How Do NCAA Statistics Translate to the NBA? (NICHOLS 09)
Posted: Fri Apr 22, 2011 2:16 am
Author Message
JNichols42887
Joined: 18 Aug 2005
Posts: 118
PostPosted: Wed Mar 18, 2009 11:25 am Post subject: How Do NCAA Statistics Translate to the NBA? Reply with quote
I just did a study over at Basketball-Statistics.com in which I attempt to see how well college stats correlate with NBA stats. You can find it here:
http://basketball-statistics.com/howdon ... henba.html
It's not incredibly complicated, but I do have a system in the works that will use multiple regression analysis to predict a player's box score stats based on a variety of factors including height, weight, NCAA experience, college numbers, etc. I know Kevin Pelton has done a lot of work in this regard and John Hollinger has his prediction system as well, so I guess this is just my stab at it.
-Jon Nichols
Back to top
View user's profile Send private message AIM Address
thref23
Joined: 13 Aug 2007
Posts: 82
PostPosted: Wed Mar 18, 2009 12:29 pm Post subject: Reply with quote
This is a topic that always interests me, especially given the limited amount of college stats available, relatively speaking. I am curious to see what you pull out of your hat.
I have not attempted to do work on the subject, but I personally anticipate that some potentially important factors not typically examined (partially because they can't easily be incorporated into a formula) are:
1.) Strength of schedule. This is probably obvious, but obviously there are huge differences in strength of schedule between certain teams.
2.) Pre-college high-school class rank. I am interested in seeing this incorporated within a formula, or a separate formula calculated for different rank classes. I feel that some players who are already well rounded at the college level may focus more on strengthening their weaknesses at the expense of putting up statistics. And among other things, I imagine that this can help predict the rate at which a player can be expected to improve as he ages. And especially with the age limit in place now, I expect that you might also occasionally get the "genius in a lower level high school class" lack of motivation syndrome at times.
(these theories might or might not help provide some sort of explanation as to why John Hollinger initially had found that it was a positive "if a player had a previous season that was better than the one just before the draft.")
3.) Height. But I am talking about height as a negative more than a positive. Meaning, height gives more of an advantage at the college level than it does the pro level. Once that advantage is gone, what happens then? A recent case study: Pat Calathes, perhaps? Of course many tall NCAA players have gone on to have nice NBA careers, but I assume they would have still been statistically desireable even if knocked a bit for their height.
(John Hollinger's system incorporates this on a certain level)
4.) Strength of teammates. This is worth examining, at least. Random example to start with, a knock on George Hill before last season's draft was that he was a bit TO prone. I would assume its hard not to be TO prone when you are the main scoring option on a not so great NCAA team. He is a little TO prone in the NBA so far, but not that bad. Of course strength of opponent also comes into play.
Back to top
View user's profile Send private message Send e-mail
JNichols42887
Joined: 18 Aug 2005
Posts: 118
PostPosted: Wed Mar 18, 2009 1:23 pm Post subject: Reply with quote
The strength of opponent, strength of teammates, and high school rank data are not included, but I do include height in my projections. I ran the numbers on DaJuan Blair because I was curious to see how he projected as a rebounder in the NBA despite only being 6'7, and he came out smelling like roses. I believe he projected to get over 10 rebounds per 36 minutes in the NBA, higher than Blake Griffin (who's 6'10). But that was despite his height, not because of it.
Back to top
View user's profile Send private message AIM Address
Kevin Pelton
Site Admin
Joined: 30 Dec 2004
Posts: 828
Location: Seattle
PostPosted: Wed Mar 18, 2009 2:31 pm Post subject: Reply with quote
I looked at rookie-year correlations back at the old APBR_analysis group. It's interesting to note that the correlations get much better when multiple seasons are used, though that might have something to do with the kind of cutoffs on the player samples. Was there a minimum career minute requirement you used, Jon?
For the most part, the relative ranking is pretty similar. I'm a little surprised rebounding came out behind blocks.
