Loading ...
Sorry, an error occurred while loading the content.

Re: [APBR_analysis] Re: Assist tracking update - Dave Berri

Expand Messages
  • bchaikin@aol.com
    Plenty of papers available through... http://www.csub.edu/~dberri/ Some abstracts and papers at http://www.csub.edu/~dberri/research.html i eventually found
    Message 1 of 24 , Feb 3, 2004
    • 0 Attachment

      Plenty of papers available through...

      http://www.csub.edu/~dberri/

      Some abstracts and papers at

      http://www.csub.edu/~dberri/research.html

      i eventually found his homepage searching on david not dave....

      his applications research page at:

      http://www.csub.edu/~dberri/researchapps.html

      has 3 links for 2002-03 NBA player rankings/studies but all appear to be inactive, so i could not review those for any comparison's to other work...

      i see only one active article link at his research page (other than the abstract synopses), the one for the PDF "who is most valuable? measuring the player's production of wins in the NBA". the paper is a 1999 paper about the 97-98 season, so i have not much to go on about berri's work other than this, and while i have had a number of stats courses i am not a stats major and thus don't understand all his work presented in the article because the derivations are not completely shown (hard to do in a published article)...

      having said that, however, the article's conclusion says, quote "...how productive is each NBA player? the methods presented (here)....... provide an accurate answer to the question...... such evaluations can obviously be utilized with respect to free agent signings, player-for-player trades, the allocation of minutes, and also to determine the impact changes in coaching methods or strategy have had on an individual's productivity...."

      it goes on to say "...with respect to the academic literature the methods reported here are also likey to be useful. previous methods cited appear to have used the incorrect functional form, an incorrect set of data, and failed to account for the importance of team tempo. the correction of each of these shortcomings has led to an accurate appraisal of playing talent that can be utilized in the study of such issues as racial discrimination and worker remuneration under the auspices of differing institutional arrangements..."....

      well that was a mouthful....

      anyone know if any of his more current research is online for perusal, or if any of his work has been used by the league or any nba teams?...
    • Dean Oliver
      ... quote ...how ... provide an ... obviously be utilized ... allocation of ... methods or ... methods ... appear to have ... failed to ... these ... that can
      Message 2 of 24 , Feb 3, 2004
      • 0 Attachment
        --- In APBR_analysis@yahoogroups.com, bchaikin@a... wrote:
        > having said that, however, the article's conclusion says,
        quote "...how
        > productive is each NBA player? the methods presented (here).......
        provide an
        > accurate answer to the question...... such evaluations can
        obviously be utilized
        > with respect to free agent signings, player-for-player trades, the
        allocation of
        > minutes, and also to determine the impact changes in coaching
        methods or
        > strategy have had on an individual's productivity...."
        >
        > it goes on to say "...with respect to the academic literature the
        methods
        > reported here are also likey to be useful. previous methods cited
        appear to have
        > used the incorrect functional form, an incorrect set of data, and
        failed to
        > account for the importance of team tempo. the correction of each of
        these
        > shortcomings has led to an accurate appraisal of playing talent
        that can be utilized
        > in the study of such issues as racial discrimination and worker
        remuneration
        > under the auspices of differing institutional arrangements..."....
        >
        > well that was a mouthful....
        >
        > anyone know if any of his more current research is online for
        perusal, or if
        > any of his work has been used by the league or any nba teams?...

        I get stuff from him every so often. I review a lot of sports econ
        these days and he's doing almost all the basketball work. Journal of
        Sports Economics doesn't appear to be online.

        It was his recent work, which included a revision on that 1999 paper,
        that we talked about a lot recently. Basically, his technique for
        identifying the value of stats is a regression of team wins vs team
        stats. He originally did it one way that had a bit of collinearity
        issues, then redid it based upon a possessions-based argument that I
        made, and things worked out a little different. It was actually very
        interesting if you look back on that 1999 paper online that his
        regression pretty much pegs the possession formula. The coefficients
        on FGA and TO are about 1. The coefficient on OR is about -1 and the
        coefficient on FTA is about 0.4. So he was getting only marginal
        performance above what teams would do with possessions. So he fixed
        that. He recently did a paper in Journal of Sports Econ on the value
        of a win in basketball -- it's about $100K in gate revenue (in 1992-
        96) and about 5-6 times that for a playoff win. I think those
        numbers double for today's money.

        His stuff got used by the Sonics before they found my stuff in 2000.
        They already used a kind of linear weights technique before and
        that's all his really is.

        I will add that Dave Berri is actually a very funny guy, funnier than
        Dave Barry the columnist. His papers don't let that on. He is
        pretty much known as the stand up comedian of econ.

        Western Econ Association meeting this summer is in Vancouver, BC. I
        will be going, as will Dave. No coincident SABR meeting this time.


        DeanO
        www.basketballonpaper.com
      • Michael Tamada
        ... From: Dean Oliver [mailto:deano@rawbw.com] Sent: Tuesday, February 03, 2004 6:20 PM [...] ... My recollection had been that when his work first came to the
        Message 3 of 24 , Feb 6, 2004
        • 0 Attachment
          -----Original Message-----
          From: Dean Oliver [mailto:deano@...]
          Sent: Tuesday, February 03, 2004 6:20 PM

          [...]

          >It was his recent work, which included a revision on that 1999 paper,
          >that we talked about a lot recently. Basically, his technique for
          >identifying the value of stats is a regression of team wins vs team
          >stats. He originally did it one way that had a bit of collinearity

          My recollection had been that when his work first came to the attention
          of this group (I think it was then that we coined the term "laugh test"
          with respect to his Rodman MVP result), that he had used a Cobb-Douglas
          (i.e. loglinear) functional form. But I see in this online paper that
          he says that linear is superior to loglinear, using, evidently,
          MacKinnon, White, and Davidson's test.

          My reaction would've been to use the Box-Cox transformation, which is
          specifically designed to find a best-fitting functional form from an
          entire spectrum of possible functional forms (including the linear
          and loglinear).

          But, the MacKinnon, White, and Davidson paper is somewhat recent (1983),
          whereas the Box-Cox transformation dates back to 1964...I seem to remember
          seeing an article several years ago about possible problems with the
          Box-Cox transformation (lack of consistent estimates when the null
          hypothesis is not satisfied), so perhaps techniques such as MacKinnon,
          White, and Davidson's have superseded it.


          --MKT
        • Dean Oliver
          One of the questions that Dave raises is this: Why do all the linear weights techniques subtract off missed field goals rather than all field goal attempts?
          Message 4 of 24 , Feb 7, 2004
          • 0 Attachment
            One of the questions that Dave raises is this:

            Why do all the linear weights techniques subtract off missed field
            goals rather than all field goal attempts? Isn't any field goal
            attempt a cost?

            So I'd pose that to some of the people here. Why not subtract FGA
            rather than (FGA - FGM)?

            DeanO
            www.basketballonpaper.com

            --- In APBR_analysis@yahoogroups.com, "Michael Tamada" <tamada@o...>
            wrote:
            > -----Original Message-----
            > From: Dean Oliver [mailto:deano@r...]
            > Sent: Tuesday, February 03, 2004 6:20 PM
            >
            > [...]
            >
            > >It was his recent work, which included a revision on that 1999 paper,
            > >that we talked about a lot recently. Basically, his technique for
            > >identifying the value of stats is a regression of team wins vs team
            > >stats. He originally did it one way that had a bit of collinearity
            >
            > My recollection had been that when his work first came to the attention
            > of this group (I think it was then that we coined the term "laugh test"
            > with respect to his Rodman MVP result), that he had used a Cobb-Douglas
            > (i.e. loglinear) functional form. But I see in this online paper that
            > he says that linear is superior to loglinear, using, evidently,
            > MacKinnon, White, and Davidson's test.
            >
            > My reaction would've been to use the Box-Cox transformation, which is
            > specifically designed to find a best-fitting functional form from an
            > entire spectrum of possible functional forms (including the linear
            > and loglinear).
            >
            > But, the MacKinnon, White, and Davidson paper is somewhat recent (1983),
            > whereas the Box-Cox transformation dates back to 1964...I seem to
            remember
            > seeing an article several years ago about possible problems with the
            > Box-Cox transformation (lack of consistent estimates when the null
            > hypothesis is not satisfied), so perhaps techniques such as MacKinnon,
            > White, and Davidson's have superseded it.
            >
            >
            > --MKT
          • wimpds
            I think he s right. It also has the nice effect of making 2 for 6 on 3 s come out the same as 3 for 6 on 2 s. ... ... paper, ... for ... team
            Message 5 of 24 , Feb 7, 2004
            • 0 Attachment
              I think he's right. It also has the nice effect of making 2 for 6
              on 3's come out the same as 3 for 6 on 2's.



