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

Re: Pro Basketball Prospectus: Initial Thoughts

Expand Messages
  • John Hollinger
    I guess I feel about linear weights the same way I feel about democracy. It s a horrible system, except when compared to everything else. In all seriousness,
    Message 1 of 28 , Oct 22, 2002
    • 0 Attachment
      I guess I feel about linear weights the same way I feel about
      democracy. It's a horrible system, except when compared to everything
      else.

      In all seriousness, there is a point I would like to clarify, which
      is that in addition to the PER (always handy for the "he's-great-no-
      way-he-sucks" debates), I developed and included -- both in the book
      and the stats on every player -- several more specific metrics for
      evaluating players' performance in individual categories, which I
      think shed a lot of light onto the types of players and where their
      strengths and weaknesses lie.




      --- In APBR_analysis@y..., "Dean Oliver" <deano@r...> wrote:
      > --- In APBR_analysis@y..., "danthestatman2002" <danthestatman@h...>
      > wrote:
      > > they had with their research - but it still seems like ALOT of
      work
      > > to get results wouldn't vary a huge deal from good linear
      weights -
      > > with the extremes looking questionable. I'm sure when I'm done
      > with
      > > the NBA rankings - 90 of my top 100 players would match DeanO's
      top
      > > 100 players. Same with Mike's. Well - we'll see eventually.
      > >
      > > ARGGHHH - still typing.
      > >
      >
      > Rather than trying to out-Tolstoy Dan, I'll make a quick comment
      here
      > on his last statement, err, second-to-last comment. It is
      > interesting how much time we all spend trying to get the greats
      > right. That frankly is almost never the question I get from
      teams.
      > They don't want to know about greats (unless they think Vin Baker
      is
      > great and he's never been). They want to know about the middle
      tier
      > guys. Comparing these guys is where things get dicey. This is why
      > MikeG's ranking of Terry Cummings as a great forward is so
      > interesting and we spent weeks telling him why he's crazy. Most
      > people would not place Cummings among the greats. He was a nice
      > player. Identifying that kind of discrepancy between opinions and
      > then making the right decision can turn a scout/coach/GM into a
      star.
      >
      > There are definite problems. Players can adopt different roles
      when
      > changing systems and predicting that has been hard (I'm getting
      > better at it with some of the individual win-loss records I posted
      > long ago). And, frankly, talent is probably only 50-60% of
      > performance for average to marginal players. Performance reflects
      > probably 80-90% talent in great players for whom systems are always
      > built and for whom the teammates make little difference.
      >
      > So what exactly are your weights? Sounds like a lot of them
      between
      > 0.5 and 1.5, which would be like the averages of most of them. The
      > most controversial ones are offensive rebounds, steals, and
      > turnovers. The weight on OR's can be as high as nearly 4 pts or as
      > low as about 0.6 in systems I've seen. The weight on steals
      between
      > about 0.6 and 2.5. The average almost always ends up around 1
      unless
      > you and JohnH have some funky weights. That's why, if I ever find
      a
      > need for such a formula (mainly to try to check what you jokers are
      > doing), I just use 1 as a weight on everything. I'll try to get
      > permission to post all these things sometime soon. Or maybe I'll
      > only send them to people who ask.
      >
      > > > > Uh oh - wife calling - we gotta run.
      > > >
      > > > Does she approve of all these linear weights you're using?
      > > >
      > > > DeanO
      > >
      > > weights make Dan strong - wifey like strong Dan.
      >
      > Hm. Sorry I asked, I think.
      >
      > DeanO
    • danthestatman2002
      ... everything ... Agreed. Plus, if you want to quantify performance over a long period of time - it makes it simpler. Your book STILL hasn t gotten here.
      Message 2 of 28 , Oct 22, 2002
      • 0 Attachment
        --- In APBR_analysis@y..., "John Hollinger" <alleyoop2@y...> wrote:
        > I guess I feel about linear weights the same way I feel about
        > democracy. It's a horrible system, except when compared to
        everything
        > else.
        >

        Agreed. Plus, if you want to quantify performance over a long period
        of time - it makes it simpler.

        Your book STILL hasn't gotten here. Suck. Soon I hope.
      • Dean Oliver
        ... everything ... In democracy, everyone gets to vote for their preferred candidate. In linear weights, everyone gets to choose their own weights. In
        Message 3 of 28 , Oct 22, 2002
        • 0 Attachment
          --- In APBR_analysis@y..., "John Hollinger" <alleyoop2@y...> wrote:
          > I guess I feel about linear weights the same way I feel about
          > democracy. It's a horrible system, except when compared to
          everything
          > else.

