Author Message
Ed Küpfer
Joined: 30 Dec 2004
Posts: 785
Location: Toronto
PostPosted: Mon Jan 17, 2005 12:15 am Post subject: 4 Factors, times two Reply with quote
I gotta tell you, I really like the four factors idea. At the team level, it's a great compromise between the simplicity of ORTGs and DRTGs, and pouring over the multitudes of stats that track every single thing that happens on the floor.
(Don't know what I'm talking about? Dean Oliver discusses the concept of four factors here, and I wrote a little something on it too.)
Dean mentions his weights in his article -- shooting percentage (10), turnovers (5-6), offensive rebounding (4-5), getting to the foul line (2-3) -- derived from his Roboscout program. I wanted to see if I could duplicate them.
Taking every team from 1974 to 2004, I regressed the 8 factors (4 on offense, 4 on defense) against win%. Specifically, the factors I used were standardized (by season) EFG%, OR%, TO%, and FOUL% -- the last two being Turnovers and Fouls per possession (Foul = .44 * FTA), and I forced the intercept to equal zero. Here's how it looks:
Code:
Var Coef
O_EFF 10.0
O_TO -5.9
O_OR 4.1
O_FOUL 2.9
D_EFF -8.7
D_TO 5.6
D_OR 3.5
D_FOUL -3.9
The correlation between predicted and actual win% was 0.94, with everything being significant at 0.1%.
Pretty close to what Dean got, but I wonder why defensive shooting was relatively less important than offensive shooting?
_________________
ed
Back to top
View user's profile Send private message Send e-mail
Dan Rosenbaum
Joined: 03 Jan 2005
Posts: 541
Location: Greensboro, North Carolina
PostPosted: Mon Jan 17, 2005 2:27 am Post subject: Reply with quote
Nice work again, Ed. But if we included true field goal percentage (rather than effective field goal percentage), i.e. if we included free throws in the field goal percentage, I bet we could knock it down to three factors: shooting, rebounding, and turnovers.
But then we would want to know what generates good true field goal percentage. My bet is that this is where assists would come into play. They would help explain why some teams have better true field goal percentage than others. It doesn't make sense to include assists in the original regression, because holding field goal percentage constant, it is hard to imagine why an assist would have any value. Its value is almost entirely subsumed in the field goal percentage.
I think sometimes this "four factors" analysis has resulted in folks devaluing assists. And I think that is a big mistake. Because assists, especially assists from big men or high volume shooters, are highly correlated in my work with being a player that helps a team win.
Point guard assists, especially from point guards that don't score a lot, are much harder to value. It appears that a lot of these assists might be "system assists," which make valuing point guards by far the most difficult position to value using box score statistics.
Back to top
View user's profile Send private message Send e-mail Visit poster's website Yahoo Messenger
benji
Joined: 31 Dec 2004
Posts: 32
PostPosted: Mon Jan 17, 2005 12:30 pm Post subject: Re: 4 Factors, times two Reply with quote
Maybe looking at the difference between offense and defense would be better than each individually?
Back to top
View user's profile Send private message Send e-mail
Ed Küpfer
Joined: 30 Dec 2004
Posts: 785
Location: Toronto
PostPosted: Mon Jan 17, 2005 2:31 pm Post subject: Re: 4 Factors, times two Reply with quote
benji wrote:
Maybe looking at the difference between offense and defense would be better than each individually?
You're right: Offensive and Defensive Ratings predict win/loss percentages better than the four*2 factors do. (Why? I don't know. Is there some information contained within RTGs nor contained in 4 Factors? I don't think so...) The problem is that while looking at RTGs can accurately tell you how good a team is, it can't tell you why the team is good. That is, RTGs can't specify which things a team does well.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
Kevin Pelton
Site Admin
Joined: 30 Dec 2004
Posts: 978
Location: Seattle
PostPosted: Mon Jan 17, 2005 8:55 pm Post subject: Re: 4 Factors, times two Reply with quote
Ed Küpfer wrote:
Pretty close to what Dean got, but I wonder why defensive shooting was relatively less important than offensive shooting?
I don't think that really surprises me. Most seasons, offensive rating tends to be a somewhat stronger indicator of team success than defensive rating. The implication I've drawn from that and the fact that there is typically more variance in offensive ratings than defensive rating is that offense tends to dictate how a matchup between two teams plays out slightly more than defense.
What's interesting to me is that fouls show as much more important on the defensive end of the court, implying that avoiding fouls is more of a skill than drawing them is.
As far as the loss in predictive value by going from two factors to eight, I'd assume that's because the relative value of each factor is not constant from team to team. Rebounding is assumedly more important for a team that's defensive-minded than an offensive one, for example.