Back to top
View user's profile Send private message Send e-mail Visit poster's website
JNichols42887
Joined: 18 Aug 2005
Posts: 118
PostPosted: Wed Mar 18, 2009 3:51 pm Post subject: Reply with quote
The cutoff was actually made for me. I used the data from B-R.com, and it didn't have the college stats of players drafted in the last 3 or so years.
Back to top
View user's profile Send private message AIM Address
BadgerCane
Joined: 05 Sep 2007
Posts: 28
Location: University of Miami-Florida
PostPosted: Wed Mar 18, 2009 4:23 pm Post subject: Reply with quote
My first concern looking here is how high the rebound and assist numbers are. My question is, when running these numbers, do you put all players in the same pot? What I'm trying to get at is that if everyone is measured together, then rebounds and assists could mean nothing more than an identification of "big man" versus "little man." I'm thinking that we could learn a lot more if, when only PG's are measured against each other, if the RSQ for assists from one level to the other changes in any significant way. Similarly, if power forwards and centers are measured against only each other, do these high rebounding numbers hold up?
Back to top
View user's profile Send private message Send e-mail AIM Address
davis21wylie2121
Joined: 13 Oct 2005
Posts: 693
Location: Atlanta, GA
PostPosted: Wed Mar 18, 2009 4:40 pm Post subject: Reply with quote
Hey Jon, any theories on why FT% doesn't have a higher (or the highest) correlation? You'd think it would be the #1 most-retained skill when you make the jump to the pros, because 15 feet uncontested is 15 feet uncontested, no matter where you go, right?
Back to top
View user's profile Send private message Visit poster's website
JNichols42887
Joined: 18 Aug 2005
Posts: 118
PostPosted: Wed Mar 18, 2009 9:03 pm Post subject: Reply with quote
BadgerCane wrote:
My first concern looking here is how high the rebound and assist numbers are. My question is, when running these numbers, do you put all players in the same pot? What I'm trying to get at is that if everyone is measured together, then rebounds and assists could mean nothing more than an identification of "big man" versus "little man." I'm thinking that we could learn a lot more if, when only PG's are measured against each other, if the RSQ for assists from one level to the other changes in any significant way. Similarly, if power forwards and centers are measured against only each other, do these high rebounding numbers hold up?
That's a good idea, and perhaps I could break it down in the future. Not all point guards had the same assists numbers in college though, so I think there's something to say about the predictive effect.
Back to top
View user's profile Send private message AIM Address
JNichols42887
Joined: 18 Aug 2005
Posts: 118
PostPosted: Wed Mar 18, 2009 9:06 pm Post subject: Reply with quote
davis21wylie2121 wrote:
Hey Jon, any theories on why FT% doesn't have a higher (or the highest) correlation? You'd think it would be the #1 most-retained skill when you make the jump to the pros, because 15 feet uncontested is 15 feet uncontested, no matter where you go, right?
I was very surprised by this as well, especially with something like blocks having a much higher correlation.
I suppose there are a couple of theories on this. Free throw shooting presumably gets better with age, and maybe just some players "mature" more than others. My other theory is that some college players only had a year or two in school, and usually they had the lower free throw rates. They just made their improvements as 20 year olds in the NBA instead of as 20 year olds in college.
Back to top
View user's profile Send private message AIM Address
Harold Almonte
Joined: 04 Aug 2006
Posts: 576
PostPosted: Thu Mar 19, 2009 10:38 am Post subject: Reply with quote
It seems that the natural skills, where there isn't a such wider room (and extra opportunities) for learning and improving (unlike 2P scoring and man defense), because they are probably a matter of instincts and genetics at every level; correlates better. 3P shooting is considered a specialized skill because the distance exclusivity, complexity, and degree of difficulty; not the same for FT shooting. But, I don't understand well why is so difficult for some players to improve FT shooting, than 3P shooting for others.