          In democracy, everyone gets to vote for their preferred candidate.

          In linear weights, everyone gets to choose their own weights.

          In democracy, candidates have debates about issues that mean
          something, but they never get resolved because none say anything
          meaningful.

          In linear weights, we have debates about issues underlying the
          weights, but end up with weights that don't say anything meaningful.

          In democracy, the ultimate election winner receives power, money,
          security guards, and a slot in the National Enquirer.

          In linear weights, Dave Heeren, Martin Manley, Gerald Secor Couzens,
          and John Hollinger have little power in pro basketball, earn a modest
          buck, don't fear for their lives, and, unless John goes nuts on us,
          won't end up showing their buttocks at a Dennis Rodman-hosted rave in
          photos from the Enquirer.

          Democratic voting on measures can very easily end up with the non-
          optimal measure when options are not simple. (Old game theory class
          coming in handy for that reference.)

          Linear weights can very easily end up with incorrect selections for
          top players because basketball is not simple.

          I should have stopped with the Dennis Rodman comment, shouldn't I?

          DeanO
        • Mike G
          ... top ... You d like to think so, but I wouldn t bet on it. Dean s system is so different (and to me, puzzling), I only expect to see about 6 of our Top 10s
          Message 4 of 28 , Oct 22, 2002
          • 0 Attachment
            --- In APBR_analysis@y..., "Dean Oliver" <deano@r...> wrote:
            > --- In APBR_analysis@y..., "danthestatman2002" <danthestatman@h...>
            > wrote:
            > > I'm sure when I'm done
            > with
            > > the NBA rankings - 90 of my top 100 players would match DeanO's
            top
            > > 100 players. Same with Mike's.

            You'd like to think so, but I wouldn't bet on it. Dean's system is
            so different (and to me, puzzling), I only expect to see about 6 of
            our Top 10s in agreement, 60 of 100, etc.

            > ...the middle tier
            > guys. Comparing these guys is where things get dicey. This is why
            > MikeG's ranking of Terry Cummings as a great forward is so
            > interesting and we spent weeks telling him why he's crazy.

            I thought you had to be crazy to gain admission here.

            Did I say Cummings was great? Or did I just lay out his stats? Was
            that in this room or in another?


            > .... frankly, talent is probably only 50-60% of
            > performance for average to marginal players. Performance reflects
            > probably 80-90% talent in great players for whom systems are always
            > built and for whom the teammates make little difference.

            Interesting, and timely. If a player like Terry Cummings can carry a
            team like the Clippers, or be a star on an excellent team like the
            Bucks, or the Admiral's right-hand man with a very good Spurs team --
            all this for 10 seasons...
            And then this same player has a debilitating injury, plus a heart
            condition -- but he goes on to another 7 or 8 years as a primo backup
            player, capable of starting any time -- then he may have proven
            something about perseverence and hard work, conditioning, setting an
            example to younger players, and a myriad of positive contributions.

            I guess the Definition of Greatness is also going on in APBR. Why is
            Willis Reed unquestionably great? Because there are more Knicks fans
            around? Why is Cummings laughably ungreat?


            > So what exactly are your weights? ...some funky weights. That's
            why, if I ever find a
            > need for such a formula (mainly to try to check what you jokers are
            > doing), I just use 1 as a weight on everything. I'll try to get
            > permission to post all these things sometime soon. Or maybe I'll
            > only send them to people who ask.

            Dean, if you're going to average everyone's weights, are you
            assigning different weights to the different weights?

            I mean, is my one rebound only worth the same as one rebound of a guy
            blowing in here off the street? I mean, come on...