              --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
              wrote:
              > One of the questions that Dave raises is this:
              >
              > Why do all the linear weights techniques subtract off missed field
              > goals rather than all field goal attempts? Isn't any field goal
              > attempt a cost?
              >
              > So I'd pose that to some of the people here. Why not subtract FGA
              > rather than (FGA - FGM)?
              >
              > DeanO
              > www.basketballonpaper.com
              >
              > --- In APBR_analysis@yahoogroups.com, "Michael Tamada"
              <tamada@o...>
              > wrote:
              > > -----Original Message-----
              > > From: Dean Oliver [mailto:deano@r...]
              > > Sent: Tuesday, February 03, 2004 6:20 PM
              > >
              > > [...]
              > >
              > > >It was his recent work, which included a revision on that 1999
              paper,
              > > >that we talked about a lot recently. Basically, his technique
              for
              > > >identifying the value of stats is a regression of team wins vs
              team
              > > >stats. He originally did it one way that had a bit of
              collinearity
              > >
              > > My recollection had been that when his work first came to the
              attention
              > > of this group (I think it was then that we coined the
              term "laugh test"
              > > with respect to his Rodman MVP result), that he had used a Cobb-
              Douglas
              > > (i.e. loglinear) functional form. But I see in this online
              paper that
              > > he says that linear is superior to loglinear, using, evidently,
              > > MacKinnon, White, and Davidson's test.
              > >
              > > My reaction would've been to use the Box-Cox transformation,
              which is
              > > specifically designed to find a best-fitting functional form
              from an
              > > entire spectrum of possible functional forms (including the
              linear
              > > and loglinear).
              > >
              > > But, the MacKinnon, White, and Davidson paper is somewhat recent
              (1983),
              > > whereas the Box-Cox transformation dates back to 1964...I seem to
              > remember
              > > seeing an article several years ago about possible problems with
              the
              > > Box-Cox transformation (lack of consistent estimates when the
              null
              > > hypothesis is not satisfied), so perhaps techniques such as
              MacKinnon,
              > > White, and Davidson's have superseded it.
              > >
              > >
              > > --MKT
            • Michael Tamada
              ... From: Dean Oliver [mailto:deano@rawbw.com] Sent: Saturday, February 07, 2004 10:20 AM ... I conjecture (but can t prove) that it s a question of what level
              Message 6 of 24 , Feb 7, 2004
              • 0 Attachment
                -----Original Message-----
                From: Dean Oliver [mailto:deano@...]
                Sent: Saturday, February 07, 2004 10:20 AM

                >One of the questions that Dave raises is this:
                >
                >Why do all the linear weights techniques subtract off missed field
                >goals rather than all field goal attempts? Isn't any field goal
                >attempt a cost?
                >
                >So I'd pose that to some of the people here. Why not subtract FGA
                >rather than (FGA - FGM)?

                I conjecture (but can't prove) that it's a question of what level of
                detail one wants to track things at. If we decide to track FGA, we're
                noting that they put the team into a certain position, one which has
                a couple of outcomes, namely FGM or FGMiss.

                Or we can skip over that process and simply look at the outcome:
                FGM or FGMiss. I.e. ignore the FGA.

                I think it's analogous to the "possessions" vs "plays" question that
                we periodically discuss here. Or equivalently, to this question:

                Do we measure basketball teams' qualities in two areas, offense and
                defense?

                Or do we measure them in three areas: offense, defense, and
                rebounding?

                If we look as "possession-based" stats, then the rebounds get folded
                into the offense and defense measures. If we look at "play-based"
                stats, then rebounding is a skill separate from offense and defense.


                Neither technique is "correct" or "incorrect", it's a question of
                at what level of detail, or with what "granularity" (to use a term
                which seems to be one of the favorite over-used buzzwords these days)
                we want to study basketball.


                Anyway, back to FGA. I think that's a choice that we can make. Just as
                we can say that offensive rebounding is a part of offense, and defensive
                rebounding a part of defense, we can say that FGA are a part of the FGM
                and FGMiss stats. And hence can be ignored because they're already
                "taken care of" by virtue of the trivial identity FGA == FGM + FGMiss.

                But if we did want to study what happens where a team is before, and after,
                if puts up its FGA, then we wouldn't want to ignore FGA.

                An example from the interesting assist discussion we've been having: maybe
                the value of the good passes (some but not all of which lead to measurable
                assists) is that it permits the recipient to not only put up a FGA, but to
                put one up with has an enhanced % probability of going in. Or it permits
                the shooter to attempt a 3PT FGA, one which he otherwise wouldn't even
                attempt to shoot because he's well-covered (or doesn't even have the ball at
                all, since we're assuming it was a good pass which got the ball to him).

                Those I think are examples where FGA would matter and we wouldn't want to
                sweep them under the rug by parroting the FGA == FGM+FGMiss identity.


                To get to the original question, if we're using linear weights type
                formulas, and have a choice of treating FGA as a negative or FGA-FGMiss
                as a negative, I think that's a case where it wouldn't make any difference.

                An analogy: we can predict students GPAs by looking at their verbal SAT
                and their math SAT, but if we wanted to we could instead use verbal SAT
                and TOTAL SAT. And we'd get the same results, except for a factor to be
                added to the coefficient on verbal SAT.


                Aesthetically, the one problem with treating FGA as a negative as Berri
                seems to want to is that they aren't by themselves a negative per se.
                If they go in (FGM) they're part of a positive process; it's only if they
                miss (FGMiss) that they become a part of a negative process.

                Given the 24-second clock, somebody has to shoot the ball at some point,
                and it seems a bit severe to "call" a FGA a negative thing.


                --MKT
              • Mike G
                ... only if they ... point, ... The term used was cost ; and yes, it costs you something to attempt a shot. It also costs you gas money to drive to work.
                Message 7 of 24 , Feb 8, 2004
                • 0 Attachment
                  --- In APBR_analysis@yahoogroups.com, "Michael Tamada" <tamada@o...>
                  wrote:
                  >... If they go in (FGM) they're part of a positive process; it's
                  only if they
                  > miss (FGMiss) that they become a part of a negative process.
                  >
                  > Given the 24-second clock, somebody has to shoot the ball at some
                  point,
                  > and it seems a bit severe to "call" a FGA a negative thing.


                  The term used was "cost"; and yes, it costs you something to attempt
                  a shot. It also costs you gas money to drive to work.

                  Another term that comes to mind is "investment". Whether it's just
                  the time the ball is in its arc, or over the course of the game, you
                  reap the benefits of your shot's accuracy.

                  And when the clock is running down, almost Any shot is better than
                  no shot. You might get a foul; and there's about 1/3 chance of
                  getting the rebound (or somewhat less on a "forced" shot).

                  It seems any attempt to assign a negative factor to a FGA would
                  depend on the team's offensive-rebounding strength; who's in the
                  lineup; and who's taking the shot; and ultimately, on a case-by-case
                  basis.


                  I guess I don't use 'linear weights', since I don't subtract FGA, or
                  any multiple thereof. Rather, I multiply a scorer's rate by his
                  eff%. Is that the same thing?
                • Dean Oliver
                  ... Here is Dave s rationale. If a shot goes in, it goes back to the opponents. If a shot misses and doesn t get rebounded by the offense, it goes back to
                  Message 8 of 24 , Feb 8, 2004
                  • 0 Attachment
                    --- In APBR_analysis@yahoogroups.com, "Mike G" <msg_53@h...> wrote:
                    > --- In APBR_analysis@yahoogroups.com, "Michael Tamada" <tamada@o...>
                    > wrote:
                    > >... If they go in (FGM) they're part of a positive process; it's
                    > only if they
                    > > miss (FGMiss) that they become a part of a negative process.
                    > >
                    > > Given the 24-second clock, somebody has to shoot the ball at some
                    > point,
                    > > and it seems a bit severe to "call" a FGA a negative thing.
                    >
                    >
                    > The term used was "cost"; and yes, it costs you something to attempt
                    > a shot. It also costs you gas money to drive to work.
                    >
                    > Another term that comes to mind is "investment". Whether it's just
                    > the time the ball is in its arc, or over the course of the game, you
                    > reap the benefits of your shot's accuracy.
                    >
                    > And when the clock is running down, almost Any shot is better than
                    > no shot. You might get a foul; and there's about 1/3 chance of
                    > getting the rebound (or somewhat less on a "forced" shot).
                    >
                    > It seems any attempt to assign a negative factor to a FGA would
                    > depend on the team's offensive-rebounding strength; who's in the
                    > lineup; and who's taking the shot; and ultimately, on a case-by-case
                    > basis.
                    >

                    Here is Dave's rationale. If a shot goes in, it goes back to the
                    opponents. If a shot misses and doesn't get rebounded by the offense,
                    it goes back to the opponents. Given that we track offensive rebounds
                    and can count them separately, the cost of the shot is the same
                    whether it's made or missed. The return on the shot is already
                    reflected in whether points are scored.

