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

Value of Assists

Expand Messages
  • nick@lab.net
    I just joined this group so apologies if the following throughts are old news. ... I ve been trying to think of a good way to measure the value of an assist
    Message 1 of 19 , Jan 2, 2004
    • 0 Attachment
      I just joined this group so apologies if the following throughts are old news.

      > From: "Kevin Pelton" <kpelton08@...>
      > The value of assists? To get a legitimate answer would probably
      > require at least a season-long or league-wide study where volunteers
      > tracked whether unmade shots would have been credited an assist or
      > not. Ask DeanO how difficult that would be.

      I've been trying to think of a good way to measure the value of an assist
      since John Hollinger's first prospectus came out.

      Here's my current thinking on the subject.

      To get a perfectly accurate number we would need two stats that aren't
      currently recorded -- times in which a pass is made that would result
      an an assist if a shot was made (potential assists) in which the player
      receiving the pass gets fouled and potential assists in which the
      player receiving the pass misses the shot.

      The latter is of crucial importance. Imagine a jump shooter who can shoot
      35% if they have to create off the dribble, and 65% if they can catch and
      shoot. Then imagine that the team runs 40 plays for the shooter, 20 plays
      in which he has to create off the dribble (no potential assist), and 20
      in which he receives a pass (potential assist). At the end of that sequence
      he has shot 20-40 (40 pts) and the pg has racked up 13 assists. If there
      were no pg to pass the ball he would shoot 14-40 (28 pts). In this
      case the marginal value of an assist is clearly (40-28)/13 or ~ 1 pt.

      If we were to just look at the first situation (20-40, 13 assists) and
      say "he went 7-27 on shots that were not assisted" we would assume that
      his FG% would be 26% without a pg to pass the ball (and 100% when
      generating an assist) and would set the value of an assist at 1.48 pts. To
      get the true value of an assist we need to understand that not every
      potential assist results in a basket.

      Lacking any way to compute those percentages directly (if we had the two
      stats I mentioned we could compute PSA for situations involving a potential
      assist, and PSA for situations not involving an assist) we have to arrive
      at the numbers indirectly.

      I decided that it made more sense to look at assists as a team stat rather
      than an individual stat. I think this because assists are usually the
      result of an offense running properly, and that's a team effort. I also
      think that we have to move from the individual to the team level to have a
      closed system that includes both the passer and the scorer.

      I took all of the team data from 88-89 - 01-02 and computed possessions,
      assist/possession (app) and points per possession (ppp).

      I then did a simple linear regression on the relationship between app and
      ppp and generated the following values:

      PPP = 1.0733* APP + 0.7794

      This sets the marginal value of an assist at 1.07 (changing one posession
      from 0 to 1 asssits increased the predicted team score by 1.07 points).

      This methodology has a couple problems, some intentional some not. It
      creates the chicken/ egg problem of whether the correlation between assists
      and offense is caused by good offense creating assist oppportunities or
      visa versa. Personally I think this question is a little like the question
      about do coaches succeed because of talented players or visa versa. Yes a
      talented offense makes things easier for players with good passing skills,
      but it also takes special skills to play in that offense.

      You could, if you wanted, phrase the question as "did moving from Cleveland
      to sactown suddenly make Brad Miller a better player?" My methodology
      implies an answer of, "if he is contributing more to the team success now
      than he did in cleveland we have to treat him as if he is a better player."
      Perhaps he's benefiting from the team, perhaps he has a chance to show
      skills that he had all along but couldn't get full benefit from before.

      In addition, I agree with John Hollinger that, intuitively, this seems like
      too high a value to place on assists. But I still think it's interesting to
      look at the numbers this way and it's the best statistical assessment that
      I can think of.

      I'm curious to know what everyone makes of that and if someone wants to run
      the calculation on a larger dataset.

      It would also be interesting to look at the differences between pre-1994-95
      and post 1994-95, as the number of 3-pointers attempted increased
      dramatically that year.

      Comments would be appreciated.

      Nick Scholtz
    • Michael Tamada
      ... From: nick@lab.net [mailto:nick@lab.net] Sent: Friday, January 02, 2004 1:08 PM [...excellent summary of the principles deleted] ... Fascinating, I thought
      Message 2 of 19 , Jan 2, 2004
      • 0 Attachment
        -----Original Message-----
        From: nick@... [mailto:nick@...]
        Sent: Friday, January 02, 2004 1:08 PM

        [...excellent summary of the principles deleted]

        >I took all of the team data from 88-89 - 01-02 and computed possessions,
        >assist/possession (app) and points per possession (ppp).
        >
        >I then did a simple linear regression on the relationship between app and
        >ppp and generated the following values:
        >
        >PPP = 1.0733* APP + 0.7794
        >
        >This sets the marginal value of an assist at 1.07 (changing one posession
        >from 0 to 1 asssits increased the predicted team score by 1.07 points).

        Fascinating, I thought that similar research had been done before and
        found a much smaller coefficient. But I may be mis-remembering. If so,
        this is a good start.

        >This methodology has a couple problems, some intentional some not. It
        >creates the chicken/ egg problem of whether the correlation between assists
        >and offense is caused by good offense creating assist oppportunities or
        >visa versa. Personally I think this question is a little like the question
        >about do coaches succeed because of talented players or visa versa. Yes a
        >talented offense makes things easier for players with good passing skills,
        >but it also takes special skills to play in that offense.

        [...]

        >In addition, I agree with John Hollinger that, intuitively, this seems like
        >too high a value to place on assists. But I still think it's interesting to
        >look at the numbers this way and it's the best statistical assessment that
        >I can think of.

        I agree with both of you -- for an individual basket. I.e. if someone
        makes a 2-pointer, do we want to give half of the credit (1.07 points) to
        the passer? If the pass turned what would've been a 25% FG attempt into a
        75% FG attempt, then yes we would, but do most assists have that large an
        impact? Some do, but many do not.

        But here's another explanation for the large coefficient that you found: it
        may be due to the "missing assists" that you mentioned earlier (good passes
        leading to missed shots, and also good passes leading to FTs -- neither of
        these get recorded as assists).

        I.e. the presence of the good passers and the oodles of assists on the
        high APP teams leads to high PPP. But some of those wonderful passes
        don't lead directly to baskets -- instead they lead to missed FGs or
        to FTAs. And do not get recorded as assists.

        The few good passes which do get recorded as assists will then be "overvalued"
        by the regression, because what looks like a relatively small number of good
        passes is in fact only the tip of the iceberg: there's a number of other
        good passes being made that aren't being recorded.

        The average value of a good pass is thus less than it appears from the
        regression, because the few recorded ones (the assists) are being given
        *all* the credit for the overall improved offense.

        I put "overvalued" in quotation marks because if my explanation is correct,
        then it may in fact be a good approximation to give those assists 1.07 points
        of credit. This may indeed cause that one individual assist to be overvalued,
        but in doing so we hopefully correct for our failure to give the passer any
        credit at all for his other good passes -- the ones that led to missed FGs
        or to FTAs.

        The ideal solution of course would be to record all such assist-worthy
        passes, as I mentioned in my previous comment about DeanO's play-by-play
        notation system. Then we could e.g. give Kidd a +.5 or whatever for
        making a good pass, and Martin a -1.0 or whatever for missing the resulting
        better-than-average-chance shot, instead of the usual -.5 or whatever.


        The other problem though is the one that you mentioned: the possible
        reverse causality of PPP and APP, or perhaps even more likely, that
        plain old "good offense" or "bad defense" or both leads to BOTH PPP and
        APP rising. Which, as you point out, would cause the regression to overvalue
        assists -- and to truly overvalue them, not simply to make up for the
        missing assists. This is an example of what econometricians call the
        simultaneity problem.

