.

Sorry, I didn't mean floor.  I meant Allan Houston as a "rough baseline" for comparison.  Houston is more like the median.  Floor is maybe Jordan Clarkson.

[DefenseWinsChamps]

Interesting. Do you have any predictive models for the draft?

[DefenseWinsChamps]

PickNRoll I think the communication breakdown came when we used the terms early and late differently.

Every stat is arbitrary to a certain degree (think about our friendly conversation about Bender's vertical yesterday), but I still think this adds to the conversation. It should not be the only tool to evaluate players, but it helps.

I don't think I've asked, but if you had to pick someone in this draft, who would you take?

I'd take Murray, then Poeltl, Dunn, Hield, Brown.  Something like that.  Luwawu, Valentine, D. Jackson, and Sabonis in my second tier.  Ron Baker as my 2nd round flier.

How good do you think Murray could be?

Allan Houston as a baseline.

Like floor? Wow.

Do you think he can transition to point?

Nope.  Shooting guard, but he can do at least as much ballhandling as Klay Thompson.  He's a scorer.

Sorry, I didn't mean floor.  I meant Allan Houston as a "rough baseline" for comparison.  Houston is more like the median.  Floor is maybe Jordan Clarkson.

Then ceiling is?

[PickNRoll]

PickNRoll I think the communication breakdown came when we used the terms early and late differently.

Every stat is arbitrary to a certain degree (think about our friendly conversation about Bender's vertical yesterday), but I still think this adds to the conversation. It should not be the only tool to evaluate players, but it helps.

I don't think I've asked, but if you had to pick someone in this draft, who would you take?

I'd take Murray, then Poeltl, Dunn, Hield, Brown.  Something like that.  Luwawu, Valentine, D. Jackson, and Sabonis in my second tier.  Ron Baker as my 2nd round flier.

How good do you think Murray could be?

Allan Houston as a baseline.

Like floor? Wow.

Do you think he can transition to point?

Nope.  Shooting guard, but he can do at least as much ballhandling as Klay Thompson.  He's a scorer.

Sorry, I didn't mean floor.  I meant Allan Houston as a "rough baseline" for comparison.  Houston is more like the median.  Floor is maybe Jordan Clarkson.

Then ceiling is?

Tough to say.  I guess I have him pegged in a pretty narrow range.  Not as good as Ray Allen.  Too small to defend like Klay.  Not as good as Dwayne Wade.  So whatever the gap is between Allan Houston and DWade, I'd put his ceiling in there.  Dale Ellis in his prime?

[DefenseWinsChamps]

PickNRoll I think the communication breakdown came when we used the terms early and late differently.

Every stat is arbitrary to a certain degree (think about our friendly conversation about Bender's vertical yesterday), but I still think this adds to the conversation. It should not be the only tool to evaluate players, but it helps.

I don't think I've asked, but if you had to pick someone in this draft, who would you take?

I'd take Murray, then Poeltl, Dunn, Hield, Brown.  Something like that.  Luwawu, Valentine, D. Jackson, and Sabonis in my second tier.  Ron Baker as my 2nd round flier.

How good do you think Murray could be?

Allan Houston as a baseline.

Like floor? Wow.

Do you think he can transition to point?

Nope.  Shooting guard, but he can do at least as much ballhandling as Klay Thompson.  He's a scorer.

Sorry, I didn't mean floor.  I meant Allan Houston as a "rough baseline" for comparison.  Houston is more like the median.  Floor is maybe Jordan Clarkson.

Then ceiling is?

Tough to say.  I guess I have him pegged in a pretty narrow range.  Not as good as Ray Allen.  Too small to defend like Klay.  Not as good as Dwayne Wade.  So whatever the gap is between Allan Houston and DWade, I'd put his ceiling in there.  Dale Ellis in his prime?

Hmm. I'm a bit higher on him developing point skills at the next level. I compare him to Lillard, who is not a great passer/distributor, but can create for others. I like the Clarkson floor comparison. If we et something between Clarkson and Lillard with the 3rd, I'd be happy.