Back to top
View user's profile Send private message Send e-mail Visit poster's website
benji
Joined: 31 Dec 2004
Posts: 32
PostPosted: Mon Jan 17, 2005 11:03 pm Post subject: Re: 4 Factors, times two Reply with quote
Ed Küpfer wrote:
You're right: Offensive and Defensive Ratings predict win/loss percentages better than the four*2 factors do. (Why? I don't know. Is there some information contained within RTGs nor contained in 4 Factors? I don't think so...)
Actually, I meant, since you split the four factors into offense and defense...maybe finding the difference in each of the four would work better...I probably wasn't (and still aren't, heh) clear enough...
As for the ratings predicting it better...I like the Wizards as an example:
http://members.rogers.com/edkupfer/nba/ ... sgraph.htm
They and their opponents are both below average at eFG%, and equal on to and foul. But their rebounding is so good it seems makes up for it all. Of you don't subtract offensive rebounds (or add them even) their points per 100 difference from their opponents becomes nearly nonexistant and their offense doesn't look so great...but when you're counting them their offense is one of the best in the league. In this case it appears (to me at least) that offensive rebounding is more important to them than the average team to make up for their low shooting.
The weights on average might be the ones noted, but it may be different for each team and that's why the offense and defense ratings predict the win% better.
Back to top
View user's profile Send private message Send e-mail
HoopStudies
Joined: 30 Dec 2004
Posts: 705
Location: Near Philadelphia, PA
PostPosted: Tue Jan 18, 2005 12:16 pm Post subject: Reply with quote
Dan Rosenbaum wrote:
It doesn't make sense to include assists in the original regression, because holding field goal percentage constant, it is hard to imagine why an assist would have any value. Its value is almost entirely subsumed in the field goal percentage.
I think sometimes this "four factors" analysis has resulted in folks devaluing assists. And I think that is a big mistake. Because assists, especially assists from big men or high volume shooters, are highly correlated in my work with being a player that helps a team win.
Point guard assists, especially from point guards that don't score a lot, are much harder to value. It appears that a lot of these assists might be "system assists," which make valuing point guards by far the most difficult position to value using box score statistics.
Dan says this well.
Assists in some perspectives are valueless, but they clearly have value. At the team level, assists add nothing to explaining winning that isn't in efffg%. But efffg% and assists are related. I don't want to say one causes the other even though that is a tacit assumption of assists. At the large scale, you have "system assists" as Dan says that are a result of just having guys at the point slot (there are "system assists" at pretty much every position, but PG has most).
Guys who get double-teamed for their scoring -- if they have good assist totals, it is huge. Dan's work shows it. My work shows it. It's theoretically sensible. Typically these are big men, but it can be other guys. MJ was a good example of a non-post player who had to be double-teamed. Kobe gets good value for his assists. There are others but not a lot.
I don't think this is a surprise if you step back from the basic four factor analysis. But, as Dan said, some have devalued assists. I remember that even Dan initially undervalued them (a long time ago now) in one sense by implementing my indiv offensive rating formula with very little (or no?) weight on assists.
As to why the four factors don't predict win/loss as well, it's simple. The four factors go towards explaining ratings. Fundamentally, ratings determine winning. If you have some regression to try to explain ratings, you can't explain those at 100%, so you will do worse in explaining winning.
And, yeah, I think four factor analysis, in its basic simplicity, is one of the best things I've done. So incredibly obvious and powerful that it's amazing that it took so long to just come out and say it.
_________________
Dean Oliver
Author, Basketball on Paper
The postings are my own & don't necess represent positions, strategies or opinions of employers.
Back to top
View user's profile Send private message Visit poster's website
Ed Küpfer
Joined: 30 Dec 2004
Posts: 785
Location: Toronto
PostPosted: Tue Jan 18, 2005 1:17 pm Post subject: Reply with quote
WRT assists, I will repeat my position.
1. My little study of 30 or so games showed that when a player shot in a potential assist situation, there was about a 20% increase in FG%.
2. I had a lot of trouble reconciling my assist scoresheets with official boxscores. Either the official scorers used a much looser definition of assist than I can even conceive, or they were padding the numbers.
3. Subsequently, I had a conversation with a guy employed by one of the NBA stadiums to run the stats displays on the stadium screens. He told me that he felt the numbers were innaccurate due to the poor ability of the official scorers, many of whom received their jobs due to nepotism.