        I need to think about this more, but partial help may come from looking
        at individual game scores, instead of season totals (which is what I assume
        you based your regression on). This would permit a couple of things: a
        time-series approach rather than the cross-sectional regression that you
        ran (however, non-stationarity which has been mentioned in the hot hand
        context, is a potential problem here). And also it might permit us to
        bring in additional explanatory variables, to use what econometricians
        call an instrumental variables approach to solve the simultaneity problem.
        I'm thinking in particular that teams' season offensive and defensive
        PPP's might be used as instruments when looking at one individual box score.
        But I need to think this through more.

        >I'm curious to know what everyone makes of that and if someone wants to run
        >the calculation on a larger dataset.

        I think 1989-2002 is in general plenty big; to look at additional seasons
        runs the risk of looking at years in which the NBA was a very different
        animal from what it is now. Your 3-pointer example below is an example
        of the hazards.

        One crude but simple way in which you could correct for a little of
        the simultaneity bias: because you have something close to panel data
        (a cross-section of teams followed across 14 seasons, or equivalently
        several dozen time series of team data) instead of running a univariate
        regression with APP as the only explanatory variables, run a "fixed effects"
        multivariate regression: put in dummy variables (binary variables) for teams
        and years. This will likely reduce the size of the APP coefficient, to
        more believable proportions.

        >It would also be interesting to look at the differences between pre-1994-95
        >and post 1994-95, as the number of 3-pointers attempted increased
        >dramatically that year.


        --MKT
      • Mike G
        ... you found: it ... I try to assign weights that correspond to minutes played. For some guys [Bruce Bowen], there are no stats that you can jack up to
        Message 3 of 19 , Jan 3, 2004
        • 0 Attachment
          --- In APBR_analysis@yahoogroups.com, "Michael Tamada" <tamada@o...>
          wrote:
          > -----Original Message-----
          > From: nick@l... [mailto:nick@l...]
          > Sent: Friday, January 02, 2004 1:08 PM
          > > ...sets the marginal value of an assist at 1.07 ...

          > >... intuitively, this seems like
          > >too high a value to place on assists.

          >
          > ...But here's another explanation for the large coefficient that
          you found: it
          > may be due to the "missing assists" ...

          I try to assign weights that correspond to minutes played. For some
          guys [Bruce Bowen], there are no stats that you can "jack up" to
          justify their minutes. But point guards are responsible for lots of
          things that don't show up -- except for the occasional assist.

          They also bring the ball up the court. How many guys can do that?
          Plenty, but not nearly everyone. Thus they spare the team a lot of
          turnovers they'd otherwise have.

          PG is also the position that gets stuck with launching shotclock-
          beating (low-%) shots. So their shooting % tends to take a beating.

          And then there are the turnovers that are assigned to the passer
          because the would-be receiver wasn't looking.

          So I've gone from 1.25 to 1.33 as a "weight" for assists. This
          seems to give PGs their due in player rankings, commensurate with
          the minutes they play.
        • dlirag@hotmail.com
          ... possessions, ... app and ... posession ... points). ... and ... If so, ... Here s a list of some Usenet threads discussing the possible value of assists:
          Message 4 of 19 , Jan 4, 2004
          • 0 Attachment
            --- In APBR_analysis@yahoogroups.com, "Michael Tamada" <tamada@o...>
            wrote:
            > -----Original Message-----
            > From: nick@l... [mailto:nick@l...]
            > Sent: Friday, January 02, 2004 1:08 PM
            >
            > [...excellent summary of the principles deleted]
            >
            > >I took all of the team data from 88-89 - 01-02 and computed
            possessions,
            > >assist/possession (app) and points per possession (ppp).
            > >
            > >I then did a simple linear regression on the relationship between
            app and
            > >ppp and generated the following values:
            > >
            > >PPP = 1.0733* APP + 0.7794
            > >
            > >This sets the marginal value of an assist at 1.07 (changing one
            posession
            > >from 0 to 1 asssits increased the predicted team score by 1.07
            points).
            >
            > Fascinating, I thought that similar research had been done before
            and
            > found a much smaller coefficient. But I may be mis-remembering.
            If so,
            > this is a good start.

            Here's a list of some Usenet threads discussing the possible value of
            assists:

            http://groups.google.com/groups?as_q=value%20regression&as_oq=assist%
            20assists&safe=images&ie=UTF-8&oe=UTF-
            8&as_ugroup=*basketball*&as_scoring=d&lr=&hl=en
          • Franklin X
            The major problem I have with assists, is that I believe it isn t a very good individual statistic. It relies too much upon what another player does with the
            Message 5 of 19 , Jan 4, 2004
            • 0 Attachment
              The major problem I have with assists, is that I believe it isn't a
              very good individual statistic. It relies too much upon what another
              player does with the ball. Jason Kidd could get 15 assists, or 0
              assists, it all depends on whether his teammates make shots.
              Something Mr Kidd has no control over.

              It is similiar to the save category in baseball, a statistic that
              doesn't truly measure the effectiveness of a player.

              Of course passing skill and creating high percentage opportunities
              for your teammates is important, but assists alone does not measure
              this.

              thoughts?

              vr,

              Xei
            • Kevin Pelton
              ... I don t buy this, and here s why. In this column, http://www.hoopsworld.com/article_5259.shtml I took a look at the stickiness of various statistical
              Message 6 of 19 , Jan 5, 2004
              • 0 Attachment
                --- In APBR_analysis@yahoogroups.com, "Franklin X" <xeifrank@y...>
                wrote:
                > The major problem I have with assists, is that I believe it isn't
                > a very good individual statistic. It relies too much upon what
                > another player does with the ball. Jason Kidd could get 15
                > assists, or 0 assists, it all depends on whether his teammates
                > make shots. Something Mr Kidd has no control over.

                I don't buy this, and here's why.

                In this column,
                http://www.hoopsworld.com/article_5259.shtml

                I took a look at the "stickiness" of various statistical skills,
                using players who played at least 500 minutes in the consecutive-
                season pairs of 00/01-01/02 and 01/02-02/03.

                I didn't use assist rate per se; I actually used a rating I invented
                I call "pass" rating -- assists^2/turnovers*minutes -- that still
                tracks assists pretty well. I found 0.889 correlation between the
                player's pass rating in Year 1 and Year 2, which was lower than the
                correlations for rebound rate and block rate but higher than the
                correlations for "shoot" (which combines free-throw and three-point
                percentages) and two-point percentage.

                If assists were really that context-dependent, I don't think we'd
                see this high of a correlation. Too many players change teams, and
                too many teams change players, each season.

                If you look at an extreme -- say, trading Gary Payton from this
                year's Lakers to last year's Nuggets -- obviously there's going to
                be a significant change in his assists. For most moves, I don't
                think the change is going to be significant. After all, Jason Kidd
                has been amongst the league leaders in assists everywhere he's gone,
                as was Mark Jackson, to name two top passers who have moved around.