                    Subtracting off missed shots rather than total shots does make a
                    difference in how players are evaluated. If you subtract off missed
                    fg and ft, you get a ranking of (using 1 as the coefficients)

                    Rank Player
                    1 Garnett,Kevin
                    2 O'Neal,Shaquille
                    3 Duncan,Tim
                    4 McGrady,Tracy
                    5 nowitzki,dirk
                    6 Bryant,Kobe
                    7 Webber,Chris
                    8 O'Neal,Jermaine
                    9 Sabonis,Arvydas
                    10 brand,elton
                    11 Kidd,Jason
                    12 Malone,Karl
                    13 ming,yao
                    14 gasol,pau
                    15 Bradley,Shawn
                    16 miller,brad
                    17 boozer,carlos
                    18 Pierce,Paul
                    19 Cassell,Sam
                    20 marion,shawn
                    21 Wallace,Ben
                    22 kirilenko,andrei
                    23 Marshall,Donyell
                    24 Nash,Steve
                    25 Stockton,John
                    26 Allen,Ray
                    27 Ilgauskas,Zydrun
                    28 randolph,zach
                    29 Abdur-Rahim,Shar
                    30 swift,stromile

                    If you subtract off just FGA and FTA, not misses, you get the ranking
                    on the right below (with the ranking above on the left):

                    Rank Player Rank Player
                    1 Garnett,Kevin 1 Wallace,Ben
                    2 O'Neal,Shaquille 2 Garnett,Kevin
                    3 Duncan,Tim 3 Bradley,Shawn
                    4 McGrady,Tracy 4 Duncan,Tim
                    5 nowitzki,dirk 5 Foyle,Adonal
                    6 Bryant,Kobe 6 Sabonis,Arvydas
                    7 Webber,Chris 7 Cato,Kelvin
                    8 O'Neal,Jermaine 8 boozer,carlos
                    9 Sabonis,Arvydas 9 Brown,P.J.
                    10 brand,elton 10 O'Neal,Shaquille
                    11 Kidd,Jason 11 Laettner,Christi
                    12 Malone,Karl 12 Grant,Brian
                    13 ming,yao 13 nowitzki,dirk
                    14 gasol,pau 14 Robinson,David
                    15 Bradley,Shawn 15 brand,elton
                    16 miller,brad 16 Stockton,John
                    17 boozer,carlos 17 Marshall,Donyell
                    18 Pierce,Paul 18 Ward,Charlie
                    19 Cassell,Sam 19 clark,keon
                    20 marion,shawn 20 Battie,Tony
                    21 Wallace,Ben 21 Kidd,Jason
                    22 kirilenko,andrei 22 Divac,Vlade
                    23 Marshall,Donyell 23 miller,brad
                    24 Nash,Steve 24 ming,yao
                    25 Stockton,John 25 marion,shawn
                    26 Allen,Ray 26 lafrentz,raef
                    27 Ilgauskas,Zydrun 27 murphy,troy
                    28 randolph,zach 28 bradley,michael
                    29 Abdur-Rahim,Shar 29 Weatherspoon,Cla
                    30 swift,stromile 30 Webber,Chris

                    Basically, guards get killed by using just FGA and FTA.

                    >
                    > I guess I don't use 'linear weights', since I don't subtract FGA, or
                    > any multiple thereof. Rather, I multiply a scorer's rate by his
                    > eff%. Is that the same thing?

                    I did not know this. I guess Kevin P also does something like this.
                    The form I've typically seen is (and the form used above)

                    pts + reb + ast + stl + blk - tov - missedFT - missed FG

                    I thought that this was what MikeG and a few others were doing. I
                    know that Bob Bellotti, Dave Heeren, John H, and a bunch of others use
                    a form like this.

                    So you multiply PTS/min by an effective shooting percentage (including
                    free throws)? How does this get combined with reb, ast, etc.?

                    DeanO
                    www.basketballonpaper.com
                  • Kevin Pelton
                    ... I basically start with the formula for true shooting percentage (or whatever you want to call it) -- PTS/(FGA+(.44*FTA)) -- and add the other positives
                    Message 9 of 24 , Feb 8, 2004
                    • 0 Attachment
                      > I did not know this. I guess Kevin P also does something like
                      > this. The form I've typically seen is (and the form used above)
                      >
                      > pts + reb + ast + stl + blk - tov - missedFT - missed FG
                      >
                      > I thought that this was what MikeG and a few others were doing. I
                      > know that Bob Bellotti, Dave Heeren, John H, and a bunch of others
                      > use a form like this.
                      >
                      > So you multiply PTS/min by an effective shooting percentage
                      > including free throws)? How does this get combined with reb, ast,
                      > etc.?

                      I basically start with the formula for true shooting percentage (or
                      whatever you want to call it) -- PTS/(FGA+(.44*FTA)) -- and add the
                      other positives (REB, AST, STL, BLK) to the numerator and
                      other "opportunities" -- minutes and turnovers, plus part of
                      the "positives" to keep non-shooters from being overrated -- to the
                      denominator.

                      I don't think I've ever seen anyone who does it quite the way I do,
                      though MikeG's appears to be in the same spirit, more or less.

                      As far as the FGM/FGA argument, I look at it this way: Would you
                      ever evaluate a player's offensive efficiency by looking just at
                      missed field goals, not at attempts? Of course not. You look at
                      PTS/POS, not PTS/MissPOS. So why should linear weights treat that
                      any differently?

                      (You'd have to use something like 1/2 for the coefficient of FGA and
                      FTA, though, wouldn't you? Isn't that why DeanO's rankings change so
                      much?)

                      How you treat offensive rebounds does affect things. I think DeanO
                      and JohnH both have the value of a missed field goal as something
                      less than one possession to account for offensive rebounds. Since I
                      treat offensive rebounding as a separate category in my non-linear
                      weights work, I don't.
                    • wimpds
                      The problem here is with the coefficient. Any choice is somewhat arbitrary, but using one is pretty bad. You ll get a much more reasonable list if you use a
                      Message 10 of 24 , Feb 8, 2004
                      • 0 Attachment
                        The problem here is with the coefficient. Any choice is somewhat
                        arbitrary, but using one is pretty bad. You'll get a much more
                        reasonable list if you use a coefficient that will roughly
                        approximate the list you get substracting field goal misses.
                        Something like -.5*(fga +.45*fta).