The last point is hardly evidence of anything, but it served to reinforce my feeling of unease with the official assist numbers. As I result, I am using the old fashioned solution to problems of every kind: I am ignoring it, and pretending no problem exists.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
Dan Rosenbaum
Joined: 03 Jan 2005
Posts: 541
Location: Greensboro, North Carolina
PostPosted: Tue Jan 18, 2005 2:05 pm Post subject: Reply with quote
Ed Küpfer wrote:
WRT assists, I will repeat my position.
1. My little study of 30 or so games showed that when a player shot in a potential assist situation, there was about a 20% increase in FG%.
2. I had a lot of trouble reconciling my assist scoresheets with official boxscores. Either the official scorers used a much looser definition of assist than I can even conceive, or they were padding the numbers.
3. Subsequently, I had a conversation with a guy employed by one of the NBA stadiums to run the stats displays on the stadium screens. He told me that he felt the numbers were innaccurate due to the poor ability of the official scorers, many of whom received their jobs due to nepotism.
The last point is hardly evidence of anything, but it served to reinforce my feeling of unease with the official assist numbers. As I result, I am using the old fashioned solution to problems of every kind: I am ignoring it, and pretending no problem exists.
I am not so worried about the errors in these assist stats. It is a problem, but given the whole host of problems we have to deal with, it seems like a small one to me.
I have tried doing a little of this charting myself recently, and I noticed a few problems.
1. What did you do on breakaway fastbreaks, where there is no possibility of an assist (except in those cases where an assist faciliatates a more exciting dunk)? If we are trying to measure the benefit of an assist, it seems that these attempts should be ignored.
2. What did you do with offensive rebounds that almost immediately led to shots? There usually is no chance of assist on these field goal attempts, so it seems that they also should not be counted.
3. What did you do with foul shots that did not come as the result of a shot attempt? In this case it is unclear whether to count these attempts as assisted or unassisted, since there is no field goal attempt.
Trying to count these cases, I suspect, would result in unassisted field goals generating more points. So if this is what you did, I suspect that the "assist" benefit that you measured might be understated.
Back to top
View user's profile Send private message Send e-mail Visit poster's website Yahoo Messenger
Ed Küpfer
Joined: 30 Dec 2004
Posts: 785
Location: Toronto
PostPosted: Tue Jan 18, 2005 2:19 pm Post subject: Reply with quote
Dan Rosenbaum wrote:
1. What did you do on breakaway fastbreaks, where there is no possibility of an assist (except in those cases where an assist faciliatates a more exciting dunk)?
I was being as generous as possible: I almost always gave an assist to a passer on a fast break, unless the break came on a steal and the stealer took the other way on his own.
Dan Rosenbaum wrote:
2. What did you do with offensive rebounds that almost immediately led to shots?
These maybe should've been counted separately.
Dan Rosenbaum wrote:
3. What did you do with foul shots that did not come as the result of a shot attempt?
I did not give an assist. The logic being, the player did something on his own to get fouled, and he should get as much credit as a player getting fouled on an iso play.
I'm not wedded to this, though. I can see tracking these separately.
Dan Rosenbaum wrote:
Trying to count these cases, I suspect, would result in unassisted field goals generating more points. So if this is what you did, I suspect that the "assist" benefit that you measured might be understated.
Yes. I was being as conservative as possible -- I simply couldn't believe how off my assist numbers were. I'd like to know if the errors are random or systemic.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
kjb
Joined: 03 Jan 2005
Posts: 864
Location: Washington, DC
PostPosted: Tue Jan 18, 2005 5:30 pm Post subject: Reply with quote
Ed: I'm seeing something similar as I'm tracking defensive stats for the Wizards. I'm tracking both blocks and steals, even though I'll use the official numbers when I'm putting together my defensive box scores. But I am noticing significant discrepancies between what I record and what the official box says. Specifically, I credit more blocks and fewer steals than does the official scorekeeper.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Ed Küpfer
Joined: 30 Dec 2004
Posts: 785
Location: Toronto
PostPosted: Tue Jan 18, 2005 5:52 pm Post subject: Reply with quote
WizardsKev wrote:
I credit more blocks and fewer steals than does the official scorekeeper.
So it wasn't just me -- I am happy to hear that Smile Can you estimate by how much the numbers are off? My assist counts were off by between 5-10%.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
kjb
Joined: 03 Jan 2005
Posts: 864
Location: Washington, DC
PostPosted: Tue Jan 18, 2005 6:56 pm Post subject: Reply with quote
I'll look through my scoresheets and see. I do remember one game in particular where Hughes was credited with 6 steals while I had him with 4, and Haywood was credited with 3 blocks when I had him with 5. The next game, Haywood officially had 1 block, when I had him at 3. Steals I can kinda understand because a lot of times, one guy pokes the ball loose, the offensive player tips it, another defender grabs it. But blocks are easy -- if the defender deflects the shot and it doesn't go in, it's a block.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
gabefarkas
Joined: 31 Dec 2004
Posts: 1313
Location: Durham, NC
PostPosted: Tue Jan 18, 2005 10:28 pm Post subject: Reply with quote
Dan Rosenbaum wrote:
I am not so worried about the errors in these assist stats. It is a problem, but given the whole host of problems we have to deal with, it seems like a small one to me.