                I think a better argument against the evaluative power of assists is
                the different roles players play in different offenses. Did Mike
                Bibby become a worse passer when he went from Vancouver to
                Sacramento and lost three assists off his average? No, he just had
                the ball in his hands less. Vice versa for Brad Miller.
              • Franklin X
                Well, assist totals can be similiar from year to year or team to team, just like the saves category in baseball, but it still doesn t do anything about the
                Message 7 of 19 , Jan 5, 2004
                • 0 Attachment
                  Well, assist totals can be similiar from year to year or team to
                  team, just like the saves category in baseball, but it still doesn't
                  do anything about the fact that an assist is completely out of the
                  control of the person who gets credit for an assist. Basketball has
                  many stats like this, there are too many stats in basketball that are
                  dependent on what other people do. I admire those of you who are
                  able to study and quantify this stuff... :)

                  vr,

                  Xei

                  --- In APBR_analysis@yahoogroups.com, "Kevin Pelton" <kpelton08@h...>
                  wrote:
                  > --- In APBR_analysis@yahoogroups.com, "Franklin X" <xeifrank@y...>
                  > wrote:
                  > > The major problem I have with assists, is that I believe it isn't
                  > > a very good individual statistic. It relies too much upon what
                  > > another player does with the ball. Jason Kidd could get 15
                  > > assists, or 0 assists, it all depends on whether his teammates
                  > > make shots. Something Mr Kidd has no control over.
                  >
                  > I don't buy this, and here's why.
                  >
                  > In this column,
                  > http://www.hoopsworld.com/article_5259.shtml
                  >
                  > I took a look at the "stickiness" of various statistical skills,
                  > using players who played at least 500 minutes in the consecutive-
                  > season pairs of 00/01-01/02 and 01/02-02/03.
                  >
                  > I didn't use assist rate per se; I actually used a rating I
                  invented
                  > I call "pass" rating -- assists^2/turnovers*minutes -- that still
                  > tracks assists pretty well. I found 0.889 correlation between the
                  > player's pass rating in Year 1 and Year 2, which was lower than the
                  > correlations for rebound rate and block rate but higher than the
                  > correlations for "shoot" (which combines free-throw and three-point
                  > percentages) and two-point percentage.
                  >
                  > If assists were really that context-dependent, I don't think we'd
                  > see this high of a correlation. Too many players change teams, and
                  > too many teams change players, each season.
                  >
                  > If you look at an extreme -- say, trading Gary Payton from this
                  > year's Lakers to last year's Nuggets -- obviously there's going to
                  > be a significant change in his assists. For most moves, I don't
                  > think the change is going to be significant. After all, Jason Kidd
                  > has been amongst the league leaders in assists everywhere he's
                  gone,
                  > as was Mark Jackson, to name two top passers who have moved around.
                  >
                  > I think a better argument against the evaluative power of assists
                  is
                  > the different roles players play in different offenses. Did Mike
                  > Bibby become a worse passer when he went from Vancouver to
                  > Sacramento and lost three assists off his average? No, he just had
                  > the ball in his hands less. Vice versa for Brad Miller.
                • Michael Tamada
                  Well, in basketball just about every statistic is dependent upon what a player s teammates do. Even rebounds are harder to get if your teammates fail to block
                  Message 8 of 19 , Jan 5, 2004
                  • 0 Attachment
                    Well, in basketball just about every statistic is dependent
                    upon what a player's teammates do. Even rebounds are harder
                    to get if your teammates fail to block out the opposing
                    rebounders.

                    But we're not about to therefore stop tracking rebounds,
                    nor are we going to stop tracking assists. Of course, the
                    art/science comes in figuring out how to best utilize
                    those raw statistics (convert rebounds into rebounds per
                    minute -- but that fails to correct for FG% and game pace;
                    better is to convert them into rebound percentages; I suspect
                    that rebound odds would be best of all).

                    Assists are trickier than rebounds because they are indeed
                    highly dependent upon the performance of teammates.


                    But your question was already answered in NickS's original
                    posting, plus one of MikeG's replies:

                    1. If the NBA started tracking not just the good passes
                    that led to assists, but also the good passes that led to
                    missed FGAs and the good passes that led to fouls and FTAs,
                    we can then look simply at "good passes" (i.e. assist-worthy
                    passes) without worrying about the contaminating impact of
                    whether the passer's teammate buried the shot or not.

                    2. Unfortunately, we're not in that nice situation. What
                    then, to make of assists? NickS's work starts us on the
                    road to emprically figuring out the average or typical value
                    of an assist (I have to look at the Usenet thread that
                    someone cited). The other approach which is more immediately
                    useful is what MikeG and others have done: recognize that
                    while any one assist is not particularly valuable, what assists
                    probably do is provide possible evidence or proxies for a
                    playmaker's contributions. Hence MikeG pumps up their value
                    because an 8-assist player, compared to a 5-assist player,
                    is likely making contributions to the team above and beyond
                    those inherent in the 3 assisted baskets.

                    The situation is analogous to sabrmetricians trying to measure
                    defense. Putouts, assists, and errors by themselves are not
                    good ways of evaluating defense. Even the more sophisticated
                    techniques such as zone ratings are vulnerable to contamination
                    from the performance of a player's teammates (fleet or strong-
                    armed centerfielders taking fly balls away from their leftfield
                    teammates; slow-releasing second basemen costing shortstops
                    some deserved double-plays; etc.).


                    But if you're looking for some completely teammate-independent,
                    context-independent statistic, no you're not going to find one
                    in basketball. You won't find one in baseball either, but
                    due to the nature of the game many (not all) of the statistics
                    there can be made somewhat context-independent.


                    I'd wager that assist statistics for starting point guards
                    do show more of what KevinP call "stickiness" than save
                    statistics do for closers. In baseball, sabrmetricians
                    have the luxury of throwing out the save statistic entirely
                    and looking instead at runs prevented or DIPS or whatever.
                    In basketball we can't throw out the assist statistic because
                    then we'd have nothing (except for turnovers).

                    But beware of throwing out the save statistic: to simply
                    look at a pitcher's stats (defense independent or otherwise)
                    per batter faced is still failling to correct for an important
                    teammate- or context-dependent factor: the leverage value of
                    the inning. Not all innings or batters faced are created equal.
                    A reliever who faces three batters and mows them down in the
                    7th inning of a 12-0 ballgame has not made the same contribution
                    to victory as one who comes in with runners on the corners
                    in the 9th inning of a 3-2 ballgame. Some sabrmetricians are
                    starting to take these inning-and-score contexts into account,
                    but much work remains to be done.



                    --MKT


                    -----Original Message-----
                    From: Franklin X [mailto:xeifrank@...]
                    Sent: Monday, January 05, 2004 3:31 AM


                    Well, assist totals can be similiar from year to year or team to
                    team, just like the saves category in baseball, but it still doesn't
                    do anything about the fact that an assist is completely out of the
                    control of the person who gets credit for an assist. Basketball has
                    many stats like this, there are too many stats in basketball that are
                    dependent on what other people do. I admire those of you who are
                    able to study and quantify this stuff... :)

                    vr,

                    Xei

                    --- In APBR_analysis@yahoogroups.com, "Kevin Pelton" <kpelton08@h...>
                    wrote:
                    > --- In APBR_analysis@yahoogroups.com, "Franklin X" <xeifrank@y...>
                    > wrote:
                    > > The major problem I have with assists, is that I believe it isn't
                    > > a very good individual statistic. It relies too much upon what
                    > > another player does with the ball. Jason Kidd could get 15
                    > > assists, or 0 assists, it all depends on whether his teammates
                    > > make shots. Something Mr Kidd has no control over.
                    >
                    > I don't buy this, and here's why.
                    >
                    > In this column,
                    > http://www.hoopsworld.com/article_5259.shtml
                    >
                    > I took a look at the "stickiness" of various statistical skills,
                    > using players who played at least 500 minutes in the consecutive-
                    > season pairs of 00/01-01/02 and 01/02-02/03.
                    >
                    > I didn't use assist rate per se; I actually used a rating I
                    invented
                    > I call "pass" rating -- assists^2/turnovers*minutes -- that still
                    > tracks assists pretty well. I found 0.889 correlation between the
                    > player's pass rating in Year 1 and Year 2, which was lower than the
                    > correlations for rebound rate and block rate but higher than the
                    > correlations for "shoot" (which combines free-throw and three-point
                    > percentages) and two-point percentage.
                    >
                    > If assists were really that context-dependent, I don't think we'd
                    > see this high of a correlation. Too many players change teams, and
                    > too many teams change players, each season.
                    >
                    > If you look at an extreme -- say, trading Gary Payton from this
                    > year's Lakers to last year's Nuggets -- obviously there's going to
                    > be a significant change in his assists. For most moves, I don't
                    > think the change is going to be significant. After all, Jason Kidd
                    > has been amongst the league leaders in assists everywhere he's
                    gone,
                    > as was Mark Jackson, to name two top passers who have moved around.
                    >
                    > I think a better argument against the evaluative power of assists
                    is
                    > the different roles players play in different offenses. Did Mike
                    > Bibby become a worse passer when he went from Vancouver to
                    > Sacramento and lost three assists off his average? No, he just had
                    > the ball in his hands less. Vice versa for Brad Miller.