                        --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
                        wrote:
                        > --- In APBR_analysis@yahoogroups.com, "Mike G" <msg_53@h...> wrote:
                        > > --- In APBR_analysis@yahoogroups.com, "Michael Tamada"
                        <tamada@o...>
                        > > wrote:
                        > > >... If they go in (FGM) they're part of a positive process;
                        it's
                        > > only if they
                        > > > miss (FGMiss) that they become a part of a negative process.
                        > > >
                        > > > Given the 24-second clock, somebody has to shoot the ball at
                        some
                        > > point,
                        > > > and it seems a bit severe to "call" a FGA a negative thing.
                        > >
                        > >
                        > > The term used was "cost"; and yes, it costs you something to
                        attempt
                        > > a shot. It also costs you gas money to drive to work.
                        > >
                        > > Another term that comes to mind is "investment". Whether it's
                        just
                        > > the time the ball is in its arc, or over the course of the game,
                        you
                        > > reap the benefits of your shot's accuracy.
                        > >
                        > > And when the clock is running down, almost Any shot is better
                        than
                        > > no shot. You might get a foul; and there's about 1/3 chance of
                        > > getting the rebound (or somewhat less on a "forced" shot).
                        > >
                        > > It seems any attempt to assign a negative factor to a FGA would
                        > > depend on the team's offensive-rebounding strength; who's in the
                        > > lineup; and who's taking the shot; and ultimately, on a case-by-
                        case
                        > > basis.
                        > >
                        >
                        > Here is Dave's rationale. If a shot goes in, it goes back to the
                        > opponents. If a shot misses and doesn't get rebounded by the
                        offense,
                        > it goes back to the opponents. Given that we track offensive
                        rebounds
                        > and can count them separately, the cost of the shot is the same
                        > whether it's made or missed. The return on the shot is already
                        > reflected in whether points are scored.
                        >
                        > Subtracting off missed shots rather than total shots does make a
                        > difference in how players are evaluated. If you subtract off
                        missed
                        > fg and ft, you get a ranking of (using 1 as the coefficients)
                        >
                        > Rank Player
                        > 1 Garnett,Kevin
                        > 2 O'Neal,Shaquille
                        > 3 Duncan,Tim
                        > 4 McGrady,Tracy
                        > 5 nowitzki,dirk
                        > 6 Bryant,Kobe
                        > 7 Webber,Chris
                        > 8 O'Neal,Jermaine
                        > 9 Sabonis,Arvydas
                        > 10 brand,elton
                        > 11 Kidd,Jason
                        > 12 Malone,Karl
                        > 13 ming,yao
                        > 14 gasol,pau
                        > 15 Bradley,Shawn
                        > 16 miller,brad
                        > 17 boozer,carlos
                        > 18 Pierce,Paul
                        > 19 Cassell,Sam
                        > 20 marion,shawn
                        > 21 Wallace,Ben
                        > 22 kirilenko,andrei
                        > 23 Marshall,Donyell
                        > 24 Nash,Steve
                        > 25 Stockton,John
                        > 26 Allen,Ray
                        > 27 Ilgauskas,Zydrun
                        > 28 randolph,zach
                        > 29 Abdur-Rahim,Shar
                        > 30 swift,stromile
                        >
                        > If you subtract off just FGA and FTA, not misses, you get the
                        ranking
                        > on the right below (with the ranking above on the left):
                        >
                        > Rank Player Rank Player
                        > 1 Garnett,Kevin 1 Wallace,Ben
                        > 2 O'Neal,Shaquille 2 Garnett,Kevin
                        > 3 Duncan,Tim 3 Bradley,Shawn
                        > 4 McGrady,Tracy 4 Duncan,Tim
                        > 5 nowitzki,dirk 5 Foyle,Adonal
                        > 6 Bryant,Kobe 6 Sabonis,Arvydas
                        > 7 Webber,Chris 7 Cato,Kelvin
                        > 8 O'Neal,Jermaine 8 boozer,carlos
                        > 9 Sabonis,Arvydas 9 Brown,P.J.
                        > 10 brand,elton 10 O'Neal,Shaquille
                        > 11 Kidd,Jason 11 Laettner,Christi
                        > 12 Malone,Karl 12 Grant,Brian
                        > 13 ming,yao 13 nowitzki,dirk
                        > 14 gasol,pau 14 Robinson,David
                        > 15 Bradley,Shawn 15 brand,elton
                        > 16 miller,brad 16 Stockton,John
                        > 17 boozer,carlos 17 Marshall,Donyell
                        > 18 Pierce,Paul 18 Ward,Charlie
                        > 19 Cassell,Sam 19 clark,keon
                        > 20 marion,shawn 20 Battie,Tony
                        > 21 Wallace,Ben 21 Kidd,Jason
                        > 22 kirilenko,andrei 22 Divac,Vlade
                        > 23 Marshall,Donyell 23 miller,brad
                        > 24 Nash,Steve 24 ming,yao
                        > 25 Stockton,John 25 marion,shawn
                        > 26 Allen,Ray 26 lafrentz,raef
                        > 27 Ilgauskas,Zydrun 27 murphy,troy
                        > 28 randolph,zach 28 bradley,michael
                        > 29 Abdur-Rahim,Shar 29 Weatherspoon,Cla
                        > 30 swift,stromile 30 Webber,Chris
                        >
                        > Basically, guards get killed by using just FGA and FTA.
                        >
                        > >
                        > > I guess I don't use 'linear weights', since I don't subtract
                        FGA, or
                        > > any multiple thereof. Rather, I multiply a scorer's rate by his
                        > > eff%. Is that the same thing?
                        >
                        > I did not know this. I guess Kevin P also does something like
                        this.
                        > The form I've typically seen is (and the form used above)
                        >
                        > pts + reb + ast + stl + blk - tov - missedFT - missed FG
                        >
                        > I thought that this was what MikeG and a few others were doing. I
                        > know that Bob Bellotti, Dave Heeren, John H, and a bunch of others
                        use
                        > a form like this.
                        >
                        > So you multiply PTS/min by an effective shooting percentage
                        (including
                        > free throws)? How does this get combined with reb, ast, etc.?
                        >
                        > DeanO
                        > www.basketballonpaper.com
                      • Mike G
                        ... ranking ... These rankings are both wack. The one on the right is the same one, apparently, that had Rodman at the top. Now it s Ben, Bradley, and Foyle,
                        Message 11 of 24 , Feb 9, 2004
                        • 0 Attachment
                          --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
                          wrote:
                          >
                          > If you subtract off just FGA and FTA, not misses, you get the
                          ranking
                          > on the right below (with the ranking above on the left):
                          >
                          > Rank Player Rank Player
                          > 1 Garnett,Kevin 1 Wallace,Ben
                          > 2 O'Neal,Shaquille 2 Garnett,Kevin
                          > 3 Duncan,Tim 3 Bradley,Shawn
                          > 4 McGrady,Tracy 4 Duncan,Tim
                          > 5 nowitzki,dirk 5 Foyle,Adonal
                          > 6 Bryant,Kobe 6 Sabonis,Arvydas
                          > 7 Webber,Chris 7 Cato,Kelvin
                          > 8 O'Neal,Jermaine 8 boozer,carlos
                          > 9 Sabonis,Arvydas 9 Brown,P.J.
                          > 10 brand,elton 10 O'Neal,Shaquille
                          > 11 Kidd,Jason 11 Laettner,Christi
                          > 12 Malone,Karl 12 Grant,Brian
                          > 13 ming,yao 13 nowitzki,dirk
                          > 14 gasol,pau 14 Robinson,David
                          > 15 Bradley,Shawn 15 brand,elton
                          > 16 miller,brad 16 Stockton,John
                          > 17 boozer,carlos 17 Marshall,Donyell
                          > 18 Pierce,Paul 18 Ward,Charlie
                          > 19 Cassell,Sam 19 clark,keon
                          > 20 marion,shawn 20 Battie,Tony
                          > 21 Wallace,Ben 21 Kidd,Jason
                          > 22 kirilenko,andrei 22 Divac,Vlade
                          > 23 Marshall,Donyell 23 miller,brad
                          > 24 Nash,Steve 24 ming,yao
                          > 25 Stockton,John 25 marion,shawn
                          > 26 Allen,Ray 26 lafrentz,raef
                          > 27 Ilgauskas,Zydrun 27 murphy,troy
                          > 28 randolph,zach 28 bradley,michael
                          > 29 Abdur-Rahim,Shar 29 Weatherspoon,Cla
                          > 30 swift,stromile 30 Webber,Chris
                          >
                          > Basically, guards get killed by using just FGA and FTA.

                          These rankings are both wack. The one on the right is the same one,
                          apparently, that had Rodman at the top. Now it's Ben, Bradley, and
                          Foyle, over Shaq.

                          I guess, too, it's last year's numbers.

                          The left list isn't as bad, but still has Bradley at #15.



                          > The form I've typically seen is (and the form used above)
                          >
                          > pts + reb + ast + stl + blk - tov - missedFT - missed FG
                          >
                          > I thought that this was what MikeG and a few others were
                          doing. ...> So you multiply PTS/min by an effective shooting
                          percentage (including
                          > free throws)? How does this get combined with reb, ast, etc.?

                          I don't add and subtract totals, I add and subtract "rates", which
                          are tweaked by team scoring/rebounding rates.

                          The adjustment due to eff% is relative to a standard that does not
                          change over time. This is a debatable practice, but my philosophy
                          is that one is responsible for one's shot. So 60% is 60%, and 40%
                          is 40%. Some player's eff% is very context-dependent, while others
                          take their shooting with them when they move.

                          I used to take a player's scoring rate and multiply it by eff%/.527
                          (where .527 is the arbitrary average I started with, in the '80s).
                          But I got tired of seeing the high-% teams being overrated. So now
                          I split the difference.

                          For example, Shareef Abdur-Rahim is shooting .566 eff%. With my
                          older system, I'd have multiplied his 21.5 pts/36 min by .566/.527,
                          to get 23.1 for his scoring rate.

                          Now I multipley 21.5 by ((.566+.527)/2)/.527 , for a 22.3 rate.

                          Basically, I've reduced by half, the effect of a higher- or lower-
                          than-average shooting %.