Last edited by Harold Almonte on Fri Mar 20, 2009 11:01 pm; edited 1 time in total
Back to top
View user's profile Send private message
Mountain
Joined: 13 Mar 2007
Posts: 1492
PostPosted: Thu Mar 19, 2009 1:03 pm Post subject: Reply with quote
So efficient scorers are the hardest to find just on college stats (apart from 3 pt shooting). How good are the results when you use a lottery pick or a top 7 pick as opposed to just any draft pick? How often is it worth it to draft a scorer and compared to other types? Do you have to be "double-sure" as compared to a rebound, assist or defensive guy? Or is the performance/cost value of getting an efficient scorer thru the draft above 7, 14 or period * chance of getting one > what you can do thru free-agent market or trades and more so than with the other types making draft mining and reaching for an efficient scorer worthwhile? To what point? Do too many GMs just go for a scorer as opposed to an efficient scorer? After you untangle by position is it still really hard to find an efficient perimeter position player? Is it harder with "shooters" or "drivers"?
Seems to me it would be quite important to do this translation correlation by position (and scoring type) and by draft rank. It might also make sense to at least think about role- translation of scoring performance of a #1 in a college offense versus a #2 or 3 and translations of these different experiences into the nba #1 role vs #2 or 3, #4-5, or less.
This discussion makes me think of the case of Shelden Williams. A GM is thinking or should be thinking about the issues in this thread. Reliability of rebounding vs scoring (Roy, Foye, Gay, R Brewer still on the board). Comparative rarity of big men who can fill roles to a certain performance level vs availability of perimeter players who can fill perimeter roles. Reliability of big man scoring, or efficient big men, or efficient under-height big men, or efficient under-height, heavy / slow big men moving from #1 scorer to whatever role you need or intend to you use him at.
Back to top
View user's profile Send private message
JNichols42887
Joined: 18 Aug 2005
Posts: 118
PostPosted: Fri Mar 20, 2009 10:52 am Post subject: Reply with quote
Just as a follow up, I posted an article that explains the system I'll be using to project NBA players' stats using their college stats. You can find it here:
http://basketball-statistics.com/explan ... ystem.html
JNichols42887
Joined: 18 Aug 2005
Posts: 118
PostPosted: Wed Mar 18, 2009 11:25 am Post subject: How Do NCAA Statistics Translate to the NBA? Reply with quote
I just did a study over at Basketball-Statistics.com in which I attempt to see how well college stats correlate with NBA stats. You can find it here:
http://basketball-statistics.com/howdon ... henba.html
It's not incredibly complicated, but I do have a system in the works that will use multiple regression analysis to predict a player's box score stats based on a variety of factors including height, weight, NCAA experience, college numbers, etc. I know Kevin Pelton has done a lot of work in this regard and John Hollinger has his prediction system as well, so I guess this is just my stab at it.
-Jon Nichols
Back to top
View user's profile Send private message AIM Address
thref23
Joined: 13 Aug 2007
Posts: 82
PostPosted: Wed Mar 18, 2009 12:29 pm Post subject: Reply with quote
This is a topic that always interests me, especially given the limited amount of college stats available, relatively speaking. I am curious to see what you pull out of your hat.
I have not attempted to do work on the subject, but I personally anticipate that some potentially important factors not typically examined (partially because they can't easily be incorporated into a formula) are:
1.) Strength of schedule. This is probably obvious, but obviously there are huge differences in strength of schedule between certain teams.
2.) Pre-college high-school class rank. I am interested in seeing this incorporated within a formula, or a separate formula calculated for different rank classes. I feel that some players who are already well rounded at the college level may focus more on strengthening their weaknesses at the expense of putting up statistics. And among other things, I imagine that this can help predict the rate at which a player can be expected to improve as he ages. And especially with the age limit in place now, I expect that you might also occasionally get the "genius in a lower level high school class" lack of motivation syndrome at times.
(these theories might or might not help provide some sort of explanation as to why John Hollinger initially had found that it was a positive "if a player had a previous season that was better than the one just before the draft.")