                    Yahoo! Groups Links

                    To visit your group on the web, go to:
                    http://groups.yahoo.com/group/APBR_analysis/

                    To unsubscribe from this group, send an email to:
                    APBR_analysis-unsubscribe@yahoogroups.com

                    Your use of Yahoo! Groups is subject to:
                    http://docs.yahoo.com/info/terms/
                  • Franklin X
                    Michael, thanks for the thoughtful reply. vr, Xeifrank ... doesn t ... has ... are ... ... ... isn t ... the ... point ... and
                    Message 9 of 19 , Jan 5, 2004
                    • 0 Attachment
                      Michael, thanks for the thoughtful reply.


                      vr,

                      Xeifrank

                      --- In APBR_analysis@yahoogroups.com, "Michael Tamada" <tamada@o...>
                      wrote:
                      > Well, in basketball just about every statistic is dependent
                      > upon what a player's teammates do. Even rebounds are harder
                      > to get if your teammates fail to block out the opposing
                      > rebounders.
                      >
                      > But we're not about to therefore stop tracking rebounds,
                      > nor are we going to stop tracking assists. Of course, the
                      > art/science comes in figuring out how to best utilize
                      > those raw statistics (convert rebounds into rebounds per
                      > minute -- but that fails to correct for FG% and game pace;
                      > better is to convert them into rebound percentages; I suspect
                      > that rebound odds would be best of all).
                      >
                      > Assists are trickier than rebounds because they are indeed
                      > highly dependent upon the performance of teammates.
                      >
                      >
                      > But your question was already answered in NickS's original
                      > posting, plus one of MikeG's replies:
                      >
                      > 1. If the NBA started tracking not just the good passes
                      > that led to assists, but also the good passes that led to
                      > missed FGAs and the good passes that led to fouls and FTAs,
                      > we can then look simply at "good passes" (i.e. assist-worthy
                      > passes) without worrying about the contaminating impact of
                      > whether the passer's teammate buried the shot or not.
                      >
                      > 2. Unfortunately, we're not in that nice situation. What
                      > then, to make of assists? NickS's work starts us on the
                      > road to emprically figuring out the average or typical value
                      > of an assist (I have to look at the Usenet thread that
                      > someone cited). The other approach which is more immediately
                      > useful is what MikeG and others have done: recognize that
                      > while any one assist is not particularly valuable, what assists
                      > probably do is provide possible evidence or proxies for a
                      > playmaker's contributions. Hence MikeG pumps up their value
                      > because an 8-assist player, compared to a 5-assist player,
                      > is likely making contributions to the team above and beyond
                      > those inherent in the 3 assisted baskets.
                      >
                      > The situation is analogous to sabrmetricians trying to measure
                      > defense. Putouts, assists, and errors by themselves are not
                      > good ways of evaluating defense. Even the more sophisticated
                      > techniques such as zone ratings are vulnerable to contamination
                      > from the performance of a player's teammates (fleet or strong-
                      > armed centerfielders taking fly balls away from their leftfield
                      > teammates; slow-releasing second basemen costing shortstops
                      > some deserved double-plays; etc.).
                      >
                      >
                      > But if you're looking for some completely teammate-independent,
                      > context-independent statistic, no you're not going to find one
                      > in basketball. You won't find one in baseball either, but
                      > due to the nature of the game many (not all) of the statistics
                      > there can be made somewhat context-independent.
                      >
                      >
                      > I'd wager that assist statistics for starting point guards
                      > do show more of what KevinP call "stickiness" than save
                      > statistics do for closers. In baseball, sabrmetricians
                      > have the luxury of throwing out the save statistic entirely
                      > and looking instead at runs prevented or DIPS or whatever.
                      > In basketball we can't throw out the assist statistic because
                      > then we'd have nothing (except for turnovers).
                      >
                      > But beware of throwing out the save statistic: to simply
                      > look at a pitcher's stats (defense independent or otherwise)
                      > per batter faced is still failling to correct for an important
                      > teammate- or context-dependent factor: the leverage value of
                      > the inning. Not all innings or batters faced are created equal.
                      > A reliever who faces three batters and mows them down in the
                      > 7th inning of a 12-0 ballgame has not made the same contribution
                      > to victory as one who comes in with runners on the corners
                      > in the 9th inning of a 3-2 ballgame. Some sabrmetricians are
                      > starting to take these inning-and-score contexts into account,
                      > but much work remains to be done.
                      >
                      >
                      >
                      > --MKT
                      >
                      >
                      > -----Original Message-----
                      > From: Franklin X [mailto:xeifrank@y...]
                      > Sent: Monday, January 05, 2004 3:31 AM
                      >
                      >
                      > Well, assist totals can be similiar from year to year or team to
                      > team, just like the saves category in baseball, but it still
                      doesn't
                      > do anything about the fact that an assist is completely out of the
                      > control of the person who gets credit for an assist. Basketball
                      has
                      > many stats like this, there are too many stats in basketball that
                      are
                      > dependent on what other people do. I admire those of you who are
                      > able to study and quantify this stuff... :)
                      >
                      > vr,
                      >
                      > Xei
                      >
                      > --- In APBR_analysis@yahoogroups.com, "Kevin Pelton"
                      <kpelton08@h...>
                      > wrote:
                      > > --- In APBR_analysis@yahoogroups.com, "Franklin X"
                      <xeifrank@y...>
                      > > wrote:
                      > > > The major problem I have with assists, is that I believe it
                      isn't
                      > > > a very good individual statistic. It relies too much upon what
                      > > > another player does with the ball. Jason Kidd could get 15
                      > > > assists, or 0 assists, it all depends on whether his teammates
                      > > > make shots. Something Mr Kidd has no control over.
                      > >
                      > > I don't buy this, and here's why.
                      > >
                      > > In this column,
                      > > http://www.hoopsworld.com/article_5259.shtml
                      > >
                      > > I took a look at the "stickiness" of various statistical skills,
                      > > using players who played at least 500 minutes in the consecutive-
                      > > season pairs of 00/01-01/02 and 01/02-02/03.
                      > >
                      > > I didn't use assist rate per se; I actually used a rating I
                      > invented
                      > > I call "pass" rating -- assists^2/turnovers*minutes -- that still
                      > > tracks assists pretty well. I found 0.889 correlation between the
                      > > player's pass rating in Year 1 and Year 2, which was lower than
                      the
                      > > correlations for rebound rate and block rate but higher than the
                      > > correlations for "shoot" (which combines free-throw and three-
                      point
                      > > percentages) and two-point percentage.
                      > >
                      > > If assists were really that context-dependent, I don't think we'd
                      > > see this high of a correlation. Too many players change teams,
                      and
                      > > too many teams change players, each season.
                      > >
                      > > If you look at an extreme -- say, trading Gary Payton from this
                      > > year's Lakers to last year's Nuggets -- obviously there's going
                      to
                      > > be a significant change in his assists. For most moves, I don't
                      > > think the change is going to be significant. After all, Jason
                      Kidd
                      > > has been amongst the league leaders in assists everywhere he's
                      > gone,
                      > > as was Mark Jackson, to name two top passers who have moved
                      around.
                      > >
                      > > I think a better argument against the evaluative power of assists
                      > is
                      > > the different roles players play in different offenses. Did Mike
                      > > Bibby become a worse passer when he went from Vancouver to
                      > > Sacramento and lost three assists off his average? No, he just
                      had
                      > > the ball in his hands less. Vice versa for Brad Miller.
                      >
                      >
                      >
                      >
                      > Yahoo! Groups Links
                      >
                      > To visit your group on the web, go to:
                      > http://groups.yahoo.com/group/APBR_analysis/
                      >
                      > To unsubscribe from this group, send an email to:
                      > APBR_analysis-unsubscribe@yahoogroups.com
                      >
                      > Your use of Yahoo! Groups is subject to:
                      > http://docs.yahoo.com/info/terms/
                    • bchaikin@aol.com
                      But if you re looking for some completely teammate-independent, context-independent statistic, no you re not going to find one in basketball.  you sure about
                      Message 10 of 19 , Jan 5, 2004
                      • 0 Attachment
                        But if you're looking for some completely teammate-independent, context-independent statistic, no you're not going to find one in basketball. 