                          I still don't get team rates that summarize team success. Good
                          defensive teams are still undervalued in my system. Maybe team
                          defense isn't a function of individual ability.
                        • Dean Oliver
                          ... Note that those were per minute Martin Manley credit things. The per minute stuff and variations on weights can change things. I just use these to make a
                          Message 12 of 24 , Feb 9, 2004
                          • 0 Attachment
                            --- In APBR_analysis@yahoogroups.com, "Mike G" <msg_53@h...> wrote:
                            > --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
                            > wrote:
                            > >
                            > > If you subtract off just FGA and FTA, not misses, you get the
                            > ranking
                            > > on the right below (with the ranking above on the left):
                            > >
                            > > Rank Player Rank Player
                            > > 1 Garnett,Kevin 1 Wallace,Ben
                            > > 2 O'Neal,Shaquille 2 Garnett,Kevin
                            >
                            > These rankings are both wack. The one on the right is the same one,
                            > apparently, that had Rodman at the top. Now it's Ben, Bradley, and
                            > Foyle, over Shaq.
                            >

                            Note that those were per minute Martin Manley credit things. The per
                            minute stuff and variations on weights can change things. I just use
                            these to make a point about how different things become if you
                            subtract off FGA rather than FG misses.

                            The one on the right is not the same as the one that had Rodman on
                            top, but it is based on the same principal: that each field goal
                            attempt has the same cost. That's what Dave Berri is saying.

                            >
                            > > The form I've typically seen is (and the form used above)
                            > >
                            > > pts + reb + ast + stl + blk - tov - missedFT - missed FG
                            > >
                            > > I thought that this was what MikeG and a few others were
                            > doing. ...> So you multiply PTS/min by an effective shooting
                            > percentage (including
                            > > free throws)? How does this get combined with reb, ast, etc.?
                            >
                            > I don't add and subtract totals, I add and subtract "rates", which
                            > are tweaked by team scoring/rebounding rates.

                            Per minute rates? Rates of? (See below)

                            >
                            > The adjustment due to eff% is relative to a standard that does not
                            > change over time. This is a debatable practice, but my philosophy
                            > is that one is responsible for one's shot. So 60% is 60%, and 40%
                            > is 40%. Some player's eff% is very context-dependent, while others
                            > take their shooting with them when they move.
                            >
                            > I used to take a player's scoring rate and multiply it by eff%/.527
                            > (where .527 is the arbitrary average I started with, in the '80s).
                            > But I got tired of seeing the high-% teams being overrated. So now
                            > I split the difference.
                            >
                            > For example, Shareef Abdur-Rahim is shooting .566 eff%. With my
                            > older system, I'd have multiplied his 21.5 pts/36 min by .566/.527,
                            > to get 23.1 for his scoring rate.
                            >
                            > Now I multipley 21.5 by ((.566+.527)/2)/.527 , for a 22.3 rate.
                            >

                            Do you do something similar for assists, rebs, etc.?

                            >
                            > I still don't get team rates that summarize team success. Good
                            > defensive teams are still undervalued in my system. Maybe team
                            > defense isn't a function of individual ability.

                            Team defense is definitely something that involves more than
                            individual ability. A coach determines where players go in that D.
                            That coach can do that in such a way that it doesn't take best
                            advantage of player ability. Ludicrous example: Man to man defense
                            with Tim Duncan guarding the opposing point guard and Tony Parker
                            guarding a center. More realistic example: Dallas Mavericks playing
                            straight man D with no help (which they don't do).

                            But, despite the fact that team defense is a function of choices, what
                            we are all measuring is performance -- what players do. The players
                            implement what a coach says. If a measurement is still missing good
                            defensive teams, it's missing part of performance and missing part of
                            what we'd like to predict.

                            DeanO
                            www.basketballonpaper.com
                            "Oliver goes beyond stats to dissect what it takes to win. His breezy
                            style makes for enjoyable reading, but there are plenty of points of
                            wisdom as well. This book can be appreciated by fans, players,
                            coaches and executives, but more importantly it can be used as a text
                            book for all these groups. You are sure to learn something you didn't
                            know about basketball here." Pete Palmer, co-author, Hidden Game of
                            Baseball and Hidden Game of Football
                          • John Hollinger
                            I think this gets to one of the weaknesses of Berri s system, and how he ended up with Rodman as the top player in his original study. This whole time, he s
                            Message 13 of 24 , Feb 11, 2004
                            • 0 Attachment
                              I think this gets to one of the weaknesses of Berri's system, and how
                              he ended up with Rodman as the top player in his original study.

                              This whole time, he's been calculating the MARGINAL value of made
                              shots, but the ABSOLUTE value of rebounds, steals, turnovers and
                              blocks, which has tended to massively overrate big men in general and
                              the Rodman/Wallace types in particular.



                              --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
                              wrote:
                              > One of the questions that Dave raises is this:
                              >
                              > Why do all the linear weights techniques subtract off missed field
                              > goals rather than all field goal attempts? Isn't any field goal
                              > attempt a cost?
                              >
                              > So I'd pose that to some of the people here. Why not subtract FGA
                              > rather than (FGA - FGM)?
                              >
                              > DeanO
                              > www.basketballonpaper.com
                              >
                              > --- In APBR_analysis@yahoogroups.com, "Michael Tamada" <tamada@o...>
                              > wrote:
                              > > -----Original Message-----
                              > > From: Dean Oliver [mailto:deano@r...]
                              > > Sent: Tuesday, February 03, 2004 6:20 PM
                              > >
                              > > [...]
                              > >
                              > > >It was his recent work, which included a revision on that 1999
                              paper,
                              > > >that we talked about a lot recently. Basically, his technique
                              for
                              > > >identifying the value of stats is a regression of team wins vs
                              team
                              > > >stats. He originally did it one way that had a bit of
                              collinearity
                              > >
                              > > My recollection had been that when his work first came to the
                              attention
                              > > of this group (I think it was then that we coined the term "laugh
                              test"
                              > > with respect to his Rodman MVP result), that he had used a Cobb-
                              Douglas
                              > > (i.e. loglinear) functional form. But I see in this online paper
                              that
                              > > he says that linear is superior to loglinear, using, evidently,
                              > > MacKinnon, White, and Davidson's test.
                              > >
                              > > My reaction would've been to use the Box-Cox transformation,
                              which is
                              > > specifically designed to find a best-fitting functional form from
                              an
                              > > entire spectrum of possible functional forms (including the linear
                              > > and loglinear).
                              > >
                              > > But, the MacKinnon, White, and Davidson paper is somewhat recent
                              (1983),
                              > > whereas the Box-Cox transformation dates back to 1964...I seem to
                              > remember
                              > > seeing an article several years ago about possible problems with
                              the
                              > > Box-Cox transformation (lack of consistent estimates when the
                              null
                              > > hypothesis is not satisfied), so perhaps techniques such as
                              MacKinnon,
                              > > White, and Davidson's have superseded it.
                              > >
                              > >
                              > > --MKT
                            • John Hollinger
                              But in reality those two aren t equivalent. I d rather go 2 for 6 on 3s, because I have an extra chance at an offensive rebound. ... recent ... to ... with
                              Message 14 of 24 , Feb 11, 2004
                              • 0 Attachment
                                But in reality those two aren't equivalent. I'd rather go 2 for 6 on
                                3s, because I have an extra chance at an offensive rebound.

                                --- In APBR_analysis@yahoogroups.com, "wimpds" <wimpds@y...> wrote:
                                > I think he's right. It also has the nice effect of making 2 for 6
                                > on 3's come out the same as 3 for 6 on 2's.
                                >
                                >
                                >
                                > --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
                                > wrote:
                                > > One of the questions that Dave raises is this:
                                > >
                                > > Why do all the linear weights techniques subtract off missed field
                                > > goals rather than all field goal attempts? Isn't any field goal
                                > > attempt a cost?
                                > >
                                > > So I'd pose that to some of the people here. Why not subtract FGA
                                > > rather than (FGA - FGM)?
                                > >
                                > > DeanO
                                > > www.basketballonpaper.com
                                > >
                                > > --- In APBR_analysis@yahoogroups.com, "Michael Tamada"
                                > <tamada@o...>
                                > > wrote:
                                > > > -----Original Message-----
                                > > > From: Dean Oliver [mailto:deano@r...]
                                > > > Sent: Tuesday, February 03, 2004 6:20 PM
                                > > >
                                > > > [...]
                                > > >
                                > > > >It was his recent work, which included a revision on that 1999
                                > paper,
                                > > > >that we talked about a lot recently. Basically, his technique
                                > for
                                > > > >identifying the value of stats is a regression of team wins vs
                                > team
                                > > > >stats. He originally did it one way that had a bit of
                                > collinearity
                                > > >
                                > > > My recollection had been that when his work first came to the
                                > attention
                                > > > of this group (I think it was then that we coined the
                                > term "laugh test"
                                > > > with respect to his Rodman MVP result), that he had used a Cobb-
                                > Douglas
                                > > > (i.e. loglinear) functional form. But I see in this online
                                > paper that
                                > > > he says that linear is superior to loglinear, using, evidently,
                                > > > MacKinnon, White, and Davidson's test.
                                > > >
                                > > > My reaction would've been to use the Box-Cox transformation,
                                > which is
                                > > > specifically designed to find a best-fitting functional form
                                > from an
                                > > > entire spectrum of possible functional forms (including the
                                > linear
                                > > > and loglinear).
                                > > >
                                > > > But, the MacKinnon, White, and Davidson paper is somewhat
                                recent
                                > (1983),
                                > > > whereas the Box-Cox transformation dates back to 1964...I seem
                                to
                                > > remember
                                > > > seeing an article several years ago about possible problems
                                with
                                > the
                                > > > Box-Cox transformation (lack of consistent estimates when the
                                > null
                                > > > hypothesis is not satisfied), so perhaps techniques such as
                                > MacKinnon,
                                > > > White, and Davidson's have superseded it.
                                > > >
                                > > >
                                > > > --MKT
                              • wimpds
                                But that just amplifies my point. If you subtract FGmisses then you re giving an advantage to the guy who goes 3-6 on twos. (Of course if you choose a
                                Message 15 of 24 , Feb 11, 2004
                                • 0 Attachment
                                  But that just amplifies my point. If you subtract FGmisses then
                                  you're giving an advantage to the guy who goes 3-6 on twos. (Of
                                  course if you choose a certain coefficient then you're basically
                                  taking scoring out of the equation entirely.)