3.) Height. But I am talking about height as a negative more than a positive. Meaning, height gives more of an advantage at the college level than it does the pro level. Once that advantage is gone, what happens then? A recent case study: Pat Calathes, perhaps? Of course many tall NCAA players have gone on to have nice NBA careers, but I assume they would have still been statistically desireable even if knocked a bit for their height.
(John Hollinger's system incorporates this on a certain level)
4.) Strength of teammates. This is worth examining, at least. Random example to start with, a knock on George Hill before last season's draft was that he was a bit TO prone. I would assume its hard not to be TO prone when you are the main scoring option on a not so great NCAA team. He is a little TO prone in the NBA so far, but not that bad. Of course strength of opponent also comes into play.
Back to top
View user's profile Send private message Send e-mail
JNichols42887
Joined: 18 Aug 2005
Posts: 118
PostPosted: Wed Mar 18, 2009 1:23 pm Post subject: Reply with quote
The strength of opponent, strength of teammates, and high school rank data are not included, but I do include height in my projections. I ran the numbers on DaJuan Blair because I was curious to see how he projected as a rebounder in the NBA despite only being 6'7, and he came out smelling like roses. I believe he projected to get over 10 rebounds per 36 minutes in the NBA, higher than Blake Griffin (who's 6'10). But that was despite his height, not because of it.
Back to top
View user's profile Send private message AIM Address
Kevin Pelton
Site Admin
Joined: 30 Dec 2004
Posts: 828
Location: Seattle
PostPosted: Wed Mar 18, 2009 2:31 pm Post subject: Reply with quote
I looked at rookie-year correlations back at the old APBR_analysis group. It's interesting to note that the correlations get much better when multiple seasons are used, though that might have something to do with the kind of cutoffs on the player samples. Was there a minimum career minute requirement you used, Jon?
For the most part, the relative ranking is pretty similar. I'm a little surprised rebounding came out behind blocks.
Back to top
View user's profile Send private message Send e-mail Visit poster's website
JNichols42887
Joined: 18 Aug 2005
Posts: 118
PostPosted: Wed Mar 18, 2009 3:51 pm Post subject: Reply with quote
The cutoff was actually made for me. I used the data from B-R.com, and it didn't have the college stats of players drafted in the last 3 or so years.
Back to top
View user's profile Send private message AIM Address
BadgerCane
Joined: 05 Sep 2007
Posts: 28
Location: University of Miami-Florida
PostPosted: Wed Mar 18, 2009 4:23 pm Post subject: Reply with quote
My first concern looking here is how high the rebound and assist numbers are. My question is, when running these numbers, do you put all players in the same pot? What I'm trying to get at is that if everyone is measured together, then rebounds and assists could mean nothing more than an identification of "big man" versus "little man." I'm thinking that we could learn a lot more if, when only PG's are measured against each other, if the RSQ for assists from one level to the other changes in any significant way. Similarly, if power forwards and centers are measured against only each other, do these high rebounding numbers hold up?
Back to top
View user's profile Send private message Send e-mail AIM Address
davis21wylie2121
Joined: 13 Oct 2005
Posts: 693
Location: Atlanta, GA
PostPosted: Wed Mar 18, 2009 4:40 pm Post subject: Reply with quote
Hey Jon, any theories on why FT% doesn't have a higher (or the highest) correlation? You'd think it would be the #1 most-retained skill when you make the jump to the pros, because 15 feet uncontested is 15 feet uncontested, no matter where you go, right?
Back to top
View user's profile Send private message Visit poster's website
JNichols42887
Joined: 18 Aug 2005
Posts: 118
PostPosted: Wed Mar 18, 2009 9:03 pm Post subject: Reply with quote
BadgerCane wrote:
My first concern looking here is how high the rebound and assist numbers are. My question is, when running these numbers, do you put all players in the same pot? What I'm trying to get at is that if everyone is measured together, then rebounds and assists could mean nothing more than an identification of "big man" versus "little man." I'm thinking that we could learn a lot more if, when only PG's are measured against each other, if the RSQ for assists from one level to the other changes in any significant way. Similarly, if power forwards and centers are measured against only each other, do these high rebounding numbers hold up?