            MikeG
          • Dean Oliver
            ... ... I don t think so actually. My system doesn t exactly have an ultimate ranking to it, unless you consider career individual win-
            Message 5 of 28 , Oct 23, 2002
            • 0 Attachment
              --- In APBR_analysis@y..., "Mike G" <msg_53@h...> wrote:
              > --- In APBR_analysis@y..., "Dean Oliver" <deano@r...> wrote:
              > > --- In APBR_analysis@y..., "danthestatman2002"
              <danthestatman@h...>
              > > wrote:
              > > > I'm sure when I'm done
              > > with
              > > > the NBA rankings - 90 of my top 100 players would match DeanO's
              > top
              > > > 100 players. Same with Mike's.
              >
              > You'd like to think so, but I wouldn't bet on it. Dean's system is
              > so different (and to me, puzzling), I only expect to see about 6 of
              > our Top 10s in agreement, 60 of 100, etc.
              >

              I don't think so actually. My "system" doesn't exactly have an
              ultimate ranking to it, unless you consider career individual win-
              loss records (or indiv win-loss records per 82 games). In which
              case, my list has pretty much the same people at the top as on your
              list. (I've never compiled a real list since I'm not all that
              interested in all-time great lists, but Jordan pretty much dwarfs
              anyone in the modern day.)

              You may be remembering the game-by-game win-loss records which
              indicate how well players do their role. That list makes many people
              think hard. I'm still not sure what to do with it, though it holds a
              lot of potential power.

              > Did I say Cummings was great? Or did I just lay out his stats?
              Was
              > that in this room or in another?
              >

              Coulda been over in apbr. You have definitely backed him.

              >
              > > .... frankly, talent is probably only 50-60% of
              > > performance for average to marginal players. Performance
              reflects
              > > probably 80-90% talent in great players for whom systems are
              always
              > > built and for whom the teammates make little difference.
              >
              > Interesting, and timely. If a player like Terry Cummings can carry
              a
              > team like the Clippers, or be a star on an excellent team like the
              > Bucks, or the Admiral's right-hand man with a very good Spurs team -
              -
              > all this for 10 seasons...

              I guess that the popular acclaim at the time was that he wasn't
              great. No one at any time in his career considered him great. As
              you have pointed out, there were a lot of good forwards at the time.
              But a great player to me is one that is the best or close to the best
              at least at his position. And Cummings was never that. Your stats
              may say so. Mine definitely do not. He just wasn't a very efficient
              scorer, save a couple years. And he wasn't great defensively either.

              >
              > > So what exactly are your weights? ...some funky weights. That's
              > why, if I ever find a
              > > need for such a formula (mainly to try to check what you jokers
              are
              > > doing), I just use 1 as a weight on everything. I'll try to get
              > > permission to post all these things sometime soon. Or maybe I'll
              > > only send them to people who ask.
              >
              > Dean, if you're going to average everyone's weights, are you
              > assigning different weights to the different weights?
              >
              > I mean, is my one rebound only worth the same as one rebound of a
              guy
              > blowing in here off the street? I mean, come on...

              My recollection is that your weights are 1, though you have all sorts
              of other things in the system to balance post-season, etc.

              The problem is that we're all just blowing in off the street.
              Legitimately, I should only look at the linear weights systems that
              pro and college teams use? Is that what you're suggesting? To them,
              what system they use matters. To us, it augments smack. All of the
              numbers that are in the matrix in the book have been published,
              except for Doug Steele's version, which has been on the web and used
              for long enough that I figure I can post it.

              Trust me, I ain't saying that the average of all values is what's
              right. I am saying the variability in the weights suggests a lot of
              uncertainty in how linear weights get put together. Because of that
              variability and an inability to check any of the methods for
              accuracy, the linear weights model is simply not appropriate. It in
              no way represents the way the game is played. A point is not awarded
              for an assist on top of the points scored. No points are awarded for
              a steal or a block. That doesn't mean they're not valuable. It
              means the linear weights system assigns a surrogate value with no
              relation to the actual game implication of the action.

              My system is different, very different. It may be puzzling to you,
              but it's so different because it _does_ track what happens in a game,
              unlike linear weights. A block isn't a point. It is a partial stop,
              with a defensive rebound (if obtained) being the rest. I don't give
              any points for it. I credit the player with stopping an opponent.
              You could get to the end of a game and realize an average of 1.2
              points per possession were scored by the team and 0.6 were scored by
              the opponent. Then you could kinda sorta get at the net point value
              of that stop and that blocked shot. I make that estimate and the
              value is rarely +1.

              If all you want to do is rank players, feel free to use linear
              weights. Most teams do that, especially as one indicator for
              drafting players (that's why Dan's system will be interesting to see
              played out with college players). It does identify the best players
              pretty well (as I said our systems are pretty equivalent at the top),
              but for the not-so-obvious players, it doesn't lend any insight. It
              can place valuable role players in the same ship as bad players doing
              a lot for bad teams. They are rarely equivalent, but linear weights
              says that they could be.

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