                                  --- In APBR_analysis@yahoogroups.com, "John Hollinger"
                                  <alleyoop2@y...> wrote:
                                  > But in reality those two aren't equivalent. I'd rather go 2 for 6
                                  on
                                  > 3s, because I have an extra chance at an offensive rebound.
                                  >
                                  > --- In APBR_analysis@yahoogroups.com, "wimpds" <wimpds@y...> wrote:
                                  > > I think he's right. It also has the nice effect of making 2 for
                                  6
                                  > > on 3's come out the same as 3 for 6 on 2's.
                                  > >
                                  > >
                                  > >
                                  > > --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
                                  > > wrote:
                                  > > > One of the questions that Dave raises is this:
                                  > > >
                                  > > > Why do all the linear weights techniques subtract off missed
                                  field
                                  > > > goals rather than all field goal attempts? Isn't any field
                                  goal
                                  > > > attempt a cost?
                                  > > >
                                  > > > So I'd pose that to some of the people here. Why not subtract
                                  FGA
                                  > > > rather than (FGA - FGM)?
                                  > > >
                                  > > > DeanO
                                  > > > www.basketballonpaper.com
                                  > > >
                                  > > > --- In APBR_analysis@yahoogroups.com, "Michael Tamada"
                                  > > <tamada@o...>
                                  > > > wrote:
                                  > > > > -----Original Message-----
                                  > > > > From: Dean Oliver [mailto:deano@r...]
                                  > > > > Sent: Tuesday, February 03, 2004 6:20 PM
                                  > > > >
                                  > > > > [...]
                                  > > > >
                                  > > > > >It was his recent work, which included a revision on that
                                  1999
                                  > > paper,
                                  > > > > >that we talked about a lot recently. Basically, his
                                  technique
                                  > > for
                                  > > > > >identifying the value of stats is a regression of team wins
                                  vs
                                  > > team
                                  > > > > >stats. He originally did it one way that had a bit of
                                  > > collinearity
                                  > > > >
                                  > > > > My recollection had been that when his work first came to
                                  the
                                  > > attention
                                  > > > > of this group (I think it was then that we coined the
                                  > > term "laugh test"
                                  > > > > with respect to his Rodman MVP result), that he had used a
                                  Cobb-
                                  > > Douglas
                                  > > > > (i.e. loglinear) functional form. But I see in this online
                                  > > paper that
                                  > > > > he says that linear is superior to loglinear, using,
                                  evidently,
                                  > > > > MacKinnon, White, and Davidson's test.
                                  > > > >
                                  > > > > My reaction would've been to use the Box-Cox transformation,
                                  > > which is
                                  > > > > specifically designed to find a best-fitting functional form
                                  > > from an
                                  > > > > entire spectrum of possible functional forms (including the
                                  > > linear
                                  > > > > and loglinear).
                                  > > > >
                                  > > > > But, the MacKinnon, White, and Davidson paper is somewhat
                                  > recent
                                  > > (1983),
                                  > > > > whereas the Box-Cox transformation dates back to 1964...I
                                  seem
                                  > to
                                  > > > remember
                                  > > > > seeing an article several years ago about possible problems
                                  > with
                                  > > the
                                  > > > > Box-Cox transformation (lack of consistent estimates when
                                  the
                                  > > null
                                  > > > > hypothesis is not satisfied), so perhaps techniques such as
                                  > > MacKinnon,
                                  > > > > White, and Davidson's have superseded it.
                                  > > > >
                                  > > > >
                                  > > > > --MKT
                                • ptbnl123
                                  I disagree with the way most linear weight equations currently value field goals. In my opinion, a field goal should be worth : FG = Points Scored - Opponent
                                  Message 16 of 24 , Feb 11, 2004
                                  • 0 Attachment
                                    I disagree with the way most linear weight equations currently value
                                    field goals. In my opinion, a field goal should be worth :

                                    FG = Points Scored - Opponent Pts Per Possession

                                    Basically, with any basket a team makes, the team is getting points
                                    and a turnover. A turnover is worth the "Opponent Pts Per
                                    Possession" and that is subtracted from the points scored by the
                                    field goal to get the overall value of the made basket.

                                    So if we take the "Opponent Pts Per Possession" to be 1, then a 2ptr
                                    is worth 1 and a 3ptr is worth 2.

                                    As an illustration, we can look at the effect of making 2/6 three
                                    pointers versus 3/6 two pointers. First, let us give the value of a
                                    FG Miss as follows:

                                    FG Miss = - Opponent Rebound% x Opponent Pts Per Possession

                                    And since the average defensive rebound percentage is roughly 66%,
                                    I'll take a FG Miss to be worth -2/3.

                                    Then, making 2/6 treys would give us:
                                    2 + 2 - 2/3 - 2/3 - 2/3 - 2/3 = 1.33
                                    Making 3/6 2ptrs would give us:
                                    1 + 1 + 1 - 2/3 - 2/3 - 2/3 = 1

                                    The 2/6 treys are worth more because, as mentioned before, they give
                                    us an extra offensive rebounding opportunity which is worth about
                                    1/3 of a point.

                                    Currently, linear weight equations have the form of PTS-FGM... but
                                    taking into account the above, the equations would have the form of
                                    PTS-FGA+FGM... where, strangely, we would add some value for each
                                    miss, while subtracting some value for each attempt (i.e. 1*PTS -
                                    1*FGA + 1/3*FGM)

                                    -Jim Tang


                                    --- In APBR_analysis@yahoogroups.com, "wimpds" <wimpds@y...> wrote:
                                    > But that just amplifies my point. If you subtract FGmisses then
                                    > you're giving an advantage to the guy who goes 3-6 on twos. (Of
                                    > course if you choose a certain coefficient then you're basically
                                    > taking scoring out of the equation entirely.)
                                    >
                                    > --- In APBR_analysis@yahoogroups.com, "John Hollinger"
                                    > <alleyoop2@y...> wrote:
                                    > > But in reality those two aren't equivalent. I'd rather go 2 for
                                    6
                                    > on
                                    > > 3s, because I have an extra chance at an offensive rebound.
                                    > >
                                  • Charlie Board
                                    ... And if were playing you *I d* rather you go 2 for 6 on 3 s, becaue I have a 2 out of 3 chance of getting that extra rebound instead of you..
                                    Message 17 of 24 , Feb 11, 2004
                                    • 0 Attachment
                                      --- In APBR_analysis@yahoogroups.com, "John Hollinger"
                                      <alleyoop2@y...> wrote:
                                      > But in reality those two aren't equivalent. I'd rather go 2 for 6 on
                                      > 3s, because I have an extra chance at an offensive rebound.

                                      And if were playing you *I'd* rather you go 2 for 6 on 3's,
                                      becaue I have a 2 out of 3 chance of getting that extra
                                      rebound instead of you..