                        you sure about this? for the sake of arguement, how could you prove/disprove this?...

                        don't you think free throw shooting is pretty much teammate independent and context independent? what about player's FG%s? do you believe they change only because of varying teammates (obviously not)? if not how can you associate a change in FG% due to varying teammates knowing there are other reasons for it changing?

                        as for assists, again i can show you a number of players (PGs for example) whose ast/min are basically the same as they are traded from team to team, or change slightly to the point of a change being random, and i can also show you examples where its different...

                        on the outside this assumption sounds plausible, but how would you show this to be true using numbers?...

                        bob chaikin
                        bchaikin@...














                      • Michael Tamada
                        ... From: bchaikin@aol.com [mailto:bchaikin@aol.com] Sent: Monday, January 05, 2004 11:32 AM To: APBR_analysis@yahoogroups.com Subject: Re: [APBR_analysis] Re:
                        Message 11 of 19 , Jan 5, 2004
                        • 0 Attachment
                          -----Original Message-----
                          From: bchaikin@... [mailto:bchaikin@...]
                          Sent: Monday, January 05, 2004 11:32 AM
                          To: APBR_analysis@yahoogroups.com
                          Subject: Re: [APBR_analysis] Re: Value of Assists

                          But if you're looking for some completely teammate-independent, context-independent statistic, no you're not going to find one in basketball. 

                          you sure about this? for the sake of arguement, how could you prove/disprove this?...

                          don't you think free throw shooting is pretty much teammate independent and context independent? what about
                          [Michael Tamada]   Correct, FT% is the one statistic which is truly context independent.
                           
                           player's FG%s? do you believe they change only because of varying teammates (obviously not)? if not how can you associate a change in FG% due to varying teammates knowing there are other reasons for it changing?

                          [Michael Tamada] What are you asking here?  You correctly infer that I am not claiming that changes of teammates are the only thing which cause FG% to change.  There are a lot of things (context)  which cause players' FG% to change. 
                           
                          So what then is your question?  Wilt Chamberlain shot 51% in 1962 while averaging 50 points per game.  In 1973 he set an NBA record by shooting 73% (but he scored only 13 points per game).
                           
                          Nowhere have I claimed that it's solely Wilt's teammates that caused his FG% to rise. 
                           
                          But what I will claim (and this hardly seems controversial) is that Wilt's context changed.  His role on the team changed.  He shot less, and took largely high-percentage shots in 1973 instead of bearing the brunt of the offensive burden as he did in 1962.  (And yes, having Gail Goodrich and Jerry West around also contributed to the change in his stats, but no, it's not that alone which caused Wilt's stats to change.)
                           
                          His skills probably changed also, from 1962 to 1973.  But I don't think anyone would claim that his inherent 
                          marksmanship rose by 22 percentage points (indeed, his FT percentage  FELL from 61% to 51%).  In terms of overall skills (including youthful athleticism) he probably had better scoring ability in 1962 than he did in 1973.  Yet his 1973 FG% was higher.   Why?  Context:  Team role.  Coach's design of the offense.  And yes, teammates too.
                           

                           
                          as for assists, again i can show you a number of players (PGs for example) whose ast/min are basically the same as they are traded from team to team, or change slightly to the point of a change being random, and i can also show you examples where its different...


                          on the outside this assumption sounds plausible, but how would you show this to be true using numbers?...

                          [Michael Tamada] Again, what are you asking for?  A proof that every single player who experiences a change in teammates must show some large change in assists per game, or scoring per game?  No, not every single teammate change is going to result in some huge statistical change.  
                           
                          My point (or more accurately Xei's initial point) is that NOT EVERY TEAMMATE CHANGE IS GOING TO RESULT  IN ASSISTS PER GAME STAYING THE SAME.  Many will cause a player's assist numbers to change.  The Mike Bibby example that KevinP gave is a good one.  Here's another:  what do you think would happen if one of the better young point guards in the league, one sharing a name with a notorious president, suddenly acquired a backcourt partner who's one of the best playmakers in NBA history.  Might it not be reasonable to expect that the presidential point guard's assists will go down?
                           
                          Norm Nixon averaged 9.0 assists per game in 1979.  In 1980, with Magic Johnson around, he got 7.8, and averaged fewer than 8.0 over the four years that he shared a backcourt with Magic. 
                           
                          Trade Nixon from the Lakers to the Clippers, and presto! his assists went to 11.1 per game.  And the 1984 Clippers made a LOT fewer FGs than the 1983 Lakers did.  His assists were still a pre-Magic-esque 8.8 in 1985.  In 1986 injuries took hold and all his offensive numbers declined, and that was the end of his NBA career except for a brief 1989 comeback.
                           
                           
                          If your question is:  does every change in teammates cause a big change in statistics, the answer is no.
                          If your question is:  does every change in teammates cause no change in statistics, the answer is again no.  See Bibby and Nixon.
                           
                           
                          Let me re-quote one of your passages:
                           
                          as for assists, again i can show you a number of players (PGs for example) whose ast/min are basically the same as they are traded from team to team, or change slightly to the point of a change being random, and i can also show you examples where its different...
                          I agree 100%.   Thus I do not understand what your question is.
                           
                           
                           
                          --MKT
                        • Dean Oliver
                          ... are old news. ... volunteers ... or ... assist ... Couple comments (I m on the road and have little time... again)... 1. In my book, I surveyed the
                          Message 12 of 19 , Jan 5, 2004
                          • 0 Attachment
                            --- In APBR_analysis@yahoogroups.com, nick@l... wrote:
                            > I just joined this group so apologies if the following throughts
                            are old news.
                            >
                            > > From: "Kevin Pelton" <kpelton08@h...>
                            > > The value of assists? To get a legitimate answer would probably
                            > > require at least a season-long or league-wide study where
                            volunteers
                            > > tracked whether unmade shots would have been credited an assist
                            or
                            > > not. Ask DeanO how difficult that would be.
                            >
                            > I've been trying to think of a good way to measure the value of an
                            assist
                            > since John Hollinger's first prospectus came out.
                            >

                            Couple comments (I'm on the road and have little time... again)...