That's a good idea, and perhaps I could break it down in the future. Not all point guards had the same assists numbers in college though, so I think there's something to say about the predictive effect.
Back to top
View user's profile Send private message AIM Address
JNichols42887
Joined: 18 Aug 2005
Posts: 118
PostPosted: Wed Mar 18, 2009 9:06 pm Post subject: Reply with quote
davis21wylie2121 wrote:
Hey Jon, any theories on why FT% doesn't have a higher (or the highest) correlation? You'd think it would be the #1 most-retained skill when you make the jump to the pros, because 15 feet uncontested is 15 feet uncontested, no matter where you go, right?
I was very surprised by this as well, especially with something like blocks having a much higher correlation.
I suppose there are a couple of theories on this. Free throw shooting presumably gets better with age, and maybe just some players "mature" more than others. My other theory is that some college players only had a year or two in school, and usually they had the lower free throw rates. They just made their improvements as 20 year olds in the NBA instead of as 20 year olds in college.
Back to top
View user's profile Send private message AIM Address
Harold Almonte
Joined: 04 Aug 2006
Posts: 576
PostPosted: Thu Mar 19, 2009 10:38 am Post subject: Reply with quote
It seems that the natural skills, where there isn't a such wider room (and extra opportunities) for learning and improving (unlike 2P scoring and man defense), because they are probably a matter of instincts and genetics at every level; correlates better. 3P shooting is considered a specialized skill because the distance exclusivity, complexity, and degree of difficulty; not the same for FT shooting. But, I don't understand well why is so difficult for some players to improve FT shooting, than 3P shooting for others.
Last edited by Harold Almonte on Fri Mar 20, 2009 11:01 pm; edited 1 time in total
Back to top
View user's profile Send private message
Mountain
Joined: 13 Mar 2007
Posts: 1492
PostPosted: Thu Mar 19, 2009 1:03 pm Post subject: Reply with quote
So efficient scorers are the hardest to find just on college stats (apart from 3 pt shooting). How good are the results when you use a lottery pick or a top 7 pick as opposed to just any draft pick? How often is it worth it to draft a scorer and compared to other types? Do you have to be "double-sure" as compared to a rebound, assist or defensive guy? Or is the performance/cost value of getting an efficient scorer thru the draft above 7, 14 or period * chance of getting one > what you can do thru free-agent market or trades and more so than with the other types making draft mining and reaching for an efficient scorer worthwhile? To what point? Do too many GMs just go for a scorer as opposed to an efficient scorer? After you untangle by position is it still really hard to find an efficient perimeter position player? Is it harder with "shooters" or "drivers"?
Seems to me it would be quite important to do this translation correlation by position (and scoring type) and by draft rank. It might also make sense to at least think about role- translation of scoring performance of a #1 in a college offense versus a #2 or 3 and translations of these different experiences into the nba #1 role vs #2 or 3, #4-5, or less.
This discussion makes me think of the case of Shelden Williams. A GM is thinking or should be thinking about the issues in this thread. Reliability of rebounding vs scoring (Roy, Foye, Gay, R Brewer still on the board). Comparative rarity of big men who can fill roles to a certain performance level vs availability of perimeter players who can fill perimeter roles. Reliability of big man scoring, or efficient big men, or efficient under-height big men, or efficient under-height, heavy / slow big men moving from #1 scorer to whatever role you need or intend to you use him at.
Back to top
View user's profile Send private message
JNichols42887
Joined: 18 Aug 2005
Posts: 118
PostPosted: Fri Mar 20, 2009 10:52 am Post subject: Reply with quote
Just as a follow up, I posted an article that explains the system I'll be using to project NBA players' stats using their college stats. You can find it here:
http://basketball-statistics.com/explan ... ystem.html