                                      :-)


                                      >
                                      > --- In APBR_analysis@yahoogroups.com, "wimpds" <wimpds@y...> wrote:
                                      > > I think he's right. It also has the nice effect of making 2 for 6
                                      > > on 3's come out the same as 3 for 6 on 2's.
                                      > >
                                      > >
                                      > >
                                      > > --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
                                      > > wrote:
                                      > > > One of the questions that Dave raises is this:
                                      > > >
                                      > > > Why do all the linear weights techniques subtract off missed field
                                      > > > goals rather than all field goal attempts? Isn't any field goal
                                      > > > attempt a cost?
                                      > > >
                                      > > > So I'd pose that to some of the people here. Why not subtract FGA
                                      > > > rather than (FGA - FGM)?
                                      > > >
                                      > > > DeanO
                                      > > > www.basketballonpaper.com
                                      > > >
                                      > > > --- In APBR_analysis@yahoogroups.com, "Michael Tamada"
                                      > > <tamada@o...>
                                      > > > wrote:
                                      > > > > -----Original Message-----
                                      > > > > From: Dean Oliver [mailto:deano@r...]
                                      > > > > Sent: Tuesday, February 03, 2004 6:20 PM
                                      > > > >
                                      > > > > [...]
                                      > > > >
                                      > > > > >It was his recent work, which included a revision on that 1999
                                      > > paper,
                                      > > > > >that we talked about a lot recently. Basically, his technique
                                      > > for
                                      > > > > >identifying the value of stats is a regression of team wins vs
                                      > > team
                                      > > > > >stats. He originally did it one way that had a bit of
                                      > > collinearity
                                      > > > >
                                      > > > > My recollection had been that when his work first came to the
                                      > > attention
                                      > > > > of this group (I think it was then that we coined the
                                      > > term "laugh test"
                                      > > > > with respect to his Rodman MVP result), that he had used a Cobb-
                                      > > Douglas
                                      > > > > (i.e. loglinear) functional form. But I see in this online
                                      > > paper that
                                      > > > > he says that linear is superior to loglinear, using, evidently,
                                      > > > > MacKinnon, White, and Davidson's test.
                                      > > > >
                                      > > > > My reaction would've been to use the Box-Cox transformation,
                                      > > which is
                                      > > > > specifically designed to find a best-fitting functional form
                                      > > from an
                                      > > > > entire spectrum of possible functional forms (including the
                                      > > linear
                                      > > > > and loglinear).
                                      > > > >
                                      > > > > But, the MacKinnon, White, and Davidson paper is somewhat
                                      > recent
                                      > > (1983),
                                      > > > > whereas the Box-Cox transformation dates back to 1964...I seem
                                      > to
                                      > > > remember
                                      > > > > seeing an article several years ago about possible problems
                                      > with
                                      > > the
                                      > > > > Box-Cox transformation (lack of consistent estimates when the
                                      > > null
                                      > > > > hypothesis is not satisfied), so perhaps techniques such as
                                      > > MacKinnon,
                                      > > > > White, and Davidson's have superseded it.
                                      > > > >
                                      > > > >
                                      > > > > --MKT
                                    • Michael Tamada
                                      I see the smiley, but disregarding it ... The problem with this reasoning is that I d rather have you have a 2 out of 3 chance of getting the rebound than have
                                      Message 18 of 24 , Feb 11, 2004
                                      • 0 Attachment
                                        I see the smiley, but disregarding it ...

                                        The problem with this reasoning is that I'd rather have you
                                        have a 2 out of 3 chance of getting the rebound than have
                                        a *100%* chance of getting the ball -- which is what happens
                                        every time I make a field goal. If I go 2-6 on 3-pointers
                                        I've given you the ball only twice; if I go 3-6 on 2-pointers
                                        I've given you the ball three times.

                                        Jim Tang's recent posting is exactly right: each made field
                                        goal is accompanied by the equivalent of a turnover. Because
                                        the opponent gets the ball, guaranteed. Missed field goals
                                        do not give the opponent the ball; he has to fight for it,
                                        with only a 2/3 chance of winning.

                                        (Of course the problem with missed FGs is that they give you
                                        0 points, whereas made FGs give you 2 or 3 points.)


                                        --MKT



                                        -----Original Message-----
                                        From: Charlie Board [mailto:cboard@...]
                                        Sent: Wednesday, February 11, 2004 10:59 AM


                                        --- In APBR_analysis@yahoogroups.com, "John Hollinger"
                                        <alleyoop2@y...> wrote:
                                        > But in reality those two aren't equivalent. I'd rather go 2 for 6 on
                                        > 3s, because I have an extra chance at an offensive rebound.

                                        And if were playing you *I'd* rather you go 2 for 6 on 3's,
                                        becaue I have a 2 out of 3 chance of getting that extra
                                        rebound instead of you..

                                        :-)


                                        >
                                        > --- In APBR_analysis@yahoogroups.com, "wimpds" <wimpds@y...> wrote:
                                        > > I think he's right. It also has the nice effect of making 2 for 6
                                        > > on 3's come out the same as 3 for 6 on 2's.
                                        > >
                                        > >
                                        > >
                                        > > --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
                                        > > wrote:
                                        > > > One of the questions that Dave raises is this:
                                        > > >
                                        > > > Why do all the linear weights techniques subtract off missed field
                                        > > > goals rather than all field goal attempts? Isn't any field goal
                                        > > > attempt a cost?
                                        > > >
                                        > > > So I'd pose that to some of the people here. Why not subtract FGA
                                        > > > rather than (FGA - FGM)?
                                        > > >
                                        > > > DeanO
                                        > > > www.basketballonpaper.com
                                        > > >
                                        > > > --- In APBR_analysis@yahoogroups.com, "Michael Tamada"
                                        > > <tamada@o...>
                                        > > > wrote:
                                        > > > > -----Original Message-----
                                        > > > > From: Dean Oliver [mailto:deano@r...]
                                        > > > > Sent: Tuesday, February 03, 2004 6:20 PM
                                        > > > >
                                        > > > > [...]
                                        > > > >
                                        > > > > >It was his recent work, which included a revision on that 1999
                                        > > paper,
                                        > > > > >that we talked about a lot recently. Basically, his technique
                                        > > for
                                        > > > > >identifying the value of stats is a regression of team wins vs
                                        > > team
                                        > > > > >stats. He originally did it one way that had a bit of
                                        > > collinearity
                                        > > > >
                                        > > > > My recollection had been that when his work first came to the
                                        > > attention
                                        > > > > of this group (I think it was then that we coined the
                                        > > term "laugh test"
                                        > > > > with respect to his Rodman MVP result), that he had used a Cobb-
                                        > > Douglas
                                        > > > > (i.e. loglinear) functional form. But I see in this online
                                        > > paper that
                                        > > > > he says that linear is superior to loglinear, using, evidently,
                                        > > > > MacKinnon, White, and Davidson's test.
                                        > > > >
                                        > > > > My reaction would've been to use the Box-Cox transformation,
                                        > > which is
                                        > > > > specifically designed to find a best-fitting functional form
                                        > > from an
                                        > > > > entire spectrum of possible functional forms (including the
                                        > > linear
                                        > > > > and loglinear).
                                        > > > >
                                        > > > > But, the MacKinnon, White, and Davidson paper is somewhat
                                        > recent
                                        > > (1983),
                                        > > > > whereas the Box-Cox transformation dates back to 1964...I seem
                                        > to
                                        > > > remember
                                        > > > > seeing an article several years ago about possible problems
                                        > with
                                        > > the
                                        > > > > Box-Cox transformation (lack of consistent estimates when the
                                        > > null
                                        > > > > hypothesis is not satisfied), so perhaps techniques such as
                                        > > MacKinnon,
                                        > > > > White, and Davidson's have superseded it.
                                        > > > >
                                        > > > >
                                        > > > > --MKT





                                        Yahoo! Groups Links
                                      • Kevin Pelton
                                        ... I wouldn t say _exactly_ right. There s no reason opponent points per possession should enter the equation, because the opponents are getting the same
                                        Message 19 of 24 , Feb 11, 2004
                                        • 0 Attachment
                                          > Jim Tang's recent posting is exactly right: each made field
                                          > goal is accompanied by the equivalent of a turnover. Because
                                          > the opponent gets the ball, guaranteed. Missed field goals
                                          > do not give the opponent the ball; he has to fight for it,
                                          > with only a 2/3 chance of winning.

                                          I wouldn't say _exactly_ right. There's no reason opponent points per
                                          possession should enter the equation, because the opponents are
                                          getting the same number of possessions regardless of whether the
                                          offense makes the shot or misses it.

                                          What's being affected is the team's own "minor possessions"
                                          or "plays" -- it's getting extra shots at the basket because it's
                                          rebounding more of its own shots.

                                          So the relevant equeation would be PTS - FGA + (FGA-FGM)*OREB%*ORTG
                                          (based on plays).

                                          The added value of missed shots, above and beyond turnovers, is given
                                          by the team's offensive rebound percentage times its points per
                                          possession.

                                          Of course, this assumes that an offensive rebound "play" is just as
                                          valuable as a normal "play". In fact, most research seems to indicate
                                          that it's more valuable. (Right?)