I have tried doing a little of this charting myself recently, and I noticed a few problems.
1. What did you do on breakaway fastbreaks, where there is no possibility of an assist (except in those cases where an assist faciliatates a more exciting dunk)? If we are trying to measure the benefit of an assist, it seems that these attempts should be ignored.
2. What did you do with offensive rebounds that almost immediately led to shots? There usually is no chance of assist on these field goal attempts, so it seems that they also should not be counted.
3. What did you do with foul shots that did not come as the result of a shot attempt? In this case it is unclear whether to count these attempts as assisted or unassisted, since there is no field goal attempt.
Trying to count these cases, I suspect, would result in unassisted field goals generating more points. So if this is what you did, I suspect that the "assist" benefit that you measured might be understated.
I have to be honest and say that I'm not sure why we would need to parse these out. A basket is a basket; some are assisted and some aren't. Either way, fastbreaks, offensive rebound put-backs and non-shooting foul shots are all a part of the game. If we try to tease these things out of the data, or massage it as such, I feel like it's not really going to be as reflective of what's actually going on.
Does that make sense?
Back to top
View user's profile Send private message Send e-mail AIM Address
Dan Rosenbaum
Joined: 03 Jan 2005
Posts: 541
Location: Greensboro, North Carolina
PostPosted: Tue Jan 18, 2005 11:21 pm Post subject: Reply with quote
gabefarkas wrote:
Dan Rosenbaum wrote:
I am not so worried about the errors in these assist stats. It is a problem, but given the whole host of problems we have to deal with, it seems like a small one to me.
I have tried doing a little of this charting myself recently, and I noticed a few problems.
1. What did you do on breakaway fastbreaks, where there is no possibility of an assist (except in those cases where an assist faciliatates a more exciting dunk)? If we are trying to measure the benefit of an assist, it seems that these attempts should be ignored.
2. What did you do with offensive rebounds that almost immediately led to shots? There usually is no chance of assist on these field goal attempts, so it seems that they also should not be counted.
3. What did you do with foul shots that did not come as the result of a shot attempt? In this case it is unclear whether to count these attempts as assisted or unassisted, since there is no field goal attempt.
Trying to count these cases, I suspect, would result in unassisted field goals generating more points. So if this is what you did, I suspect that the "assist" benefit that you measured might be understated.
I have to be honest and say that I'm not sure why we would need to parse these out. A basket is a basket; some are assisted and some aren't. Either way, fastbreaks, offensive rebound put-backs and non-shooting foul shots are all a part of the game. If we try to tease these things out of the data, or massage it as such, I feel like it's not really going to be as reflective of what's actually going on.
Does that make sense?
I am very interested in determining the value of an assist. In my plus/minus work, it appears that the value of an assist is fairly high but differs a lot across types of players. I would like to learn more about this, because I think it holds a key towards understanding how to make an offense more efficient (or less efficient if we think of this from the defensive viewpoint).
If the true field goal percentage for assisted true field goal attempts is no different than that for unassisted true field goal attempts, then it can be argued that there is no value to an assist. It would appear that assists do not increase offensive efficiency.
But to make this comparison I need to compare apples and apples and not apples and oranges. Thus, I really only want to count true field goal attempts where there it is realistic for it to be assisted or unassisted. Thus, breakaway layups and putbacks are not really relevant for this analysis. Nor are free throw attempts not stemming from a true field goal attempt.
Author Message
gabefarkas
Joined: 31 Dec 2004
Posts: 1311
Location: Durham, NC
PostPosted: Tue Jan 18, 2005 11:31 pm Post subject: Reply with quote
Gotcha.
I missed the part earlier about comparing assisted vs unassisted FG%, which brings to light why you'd want to see only shots (baskets) with an actual potential for an assist. Thanks for clarifying though. It sounds pretty interesting now.
Back to top
View user's profile Send private message Send e-mail AIM Address
Kevin Pelton
Site Admin
Joined: 30 Dec 2004
Posts: 976
Location: Seattle
PostPosted: Wed Jan 19, 2005 2:03 am Post subject: Reply with quote
Ed Küpfer wrote:
I'd like to know if the errors are random or systemic.
