B-R is at it again

Moderator: Doctor MJ

sp6r=underrated
RealGM
Posts: 17,197
And1: 8,517
Joined: Jan 20, 2007
 

B-R is at it again 

Post#1 » by sp6r=underrated » Tue Jul 2, 2013 4:11 pm

They've added player projections for every active player in the NBA.
azuresou1
Head Coach
Posts: 7,416
And1: 1,072
Joined: Jun 15, 2009
   

Re: B-R is at it again 

Post#2 » by azuresou1 » Mon Jul 8, 2013 5:30 pm

Great site, I love that they are continuously adding new tools for public use.
Chicago76
Rookie
Posts: 1,134
And1: 228
Joined: Jan 08, 2006

Re: B-R is at it again 

Post#3 » by Chicago76 » Mon Jul 8, 2013 10:11 pm

I just looked at this and just from looking at a few players, I'm not sure what they've done really makes sense. A big part of player projection from a baseball concept (where this really started back in the day with PECOTA) is regression to the mean. By that, I mean that a player can have a really outstanding season batting 40 points higher than he ever has while hitting 25 more homeruns, but you expect him to regress somewhat to what he has done historically and to what other players have done as well. In baseball, you have thousands of players with very longish careers that offer solid benchmarks. Baseball players tend to peak fairly early and then sustain for a while before declining and there aren't really teammate effects (other than batting order/protection).

In basketball, it's a bit different. You have relatively shorter peaks and a longer player development trajectory (especially for bigs). Comparable players are more limited in supply. Rate stats like per 36 are also somewhat blended with mpg, where a guy might post superior rate stats early but not stay on the court due to foul trouble.

It is difficult to figure out how to regress basketball players to the mean each year. Take a 23 year old wing player. A lot of these guys post their best #s early in their careers and then flame out, like a Ron Mercer. Others tend to have more well rounded games and continue to improve. Others produce a lot of pure volume stuff for inferior teams before moving on to better teams, where their roles are more limited. Just as an example, if you were to project Paul George next year, would you put him in the Ron Mercer category, a steady production level, or on a course for improvement? For the system to be effective, it needs to figure out a way to statistically distinguish between the two ggroups to put guys on the correct trajectory. Right now, it appears that the system doesn't do this, because George is projected to do almost exactly what he did last season, ie, it's just blending all three categories. Either that or it is putting him in the stead production category. Looking through other guys, it appears that there are a lot of guys in the 22-25 yo range in the same boat.
RealRapsFan
Sixth Man
Posts: 1,680
And1: 893
Joined: Nov 18, 2012

Re: B-R is at it again 

Post#4 » by RealRapsFan » Tue Jul 9, 2013 1:37 pm

Chicago76 wrote:I just looked at this and just from looking at a few players, I'm not sure what they've done really makes sense. A big part of player projection from a baseball concept (where this really started back in the day with PECOTA) is regression to the mean. By that, I mean that a player can have a really outstanding season batting 40 points higher than he ever has while hitting 25 more homeruns, but you expect him to regress somewhat to what he has done historically and to what other players have done as well. In baseball, you have thousands of players with very longish careers that offer solid benchmarks. Baseball players tend to peak fairly early and then sustain for a while before declining and there aren't really teammate effects (other than batting order/protection).

In basketball, it's a bit different. You have relatively shorter peaks and a longer player development trajectory (especially for bigs). Comparable players are more limited in supply. Rate stats like per 36 are also somewhat blended with mpg, where a guy might post superior rate stats early but not stay on the court due to foul trouble.

It is difficult to figure out how to regress basketball players to the mean each year. Take a 23 year old wing player. A lot of these guys post their best #s early in their careers and then flame out, like a Ron Mercer. Others tend to have more well rounded games and continue to improve. Others produce a lot of pure volume stuff for inferior teams before moving on to better teams, where their roles are more limited. Just as an example, if you were to project Paul George next year, would you put him in the Ron Mercer category, a steady production level, or on a course for improvement? For the system to be effective, it needs to figure out a way to statistically distinguish between the two ggroups to put guys on the correct trajectory. Right now, it appears that the system doesn't do this, because George is projected to do almost exactly what he did last season, ie, it's just blending all three categories. Either that or it is putting him in the stead production category. Looking through other guys, it appears that there are a lot of guys in the 22-25 yo range in the same boat.


I do agree the dynamics of a sport such as basketball makes the analysis of the sport much different than that of a more 'one on one' sport like baseball. As such changes in role and usage (whether forced on a player by changes to their team or teammates, or a change taken on by a player eg. Battier turning himself into a 3 and D vs a more 'all around player' he was while younger) is bound to change a player

But is there any statistical support the above bolded statements? I don't believe that is at all true. Much like there is an average trajectory of baseball players, there will be an average trajectory of basketball players. There will always be outliers to the trajectory ofcourse, but I don't think there is any reason to believe multiple trajectories are necessary. Even if we did, how do we know what trajectory one player fits into until after its already happened? Was it a given that Mercer was going to fit differently than, say, Paul Pierce? And if we have 2 potential trajectories, why not 3? or 4? I think when we are trying to estimate how good, or how much better/worse, a player will become using a singular trajectory (or singular to position if one likes) is the logical choice.