                            1. In my book, I surveyed the different linear weights methods for
                            the weight they assign to assists relative to points. They all did
                            it different ways, some more scientific than others. That range was
                            actually one of the smaller ones for all stats. The average value
                            was about 1. Not that the average means much in terms of truth.

                            2. If you look at team success as a function of assists, assists
                            mean little. Some good teams have needed assists to be good. Others
                            haven't. It is a style thing.

                            Anyway, I don't try to put a constant point value on assists. If
                            you're looking for a rule of thumb. A point isn't bad, I suppose.

                            DeanO


                            > Here's my current thinking on the subject.
                            >
                            > To get a perfectly accurate number we would need two stats that
                            aren't
                            > currently recorded -- times in which a pass is made that would
                            result
                            > an an assist if a shot was made (potential assists) in which the
                            player
                            > receiving the pass gets fouled and potential assists in which the
                            > player receiving the pass misses the shot.
                            >
                            > The latter is of crucial importance. Imagine a jump shooter who can
                            shoot
                            > 35% if they have to create off the dribble, and 65% if they can
                            catch and
                            > shoot. Then imagine that the team runs 40 plays for the shooter, 20
                            plays
                            > in which he has to create off the dribble (no potential assist),
                            and 20
                            > in which he receives a pass (potential assist). At the end of that
                            sequence
                            > he has shot 20-40 (40 pts) and the pg has racked up 13 assists. If
                            there
                            > were no pg to pass the ball he would shoot 14-40 (28 pts). In this
                            > case the marginal value of an assist is clearly (40-28)/13 or ~ 1
                            pt.
                            >
                            > If we were to just look at the first situation (20-40, 13 assists)
                            and
                            > say "he went 7-27 on shots that were not assisted" we would assume
                            that
                            > his FG% would be 26% without a pg to pass the ball (and 100% when
                            > generating an assist) and would set the value of an assist at 1.48
                            pts. To
                            > get the true value of an assist we need to understand that not every
                            > potential assist results in a basket.
                            >
                            > Lacking any way to compute those percentages directly (if we had
                            the two
                            > stats I mentioned we could compute PSA for situations involving a
                            potential
                            > assist, and PSA for situations not involving an assist) we have to
                            arrive
                            > at the numbers indirectly.
                            >
                            > I decided that it made more sense to look at assists as a team stat
                            rather
                            > than an individual stat. I think this because assists are usually
                            the
                            > result of an offense running properly, and that's a team effort. I
                            also
                            > think that we have to move from the individual to the team level to
                            have a
                            > closed system that includes both the passer and the scorer.
                            >
                            > I took all of the team data from 88-89 - 01-02 and computed
                            possessions,
                            > assist/possession (app) and points per possession (ppp).
                            >
                            > I then did a simple linear regression on the relationship between
                            app and
                            > ppp and generated the following values:
                            >
                            > PPP = 1.0733* APP + 0.7794
                            >
                            > This sets the marginal value of an assist at 1.07 (changing one
                            posession
                            > from 0 to 1 asssits increased the predicted team score by 1.07
                            points).
                            >
                            > This methodology has a couple problems, some intentional some not.
                            It
                            > creates the chicken/ egg problem of whether the correlation between
                            assists
                            > and offense is caused by good offense creating assist
                            oppportunities or
                            > visa versa. Personally I think this question is a little like the
                            question
                            > about do coaches succeed because of talented players or visa versa.
                            Yes a
                            > talented offense makes things easier for players with good passing
                            skills,
                            > but it also takes special skills to play in that offense.
                            >
                            > You could, if you wanted, phrase the question as "did moving from
                            Cleveland
                            > to sactown suddenly make Brad Miller a better player?" My
                            methodology
                            > implies an answer of, "if he is contributing more to the team
                            success now
                            > than he did in cleveland we have to treat him as if he is a better
                            player."
                            > Perhaps he's benefiting from the team, perhaps he has a chance to
                            show
                            > skills that he had all along but couldn't get full benefit from
                            before.
                            >
                            > In addition, I agree with John Hollinger that, intuitively, this
                            seems like
                            > too high a value to place on assists. But I still think it's
                            interesting to
                            > look at the numbers this way and it's the best statistical
                            assessment that
                            > I can think of.
                            >
                            > I'm curious to know what everyone makes of that and if someone
                            wants to run
                            > the calculation on a larger dataset.
                            >
                            > It would also be interesting to look at the differences between pre-
                            1994-95
                            > and post 1994-95, as the number of 3-pointers attempted increased
                            > dramatically that year.
                            >
                            > Comments would be appreciated.
                            >
                            > Nick Scholtz
                          • nick_scholtz
                            Thanks to everyone for your thoughtful responses. I m happy that finding this group got me to organize my thoughts on the subject and I m continuing to go
                            Message 13 of 19 , Jan 5, 2004
                            • 0 Attachment
                              Thanks to everyone for your thoughtful responses.

                              I'm happy that finding this group got me to organize my thoughts on
                              the subject and I'm continuing to go through my data (one interesting
                              result: the lockout season had app rates that were far, far below
                              normal).

                              I will say that I am not a statistician (it's been seven years since I
                              last had stats training) so I am hoping that someone with more
                              experience than I could replicate the analysis.

                              Mike T, thanks for your clarity in explaining the fact that the value
                              of an assist is going to represent more than just the value of the
                              single play that results in the assist. Ideally, if my methodology is
                              sound, it should represent the complete real value of passing (as
                              measured by assists) on team offense. That was my hope in chosing to
                              use team stats, and you explained that better than I could have.

                              Kevin P: your examples of Mike Bibby losing assists and Brad Miller
                              gaining assists when being traded to the Kings is exactly why I am
                              interested in team totals rather than individual players. That
                              eliminates distribution of assists as a factor. That was intentional,
                              but it may make it harder to judge the marginal value of assists from
                              a specific position (like PG).


                              Dean Olliver wrote:
                              > 2. If you look at team success as a function of assists, assists
                              > mean little. Some good teams have needed assists to be good.
                              > Others haven't. It is a style thing.

                              Really? I admit I haven't had a chance to read your book yet but this
                              statement seems very suprising to me. As I've said I think an assist
                              almost always means that an offense is functioning well. While an
                              offense can function without generating assists I would be suprised if
                              a good offense would consistantly fail to generate assists.

                              I would be perfectly willing to believe that the number of assists for
                              the team leader in assists doesn't correlate to success but that's
                              different.

                              When I look at my data set (88-89 through 01-02) only 3 of the top 30
                              teams have below average assists/possession. In addition the
                              correlation between app and winning percentage is positive (it looks
                              like an improvement of .001 in assists/possession should generate a
                              .4% increase in win percentage).

                              In general,we know that offensive efficiency correlates to winning. So
                              if assists correlates to offensive efficiecy shouldn't it correlate to
                              winning as well?

                              For the record, the 3 top teams that ranked below average in app are
                              89-90 Trailblazers, 96-97 heat, and 94-95 Spurs. In addition the 88-89
                              Pistons and 99-00 Lakers were essentially average. The only team on
                              the list that suprises me is the Trailblazers but looking at them they
                              got lots of assists from their back-court but very few from anywhere
                              else.

                              Nick Scholtz
                            • bchaikin@aol.com
                              (A) But if you re looking for some completely teammate-independent, context-independent statistic, no you re not going to find one in basketball. (B) you sure
                              Message 14 of 19 , Jan 5, 2004
                              • 0 Attachment
                                (A) But if you're looking for some completely teammate-independent, context-independent statistic, no you're not going to find one in basketball.

                                (B) you sure about this? for the sake of arguement, how could you prove/disprove this?don't you think free throw shooting is pretty much teammate independent and context independent? 