                                          There also needs to be an adjustment made for missed free throws,
                                          half of which, similarly, have a chance of being rebounded by the
                                          offensive team.
                                        • ptbnl123
                                          There may be other ways to derive the value of a field goal, but by my method, opponents pts per possession (OPPP) is essential. In fact, your own pts per
                                          Message 20 of 24 , Feb 11, 2004
                                          • 0 Attachment
                                            There may be other ways to derive the value of a field goal, but by
                                            my method, "opponents pts per possession" (OPPP) is essential. In
                                            fact, your own pts per possession is not needed at all.

                                            The idea is to measure how each event affects your net points by the
                                            time you regain possession. So if you give up a turnover, you can
                                            expect you to have -1 net points when you get the ball back
                                            (assuming your opponent scores on average 1 pt per poss). If you
                                            miss a shot, you can expect to have -2/3 net points when you get the
                                            ball back (assuming your opponent's defensive rebound pct is 66%).
                                            If you make a 2pt shot, then you can expect have +1 net points when
                                            you get the ball back (since you scored 2pts and your opponent will
                                            score 1 pt on average).

                                            By possession, I mean everything that happens between your opponent
                                            gaining possession and you regaining it. So OPPP already takes into
                                            account factors based on plays or "minor possessions". Of course,
                                            OPPP is also dependent on your own ability on the defensive end and
                                            dependent on the situation (ie following a steal), but the point is
                                            that all that stuff can be reduced to one figure.

                                            It's true that your opponent gets the ball back regardless of
                                            whether your shot is made or missed, and that is why OPPP figures
                                            into each team-level stat.
                                            FG = Pts - OPPP
                                            FGM = -OppReb% x OPPP
                                            REB = OppReb% x OPPP
                                            TO = -OPPP
                                            STL = OPPP

                                            You do not need to know your own pts per poss to figure out the
                                            effect each stat has on your net points or "value added". By this
                                            method of accounting, it doesn't matter if it is easier to score
                                            following an offensive rebound because it is only when you actually
                                            score that the value is added.

                                            With free throws, the values are:
                                            reboundable FT = 1 - OPPP
                                            reboundable FTM = -OppReb% x OPPP
                                            non-reboundable FT = 1
                                            non-reboundable FTM = 0

                                            -Jim Tang

                                            --- In APBR_analysis@yahoogroups.com, "Kevin Pelton"
                                            <kpelton08@h...> wrote:
                                            >
                                            > I wouldn't say _exactly_ right. There's no reason opponent points
                                            per
                                            > possession should enter the equation, because the opponents are
                                            > getting the same number of possessions regardless of whether the
                                            > offense makes the shot or misses it.
                                            >
                                            > What's being affected is the team's own "minor possessions"
                                            > or "plays" -- it's getting extra shots at the basket because it's
                                            > rebounding more of its own shots.
                                            >
                                            > So the relevant equeation would be PTS - FGA + (FGA-FGM)*OREB%
                                            *ORTG
                                            > (based on plays).
                                            >
                                            > The added value of missed shots, above and beyond turnovers, is
                                            given
                                            > by the team's offensive rebound percentage times its points per
                                            > possession.
                                            >
                                            > Of course, this assumes that an offensive rebound "play" is just
                                            as
                                            > valuable as a normal "play". In fact, most research seems to
                                            indicate
                                            > that it's more valuable. (Right?)
                                            >
                                            > There also needs to be an adjustment made for missed free throws,
                                            > half of which, similarly, have a chance of being rebounded by the
                                            > offensive team.
                                          • ptbnl123
                                            ... Correction, this should say It s true that your opponent has a chance to get the ball back regardless... To address why OPPP figures into each team
                                            Message 21 of 24 , Feb 11, 2004
                                            • 0 Attachment
                                              --- In APBR_analysis@yahoogroups.com, "ptbnl123" <ptbnl123@y...>
                                              wrote:

                                              >
                                              > It's true that your opponent gets the ball back regardless of
                                              > whether your shot is made or missed, and that is why OPPP figures
                                              > into each team-level stat.

                                              Correction, this should say "It's true that your opponent has a
                                              chance to get the ball back regardless..."

                                              To address "why OPPP figures into each team level stat" despite the
                                              fact that "the opponents are getting the same number of possessions
                                              regardless of whether the offense makes the shot or misses it", I
                                              think it's an issue of quality vs quantity. The quantity of opponent
                                              possessions may be the same but the quality of them (OPPP) is what
                                              affects the value of the stats.

                                              -JT
                                            • monepeterson
                                              You can go in circles thinking about this stuff. Giving up possession is an inevitability, so I don t see how you can minus it for a missed field goal but not
                                              Message 22 of 24 , Feb 11, 2004
                                              • 0 Attachment
                                                You can go in circles thinking about this stuff.

                                                Giving up possession is an inevitability, so I don't see how you can
                                                minus it for a missed field goal but not a made one. In fact, I don't
                                                see how you can minus it at all, since you'll give up possession even
                                                if you DON'T shoot.

                                                Maybe you can average what a player does with a possession in
                                                comparison to points per MINOR possession.

                                                So the formula I guess would be points scored (whether the field goal
                                                was missed or made) minus points per minus possessions. An offensive
                                                rebound generates another minor possession. A turnover ends a major
                                                possession.

                                                Just an idea. I don't even know where to start with free throws,
                                                blocks and assists.

                                                Moné
                                              • Mike G
                                                ... on ... In a pickup game, you might do well by this strategy. But in the NBA, 6 2-pt attempts are typically accompanied by 2+ FTA -- from which, almost 2
                                                Message 23 of 24 , Feb 12, 2004
                                                • 0 Attachment
                                                  --- In APBR_analysis@yahoogroups.com, "John Hollinger"
                                                  <alleyoop2@y...> wrote:
                                                  > But in reality those two aren't equivalent. I'd rather go 2 for 6
                                                  on
                                                  > 3s, because I have an extra chance at an offensive rebound.

                                                  In a pickup game, you might do well by this strategy. But in the
                                                  NBA, 6 2-pt attempts are typically accompanied by 2+ FTA -- from
                                                  which, almost 2 more points will result.

                                                  But we've gone into rather unreal hypothetical situations anyway.
                                                  If you rebound your missed 3 right under the basket, do you pass it
                                                  back out for your next 3, rather than following it in?

                                                  The 3-point shot and the drive to the basket will always be co-
                                                  dependent, in their effectiveness. The threat of one boosts the
                                                  effectiveness of the other -- even when one may appear to be more
                                                  effective.
                                                • Mike G
                                                  ... assists, rebs, etc.? Rebound rate is a proportion of the team rebounds allowed, i.e., opponent rebounds. This differs from % of total available rebounds
                                                  Message 24 of 24 , Feb 13, 2004
                                                  • 0 Attachment
                                                    --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
                                                    wrote:
                                                    > > .. I add and subtract "rates", which
                                                    > > are tweaked by team scoring/rebounding rates.
                                                    >
                                                    > Per minute rates? Rates of? ...> Do you do something similar for
                                                    assists, rebs, etc.?

                                                    Rebound "rate" is a proportion of the team rebounds allowed, i.e.,
                                                    opponent rebounds. This differs from % of total available rebounds
                                                    (which is how I used to do it), but it seems to agree better with
                                                    cases of players moving from better- to worser-rebounding teams (and
                                                    vise versa).

                                                    The theory is that on good rebounding teams, you will be competing
                                                    against your own teammates more than if you are the only good
                                                    rebounder out there for your squad. I haven't checked the viability
                                                    of this idea, but it does seem to give better team ratings.
                                                    >
                                                    > >
                                                    > > I still don't get team rates that summarize team success. Good
                                                    > > defensive teams are still undervalued in my system. Maybe team
                                                    > > defense isn't a function of individual ability.
                                                    >
                                                    > despite the fact that team defense is a function of choices, what
                                                    > we are all measuring is performance -- what players do. The
                                                    players
                                                    > implement what a coach says. If a measurement is still missing
                                                    good
                                                    > defensive teams, it's missing part of performance and missing part
                                                    of
                                                    > what we'd like to predict.

                                                    I do agree. But do players really "improve" when they join a better-
                                                    coached team? Does "chemistry" change how good a player is for the
                                                    moment, but make for bad predictions if their environment changes?

                                                    If I can get a good estimate of a players "rebounding ability", that
                                                    is pretty much transferable to another context. Scoring, likewise.
                                                    Defense is more context-dependent.

                                                    So "performance" might be defined as "contributing to wins"; or it
                                                    might be defined as "being productive".

                                                    I guess I think some coaches get more out of the same talent; and
                                                    some player-combinations just "work better" than others. These are
                                                    interrelated, and don't necessarily define the "worth" of a player.
                                                  Your message has been successfully submitted and would be delivered to recipients shortly.