How important this is depends on what we mean by "systemic". If we mean the system by which all teams record assists, then that's probably not a big deal. Everyone gets boosted a relatively similar amount.
If we mean there are certain players or types of players -- e.g. superstars -- who are getting credit for assists other guys aren't getting, that's a much bigger concern.
I know when I scored Sonics games, I usually found that the "missing assists" I didn't have could be directly traced to one Gary Payton. Was that a coincidence? Maybe. It's still more than a bit suspicious that the best player on the team was getting extra assist credit.
I actually recall under-crediting blocks a fair amount when I did Project D Scoresheet for the Storm. That would fit with a home-team bias theory. Even if a whole defensive scoresheet Oliver-style wasn't adapted, it would help a lot to have half-credit for steals, since they're quite often started by one player and finished by another. So I tended to have some differentiation in my steals, but about the same number, I think.
I don't really remember what my Sonics D scoresheets showed.
Back to top
View user's profile Send private message Send e-mail Visit poster's website
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Wed Apr 20, 2005 3:02 pm Post subject: 4 dimensions times 4 Reply with quote
i agree with the dean's identification of 4 the highly important factors. ed's subject title of 4 times two is consistent with your recognition of both offensive and defensive sides to these factors for study. this may clash with the idea of focus, but borrowing from moses malone who said "fo, fo, fo, fo" about how to win the playoffs, what about 4 dimensions times 4 or more keys? in part to move the list on team construction for success somewhat further into moneyball / meta team construction territory. i also gave shooting shooting even higher prominence as a third category of its own in addition to offensive teamwork and defense because ultimately scoring and winning are mostly about shooting.
dimension 1 shooting
keys
3 pt shooting
points in the paint
FTs made
fast break points
dimension 2 offensive teamwork
passing in general
turnovers
offensive rebounds
offensive effiency
dimension 3 defense
defensive efficency
defense that ends offensive opportunity before a shot (sum of turnovers, steals, block prorated, offensive fouls)
offensive rebounds
defensive shooting %
dimension 4 money and leadership
salary- distribution, pay/performance efficiency such as net PER /salary ratio (with attention by each position or role), $ total and contract length and how they all lineup with each other
management skill and fit
coaching skill and fit
intangibles of teams star(s)
current NBA rules
there is nothing new here and builds on what others have written. just a different cut at the focus/inclusion balance.
Last edited by jambalaya on Wed Apr 20, 2005 5:26 pm; edited 4 times in total
Back to top
View user's profile Send private message
bchaikin
Joined: 27 Jan 2005
Posts: 678
Location: cleveland, ohio
PostPosted: Wed Apr 20, 2005 3:55 pm Post subject: Reply with quote
I'm seeing something similar as I'm tracking defensive stats for the Wizards. I'm tracking both blocks and steals, even though I'll use the official numbers when I'm putting together my defensive box scores. But I am noticing significant discrepancies between what I record and what the official box says. Specifically, I credit more blocks and fewer steals than does the official scorekeeper.
part of this most likely is due to the official scorer's definition of what is a blocked shot. the NBA Statistician's Manual clearly states in the section for blocked shots that:
"....if a player goes up into the air for a shot but the ball is stripped by his opponent before it leaves his hand and is recovered inbounds by the defensive team, then charge the offensive player with a turnover and the defensive player who stripped the ball with a steal. if immediately following the initial stripping of the ball the offensive team retains possession of the ball, there will be no statistical entries....."
got that? that means if a player goes up for a shot, is clearly blocked from behind (as we think of a block) before letting go of the shot, if the defensive team picks up that ball its a steal for the defender that blocked it and a TO for the shooter (not a BS for the defender and a FGA that was missed by the shooter), and its nothing if the offensive team gets the ball back, no BS or ST or FGA etc...
most of us as fans would consider this to be a blocked shot and whoever gets the ball gets credit for an off or def rebound...