                                (A) Correct, FT% is the one statistic which is truly context independent.

                                (B) what about player's FG%s? do you believe they change only because of varying teammates (obviously not)? if not how can you associate a change in FG% due to varying teammates knowing there are other reasons for it changing?

                                (A) What are you asking here?  You correctly infer that I am not claiming that changes of teammates are the only thing which cause FG% to change.  There are a lot of things (context)  which cause players' FG% to change. 

                                i am not asking anything. the above (top) statement was made, as if to be factual. it is in fact an assumption, and an incorrect one. FT% was just the obvious statistic to counter the assumption. i simply stated that - for the sake of arguement - the above assumption, if you believe it to be true, is hard to prove thru stats analysis. if you could prove it to be true, i'd love to see it...

                                just because there are 5 players on a team at one time during a game doesn't mean the variability in the stats is such that there aren't patterns of consistency to be found....

                                (A) So what then is your question? 

                                that if someone indeed believes the above assuption to be true to show how it is true. i can show you many examples of players who change teams (i.e. teammates) but who in fact play pretty much the same - same FG%, same assists per game or per minute, same touches/min, etc. in other words their compilation of statistics appears to be independent of the specific teammates. on the other hand i can also show you players who change dramatically with new or different teammates....

                                (A) Wilt Chamberlain shot 51% in 1962 while averaging 50 points per game.  In 1973 he set an NBA record by shooting 73% (but he scored only 13 points per game).

                                true - some players increase their FG% with decreased FGAs. yet i can easily give you innumerable examples of players who shot less FGA from one year to another and had a worse FG%, and some who shoot the same or close to the same FG% regardless of the number of FGA. this is precisely the point - at times a stat can appear to be dependent, at other times independent...

                                (A) Nowhere have I claimed that it's solely Wilt's teammates that caused his FG% to rise. 

                                a statement was made that "...if you're looking for some completely teammate independent statistic...you're not going to find one in basketball...". on the contrary you will find that (a) FT% is pretty much teammate independent and context independent and that (b) many players shoot pretty much the same FG% regardless of their teammates. that would appear to me to show that in their cases the stat of FG% is indeed teammate independent...

                                there are (were) many players who shot pretty much the same FG% season to season with varying teammates or on different teams. this would appear to show that in their cases that stat is, quote "...teammate independent, context independent...". also for many other players this is not the case...

                                what you will find is that many players, regardless of who their teammates are, still shoot the ball from the same areas on the floor on a regular basis, and consequently shoot about the same FG% from season to season. their minutes per game and/or touches per game may change, resulting in varying pts/g or pts/min, but because they are shooting from the same areas their FG% can resultingly be the same or close to the same on a consistent basis...

                                (A) But what I will claim (and this hardly seems controversial) is that Wilt's context changed.  His role on the team changed.  He shot less, and took largely high-percentage shots in 1973 instead of bearing the brunt of the offensive burden as he did in 1962.  (And yes, having Gail Goodrich and Jerry West around also contributed to the change in his stats, but no, it's not that alone which caused Wilt's stats to change.)

                                (A) His skills probably changed also, from 1962 to 1973.  But I don't think anyone would claim that his inherent marksmanship rose by 22 percentage points (indeed, his FT percentage  FELL from 61% to 51%).  In terms of overall skills (including youthful athleticism) he probably had better scoring ability in 1962 than he did in 1973.  Yet his 1973 FG% was higher.   Why?  Context:  Team role.  Coach's design of the offense.  And yes, teammates too.

                                chamberlain played 14 seasons in the nba. antoine carr played 16 seasons, and he had a career FG% of 50.3%. in 89-90 and 90-91 for the sacramento kings he averaged just under 20 pts/g (19.7) shooting 50.3% from the floor playing 31 min/g. he had different teammates than his other years in the league and his role was different - that of a scorer or a primary scorer (in the 2 seasons with SAC he was the 2nd leading scorer and 1st leading scorer on the team). the rest of his 16 year career he shot 50.3% yet only twice averaged over 10 pts/g....

                                i'm sure his role was different in SAC as with his other teams, yet he shot about the same. for carr there was no correlation between an increase or decrease in FG% with an increase or decrease in FGA. what was different is that his touches/min and min/g game increased because of his different teammates - but his FG% remained the same because he basically shot from the same areas on the floor as he had throughtout his career. so in his case (and many other players) FG% was teammate independent...

                                (B) as for assists, again i can show you a number of players (PGs for example) whose ast/min are basically the same as they are traded from team to team, or change slightly to the point of a change being random, and i can also show you examples where its different...

                                (B) on the outside this (top) assumption sounds plausible, but how would you show this to be true using numbers?...

                                (A) Again, what are you asking for?  A proof that every single player who experiences a change in teammates must show some large change in assists per game, or scoring per game?  No, not every single teammate change is going to result in some huge statistical change. 

                                simply asking that if someone believes the top assumption to be true to show it to be true using statistics...

                                (A) My point (or more accurately Xei's initial point) is that NOT EVERY TEAMMATE CHANGE IS GOING TO RESULT  IN ASSISTS PER GAME STAYING THE SAME.  Many will cause a player's assist numbers to change.  The Mike Bibby example that KevinP gave is a good one.  Here's another:  what do you think would happen if one of the better young point guards in the league, one sharing a name with a notorious president, suddenly acquired a backcourt partner who's one of the best playmakers in NBA history.  Might it not be reasonable to expect that the presidential point guard's assists will go down?

                                the fact is is that how often a player shoots, passes, gets fouled, and turns the ball over per ball possession is indeed pretty much - and for the most part - teammate and context independent, contrary to the assumption made. on the other hand how often a player touches the ball on offense is strongly influenced by his teammates. there are many cases of players truly altering their game (conscience efforts to shoot less, pass more or shoot more, pass less) when going to different teams of with new and or different teammates, but - again - for the most part these attributes are teammate and context independent, absolutely the opposite the assumption made....

                                so in pro basketball there are a number of stats that are for the most part teammate and context independent....

                                (A) Norm Nixon averaged 9.0 assists per game in 1979.  In 1980, with Magic Johnson around, he got 7.8, and averaged fewer than 8.0 over the four years that he shared a backcourt with Magic. 

                                (A) Trade Nixon from the Lakers to the Clippers, and presto! his assists went to 11.1 per game.  And the 1984 Clippers made a LOT fewer FGs than the 1983 Lakers did.  His assists were still a pre-Magic-esque 8.8 in 1985.  In 1986 injuries took hold and all his offensive numbers declined, and that was the end of his NBA career except for a brief 1989 comeback.

                                again this is just one example - i can show you players who regardless of their teammates have pretty much the same ast/min during the majority of their careers. that would tell me that for those players assists are pretty much teammate and context independent...

                                (A) If your question is:  does every change in teammates cause a big change in statistics, the answer is no.
                                If your question is:  does every change in teammates cause no change in statistics, the answer is again no.  See Bibby and Nixon.

                                (A) Let me re-quote one of your passages:

                                (B) as for assists, again i can show you a number of players (PGs for example) whose ast/min are basically the same as they are traded from team to team, or change slightly to the point of a change being random, and i can also show you examples where its different...
                                I agree 100%.  

                                (A) Thus I do not understand what your question is.

                                basketball has a number of stats that are - for many players - teammate and context independent, contrary to your assumption that "appeared" to be stated as factual. since your assumption cleary states there are none in basketball, i simply asked for arguement's sake to show how you arrived at that conclusion, and since this is a stats analysis discussion group, to show how with the stats...

                                bob chaikin
                                bchaikin@...