I remember reading an article (from WoW I believe) that showed players tend to improve over their first 3 years in the league, stay relatively constant for their next 5-7 years (or something in that range) and then start to fall off right around the age of 30. (take those numbers with a grain of salt as I didn't bother trying to track down the article so they may be off a bit). Now I don't have any personal statistical analysis that supports that, but that more or less fits my eye test. After year 3 the player you have is likely to be the player you will get if all else remains the same (injury, role). When they hit 30 we should expect their production to deteriorate. (that said, I do tend to find players whose game is based on athleticism drop off much quicker - but that in and of itself will beg alot of subjectivity based on what is athleticism vs skill, and is someone athletic or skilled)
Optimism Bias is the tendency of individuals to underestimate the likelihood they will experience adverse events. Optimistic bias cannot be reduced, and by trying to reduce the optimistic bias the end result was generally even more optimistically biased
Chicago76
Rookie
Posts: 1,134
And1: 228
Joined: Jan 08, 2006

Re: B-R is at it again 

Post#5 » by Chicago76 » Tue Jul 9, 2013 8:32 pm

RealRapsFan wrote:But is there any statistical support the above bolded statements? I don't believe that is at all true.

Much like there is an average trajectory of baseball players, there will be an average trajectory of basketball players. There will always be outliers to the trajectory ofcourse, but I don't think there is any reason to believe multiple trajectories are necessary. Even if we did, how do we know what trajectory one player fits into until after its already happened? Was it a given that Mercer was going to fit differently than, say, Paul Pierce? And if we have 2 potential trajectories, why not 3? or 4? I think when we are trying to estimate how good, or how much better/worse, a player will become using a singular trajectory (or singular to position if one likes) is the logical choice.

I remember reading an article (from WoW I believe) that showed players tend to improve over their first 3 years in the league, stay relatively constant for their next 5-7 years (or something in that range) and then start to fall off right around the age of 30. (take those numbers with a grain of salt as I didn't bother trying to track down the article so they may be off a bit). Now I don't have any personal statistical analysis that supports that, but that more or less fits my eye test. After year 3 the player you have is likely to be the player you will get if all else remains the same (injury, role). When they hit 30 we should expect their production to deteriorate. (that said, I do tend to find players whose game is based on athleticism drop off much quicker - but that in and of itself will beg alot of subjectivity based on what is athleticism vs skill, and is someone athletic or skilled)



I'm referring to pts/36, reb/36, ast/36 rate stats. I'm not making a quality assessment of those statistics in terms of efficiency, etc. And there are tons of players littered throughout the history of the league who are picked in the first round, get some serious minutes as starters right out of the box, are a more focal point of the offense, and then see their per 36 numbers decline or show no real improvement.

The reason for this is that players provide value in different ways and different skills/types of production have different aging curves. These skills are often referred to as "old man" skills or "young player" skills.

Young: things that require durability and athleticism. Some ratios and box score metrics that correlate well to durability and athleticism: minutes played, ORB rate, blk rate, stl rate, usg rate, foul draw rate (fta/2fga), etc.

Old: all types of efficiency metrics, but particularly 3pt%, ft%, low tov%, etc.

As a general rule, players who rely on young player attributes more heavily will decline more quickly than those with old man attributes. For example, a .500 TS%, below average efficiency player whose usage rate drops from 35 to 30 sees his value plummet (a McGrady), while a .550 TS% player whose usage drops the same way does not (a Bryant). A player whose foul draw rate declines and who can't combat the resulting efficiency decrease through high percentage volume, good 3 pt percentage (Wade) declines more abruptly than a very different kind of SG/SF like a Ray Allen/Reggie Miller/Dale Ellis. Height can matter too--especially when a smaller player lacks good range. Compare KJ/Isiah Thomas/Iverson to Kidd or Payton for example.

Because no two players are exactly the same, there isn't one single curve or even three curves, but in theory, an infinite number of curves. The B-R approach is very basic, basically taking a weighted average of various box score data and applying simple universal factors based upon player age. It doesn't consider player attributes. A more sophisticated approach would be to determine which players are most similar to player X at age Y, weight each similar player based upon how similar they are, and apply a customized age curve.

Going back to Mercer v. Pierce for a minute, yes, there were clear predictive indicators just looking at standard data that suggested Pierce would fare better early on. Per 36 numbers were nearly identical at 21 for both. Mercer was more efficient from the line, just as efficient inside the arc. He was slow, but Pierce wasn't a ridiculous athlete. Mercer's problem was that he didn't have range. If a slow perimeter guy can't shoot from beyond the arc, he isn't closely guarded and his defender can cheat off him. This means he can't drive and he also can't create as much for others. Pierce's assist numbers and volume increased once he figured out how to sell that threat to create more room for himself.

A couple of articles. Unfortunately, the links to APBR discussions are dead now. I remember those discussions from 2006 or so and have tried to provide the basic ideas above.


http://www.basketballprospectus.com/art ... icleid=423
http://basketballprospectus.com/unfiltered/?p=399

Return to Statistical Analysis