Back to top
View user's profile Send private message Send e-mail Visit poster's website
HoopStudies
Joined: 30 Dec 2004
Posts: 705
Location: Near Philadelphia, PA
PostPosted: Wed Apr 20, 2005 4:32 pm Post subject: Re: 4 dimensions times 4 Reply with quote
jambalaya wrote:
i agree with the dean's focus on his 4 keys. i agree with ed's additions that make 4 times two. this may clash with the idea of focus but borrowing from moses malone who said "fo, fo, fo, fo" about how to win the playoffs, what about 4 dimensions times 4 or more keys? in part to move the list on team construction for success somewhat into moneyball territory. i also gave shooting shooting even higher prominence as a third category of its own in addition to offensive teamwork and defense because ultimately scoring and winning are mostly about shooting.
dimension 1 shooting
keys
3 pt shooting
points in the paint
FTs made
fast break points
dimension 2 offensive teamwork
passing in general
turnovers
offensive rebounds
offensive effiency
dimension 3 defense
defensive efficency
defense that ends offensive opportunity before a shot (sum of turnovers, steals, block prorated, offensive fouls)
offensive rebounds
defensive shooting %
dimension 4 money and leadership
salary- distribution, pay/performance efficiency (perhaps with attention by position or role), total and contract length
management skill and fit
coaching skill and fit
intangibles of teams star(s)
current NBA rules
there is nothing new here and builds on what others have written. just a different cut at the focus/inclusion balance.
First of all, I called them the Four Factors because it sounds good, even though there were 4 for both O and D. The value of these is that they explain who wins, or at least 98% of it. It's a great way to narrow down what happened to win or lose a game.
Second, I have broken down some of the factors into fine details for use with the Sonics. There are some great insights to be gained in doing this, but, unfortunately, I can't talk about any of these Factor Keys as I'm starting to call them. It's a different breakdown than what you have above and certainly doesn't include soft/unmeasured things at this point. The value in doing this breakdown is to better understand what causes each of the Four Factors. So I think Four Factors will remain Four Factors.
Yes, there are other factors that determine whether you'll sign a player, trade a player, make a roster move, make a substitution. That's a whole different thread. What the Four Factors were about were explaining current win-loss stuff. They've bloomed into something a little more than that, but they will not stand alone when it comes to personnel moves.
_________________
Dean Oliver
Author, Basketball on Paper
The postings are my own & don't necess represent positions, strategies or opinions of employers.
Back to top
View user's profile Send private message Visit poster's website
kjb
Joined: 03 Jan 2005
Posts: 860
Location: Washington, DC
PostPosted: Wed Apr 20, 2005 4:37 pm Post subject: Reply with quote
Bob: I've seen the exact play you're citing, and I agree that's some of it. I've been tracking stuff directly from the play-by-play lately, and I'm also finding some plain ol' error -- shots and rebounds that didn't happen, missing shots and rebounds, incorrectly attributing the shooter, steals obviously attributed to the wrong guy, and some hinkty assists crediting. The errors aren't egregious -- we're talking about maybe 3-4 plays per game. It'd be nice to have a mechanism in place to correct those errors, though.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Wed Apr 20, 2005 5:09 pm Post subject: 4 factors & adding a different system description to thr Reply with quote
"... I have broken down some of the factors into fine details for use with the Sonics. There are some great insights to be gained in doing this, but, unfortunately, I can't talk about any of these Factor Keys as I'm starting to call them. It's a different breakdown than what you have above and certainly doesn't include soft/unmeasured things at this point. The value in doing this breakdown is to better understand what causes each of the Four Factors. So I think Four Factors will remain Four Factors. "
i've corrected my original post to reflect some of your clarifications about your 4 factors that i didnt state quite right.
the causes of the four factors make sense to study in detail and the insights from such study would be very valuable.
and i can understand keeping them as original intended and in-house at this time.
just used this thread as a location to add a somewhat different statement based on hearing about the 4 factors. my 4 dimensions and 4 keys move substantially from most select factors toward a more global system description but it is still intended as a touchstone list of primary factors (it doesnt included absolutely everything though it is getting fairly comprehensive) for system analysis to maximize output.
Back to top
View user's profile Send private message
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Fri Apr 22, 2005 3:36 pm Post subject: details of play, stats and digging deeper Reply with quote
you mentioned that you are working on factors keys but decline to reveal much about them.
nonetheless i can wonder aloud as an observer of the game about two different approaches for quality of play analysis and understanding the impact of details on overall success.
one would be a look at the detail of play and the numerous play actions that occur that lead to a shot, an assist or rebound. things like quality of spacing; settings picks; timing movement to get to a spot or clear it at just the right time; watching the ball; making reads off set plays; having hands ready; ball protection; foot balance; passing pace angle and location; close in tactical body positioning compared to opponent, and appropriate for spot on floor, role in play; recognizing likely shot launch timing and making the move toward the best rebounding spot at the optimal moment (not too soon to tip the opponent and give them more time to fight it); pretending to be a shot option when you are really not a main option; sliding unpredictably between passing, shooting, rebounding roles; improving help defense prospects, avoiding off ball defensive lapses that set off a chain of events leading to a high probablty shot, getting back quickly to other end or court or not; etc. the whole world of coachable player knowledge / skills.