                              • Dean Oliver
                                ... this ... if ... for ... What the book says (as it s better than my memory): There is a correlation between good passing teams and good offensive teams.
                                Message 15 of 19 , Jan 8, 2004
                                • 0 Attachment
                                  --- In APBR_analysis@yahoogroups.com, "nick_scholtz" <nick@l...>
                                  wrote:
                                  >
                                  > Dean Olliver wrote:
                                  > > 2. If you look at team success as a function of assists, assists
                                  > > mean little. Some good teams have needed assists to be good.
                                  > > Others haven't. It is a style thing.
                                  >
                                  > Really? I admit I haven't had a chance to read your book yet but
                                  this
                                  > statement seems very suprising to me. As I've said I think an assist
                                  > almost always means that an offense is functioning well. While an
                                  > offense can function without generating assists I would be suprised
                                  if
                                  > a good offense would consistantly fail to generate assists.
                                  >
                                  > I would be perfectly willing to believe that the number of assists
                                  for
                                  > the team leader in assists doesn't correlate to success but that's
                                  > different.

                                  What the book says (as it's better than my memory):

                                  "There is a correlation between good passing teams and good offensive
                                  teams. Teams that have a higher percentage of their field goals
                                  accompanied by assists tend to be better offensively. It's not an
                                  incredibly strong trend, but only five of the top 25 offenses were
                                  under the league average in this statistic."

                                  Interpret that how you will. My general approach in working with
                                  teams is to put assists pretty low on the list of stats to
                                  follow/prioritize. It has come up, especially when a team has a
                                  single star (it became an issue early this season when Flip Murray
                                  got a little too dominant in the Sonic offense). But most teams
                                  should care more about making shots and letting assists follow,
                                  rather than the other way...


                                  DeanO
                                • nick_scholtz
                                  ... Interesting. I thought that a year or two ago there was a study done on the Sonics that compared their win/loss records in games when they (1) had more
                                  Message 16 of 19 , Jan 8, 2004
                                  • 0 Attachment
                                    --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
                                    wrote:

                                    > What the book says (as it's better than my memory):
                                    >
                                    > "There is a correlation between good passing teams and good
                                    > offensive teams. Teams that have a higher percentage of their field
                                    > goals accompanied by assists tend to be better offensively. It's
                                    > not an incredibly strong trend, but only five of the top 25 offenses
                                    > were under the league average in this statistic."
                                    >
                                    > Interpret that how you will. My general approach in working with
                                    > teams is to put assists pretty low on the list of stats to
                                    > follow/prioritize. It has come up, especially when a team has a
                                    > single star (it became an issue early this season when Flip Murray
                                    > got a little too dominant in the Sonic offense). But most teams
                                    > should care more about making shots and letting assists follow,
                                    > rather than the other way...

                                    Interesting. I thought that a year or two ago there was a study done
                                    on the Sonics that compared their win/loss records in games when they
                                    (1) had more rebounds than their opponent (2) shot more free throws
                                    than their opponent (3) had fewer turn overs than their opponent OR
                                    (4) had more assists than their opponent.

                                    IIRC correctly, assists were by far the best predictor of wins, but
                                    that study only covered one team for a third of a season.

                                    Perhaps this would be a good study for 82games.com (HINT HINT).

                                    BTW, Dean, if you're still on the road, I live in Bellingham WA. I
                                    doubt you'll travel this far North, but please send me an e-mail if
                                    you're going to be in the area.

                                    Nick Scholtz
                                  • Stephen Greenwell
                                    IIRC correctly, assists were by far the best predictor of wins, but that study only covered one team for a third of a season. This is only a casual
                                    Message 17 of 19 , Jan 8, 2004
                                    • 0 Attachment
                                      IIRC correctly, assists were by far the best predictor of wins, but
                                      that study only covered one team for a third of a season.

                                      This is only a casual observation, but shouldn't assists be most directly related to winning?  That is what I would have expected going into the study.  After all, an assist only occurs on a scored basket; you need to score to win.  The teams that score the most will probably have the most assists.  The other stats, such as rebounds, free throw rate, and turnovers, don't relate directly to points.

                                      Steve Greenwell
                                    • Kevin Pelton
                                      ... http://www.sonicscentral.com/kevin7.html Is that the study you re referring to? I imagine you read it. As I explained in my message to you specifically,
                                      Message 18 of 19 , Jan 8, 2004
                                      • 0 Attachment
                                        --- In APBR_analysis@yahoogroups.com, "nick_scholtz" <nick@l...>
                                        wrote:
                                        > --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
                                        > wrote:

                                        > Interesting. I thought that a year or two ago there was a study
                                        > done on the Sonics that compared their win/loss records in games
                                        > when they (1) had more rebounds than their opponent (2) shot more
                                        > free throws than their opponent (3) had fewer turn overs than
                                        > their opponent OR (4) had more assists than their opponent.
                                        >
                                        > IIRC correctly, assists were by far the best predictor of wins, but
                                        > that study only covered one team for a third of a season.

                                        http://www.sonicscentral.com/kevin7.html

                                        Is that the study you're referring to? I imagine you read it. As I
                                        explained in my message to you specifically, again the important
                                        distinction is in looking at pure assists (or assists per
                                        possession) as opposed to assists per field goal made. A team with
                                        more assists is almost always going to have more points, which I
                                        hear correlates very strongly with winning.

                                        Just to throw it out there, here are the correlations this year for
                                        the Sonics (i.e. the number of times the team with the better mark
                                        has won the game):

                                        Assists: .844
                                        Ast/FGM: .719

                                        I was pretty surprised to see the latter be so high. And, just for
                                        the record:

                                        FG%: .813
                                        3PT%: .688
                                        Steals: .656
                                        Rebounding percentage: .594
                                        FT%: .594
                                        Turnovers: .500 (surprising)
                                        Blocks: .500
                                      • Dean Oliver
                                        ... Historically, FG% and Defensive rebounds have been the best indicators in this kind of study (see chp 6). This kind of study is, well, flawed. Defensive
                                        Message 19 of 19 , Jan 9, 2004
                                        • 0 Attachment
                                          --- In APBR_analysis@yahoogroups.com, "Kevin Pelton" <kpelton08@h...>
                                          wrote:
                                          > --- In APBR_analysis@yahoogroups.com, "nick_scholtz" <nick@l...>
                                          > wrote:
                                          > > --- In APBR_analysis@yahoogroups.com, "Dean Oliver" <deano@r...>
                                          > > wrote:
                                          >
                                          > > Interesting. I thought that a year or two ago there was a study
                                          > > done on the Sonics that compared their win/loss records in games
                                          > > when they (1) had more rebounds than their opponent (2) shot more
                                          > > free throws than their opponent (3) had fewer turn overs than
                                          > > their opponent OR (4) had more assists than their opponent.
                                          > >
                                          > > IIRC correctly, assists were by far the best predictor of wins, but
                                          > > that study only covered one team for a third of a season.
                                          >
                                          > http://www.sonicscentral.com/kevin7.html

                                          Historically, FG% and Defensive rebounds have been the best indicators
                                          in this kind of study (see chp 6). This kind of study is, well,
                                          flawed. Defensive rebounds are purely a reflection of fg%. So are
                                          total rebounds, to a lesser degree. There are lots of correlations
                                          between stats and this kind of study, while interesting, doesn't tell
                                          you how to use the info to make teams better (my goal). Kevin points
                                          out that turnovers aren't a huge predictor, which is true in the other
                                          large studies I've looked at. That doesn't mean that you can tell
                                          your team to just go out and commit turnovers. Turnovers matter an
                                          immense amount -- but this kind of study doesn't reflect it.

                                          DeanO
                                          www.basketballonpaper.com
                                        Your message has been successfully submitted and would be delivered to recipients shortly.