another possibility would be more of a statistical look at inter-player, inter-skill quality interaction. how does high quality passing improve shooting, in what ways and how much. it is well known that a jason kidd or steve nash can make the game easier and more productive for teammates. you could look at all the detail about that and perhaps measure their impact or share of their teammates improvement compared to what it is was with them removed from the game or team. same thing for shaq's impact as a post player. there are also secondary and tertiary effects of a high quality rebounder (obviously more possessions, but that also tends to tire a defense a little more and i would think that shows up in the box score, but also less room for their guy to provide help defense and blocked shots, and maybe less fast break opportunities for the opponents who must work harder than usual to get the defensive rebound against a strong offensive rebounder, and maybe more freedom for second and third scoring options to keep thinking about moving around for a shot possibility a few seconds longer knowing their designated high vloume rebounder has it covered instead of shifting into a rebounding or defensive recovery focus and maybe higher FG% on pass dump offs from the first option or more effective second shots after the rebound is grabbed.
but as the above shows, these two approaches begin to blur together. there is vast territory for analysis and insight. with all the next order effects of a key player their true value goes well beyond what shows up in the box score or even their 82 games data profile, though there is a lot there you could use to get started on the project. similarly players "handicapped" by an average or below average teammate may have unlocked potential and may be more valuable in the right setting than their stats on their current team, in their current role suggests and you could try to know more about this than their coach or GM does. similarly players blessed with a teammate with a strong skill may show better than they are alone. this is well known but i havent seen much effort publicly to peel the onion and measure the inter-player, inter-skill effects.
plenty of room to keep digging deeper. it is probably going on inside certain NBA teams. but there is no reason the fan / analyst community can't do it as well if they want to take the time. and perhaps down the line they could also cooperate and maybe get further.
Back to top
View user's profile Send private message
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Fri Apr 22, 2005 5:46 pm Post subject: HiddenRelationships Data Mining with Genetic Algorithms Reply with quote
A Search for Hidden Relationships: Data Mining with Genetic Algorithms
"This paper presents an algorithm that permits the search for dependencies among sets of data (univariate or multivariate time-series, or cross-sectional observations). The procedure is modeled after genetic theories and Darwinian concepts, such as natural selection and survival of the fittest. It permits the discovery of equations of the data-generating process in symbolic form. The genetic algorithm that is described here uses parts of equations as building blocks to breed ever better formulas. Apart from furnishing a deeper understanding of the dynamics of a process, the method also permits global predictions and forecasts. The algorithm is successfully tested with artificial and with economic time-series and also with cross-sectional data on the performance and salaries of NBA players during the 94–95 season."
requires subscription
could be very interesting
[revised link in later post below]
and here is another
Teammate effects on pay
Todd Idson and Leo Kahane
Applied Economics Letters, 2004, vol. 11, issue 12, pages 731-733
"Abstract: This paper explores the question of whether a worker's pay is affected by the attributes of co-workers. Using data from the NBA on players who have switched teams, it is found that teammate attributes do influence an individual's pay. Comparisons to the NHL suggest that differential co-worker effects reflect the nature of complementarity in different production environments. "
Access to full text is restricted to subscribers.
Last edited by jambalaya on Fri Apr 22, 2005 9:41 pm; edited 1 time in total
Back to top
View user's profile Send private message
kjb
Joined: 03 Jan 2005
Posts: 860
Location: Washington, DC
PostPosted: Fri Apr 22, 2005 8:07 pm Post subject: Reply with quote
jambalaya -- the link isn't working. Also, could you make it shorter? Maybe imbed it in a word -- it's so long that this page doesn't fit on my screen.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Fri Apr 22, 2005 9:39 pm Post subject: link Reply with quote
sorry not sure why it didnt work but
use this http://econpapers.repec.org/scripts/search.asp?ft=NBA
and look at article 19 using the document numbers at top right and then by clicking thru Downloads: (external link) you can get as far i could and tried to post that page
Back to top
View user's profile Send private message
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Sat Apr 23, 2005 3:16 pm Post subject: genetic building blocks of player stats Reply with quote
i havent seen the above article yet. the abstract indicates it talks about using parts of formula to build better ones. that sounds quite interesting. a computer might be able to take tendex, PER, off rtg, def rtg, DanVAL, winval, net PER, +/-, NBA efficiency, etc. and build a better higher order performance evaluation tool.
but stepping down a level, i wondered if there would be any value to using the genetic building block concept merely to describe players and their performance data.
perhaps you could think of dirk notwiski as having the genetic composition of both a decent power forward (on the order of 10 points and 6 rebounds) and a shooting forward (on the order of 16 points and 4 rebounds)
or instead of trying to squeeze two full personalities into one player you could say his genetic code is 40% of one player type (say a representative pure point standard) and 60% of another (say a shooting point or tweener).
an approach like this would prevent you from having to fragment the universe into so many boxes to describe the variety of players. it would make it easier to see blends. it might allow the defining of player types in the abstract then apply them to individuals and positions in a fair and varying way.
Last edited by jambalaya on Sat Apr 23, 2005 4:55 pm; edited 1 time in total
Back to top
View user's profile Send private message
Dan Rosenbaum
Joined: 03 Jan 2005
Posts: 541
Location: Greensboro, North Carolina
PostPosted: Sat Apr 23, 2005 4:17 pm Post subject: Re: HiddenRelationships Data Mining with Genetic Algorithms Reply with quote
jambalaya wrote:
A Search for Hidden Relationships: Data Mining with Genetic Algorithms
"This paper presents an algorithm that permits the search for dependencies among sets of data (univariate or multivariate time-series, or cross-sectional observations). The procedure is modeled after genetic theories and Darwinian concepts, such as natural selection and survival of the fittest. It permits the discovery of equations of the data-generating process in symbolic form. The genetic algorithm that is described here uses parts of equations as building blocks to breed ever better formulas. Apart from furnishing a deeper understanding of the dynamics of a process, the method also permits global predictions and forecasts. The algorithm is successfully tested with artificial and with economic time-series and also with cross-sectional data on the performance and salaries of NBA players during the 94–95 season."
These data mining/genetric algorithm techniques do a very good job with prediction. I have a colleague who has written a number of articles using these methodologies. We also teach a course in data mining here at UNCG.
In cases where there a lots of observations and lots of variables, it can sometimes take days or even weeks for these models to arrive at an answer. Also, it sometimes is difficult to compute the marginal effects of a change in a given variable, which means that it can sometimes be tough to know which variables are important. A lot of times it is those marginal effects that we care about, e.g. how much is a steal worth relative to an assist? And then getting standard errors for those marginal effects can be really difficult. And without standard errors, it is hard to know if our findings are telling us something important or just are due to random error.
In general, I have found the academic literature to fall into a couple camps in terms of its modeling of player productivity. The first camp are folks who have a fancy econometric technique that they want to test out, e.g. data mining. These folks generally are more interested in their technique than in explaining what makes players productive. In a lot of cases the data work is very sloppy with this group.
The second camp uses simpler techniques, such as Ordinary Least Squares (OLS), and this group usually does a better job modeling player productivity. But, in general, I would say they do a worse job than many folks on this message board. This group's primary interest usually is demonstrating some economic point using NBA data, so the authors generally do not have a strong interest in doing a great job modeling NBA productivity. Such effort usually would unnecessarily complicate their papers and distract readers from the issue they are focusing on. My sense is that this group usually has not spent a lot of time reading folks like Oliver, Hollinger, or folks on this board.
Back to top
View user's profile Send private message Send e-mail Visit poster's website Yahoo Messenger
jambalaya
Joined: 30 Jan 2005
Posts: 282
PostPosted: Sat Apr 23, 2005 5:02 pm Post subject: reply Reply with quote
thanks for the perspective on the academic research to date. i agree with your general assessments. i've posted a number of titles, brief descriptions and links to academic reasearch recently in part because i thought the academic and fan/analyst bases should be more aware of each other but i tend to think i'll probably stay more observant of the fan/analyst research because of the perspective, detail and creative evolution. but keep an eye on the academic stuff to look for new ways to move beyond current state of knowledge and analytic insight.
Back to top
View user's profile Send private message
Dan Rosenbaum
Joined: 03 Jan 2005
Posts: 541
Location: Greensboro, North Carolina
PostPosted: Sat Apr 23, 2005 5:09 pm Post subject: Re: reply Reply with quote
jambalaya wrote:
thanks for the perspective on the academic research to date. i agree with your general assessments. i've posted a number of titles, brief descriptions and links to academic reasearch recently in part because i thought the academic and fan/analyst bases should be more aware of each other but i tend to think i'll probably stay more observant of the fan/analyst research because of the perspective, detail and creative evolution. but keep an eye on the academic stuff to look for new ways to move beyond current state of knowledge and analytic insight.
Please don't read me as being overly dismissive. I made some generalizations, but there is a lot of heterogeneity out there. I just think that for many people without a statistical background, it is possible to be overly impressed by these academic articles. Sometimes a lot of equations can hide a lot of problems with issues that folks on this board have thought a lot about.