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

Re: [APBR_analysis] Re: Measuring clutch performance

Expand Messages
  • John Maxwell
    ... Actually, at least in baseball and the WNBA, this has not proven to be the case. Generally speaking, bad teams win more close games than any other kind and
    Message 1 of 19 , Nov 30, 2001
    • 0 Attachment
      > "Clutch" play by a team is often reflected in the percent of close
      > games they win.

      Actually, at least in baseball and the WNBA, this has not proven to be the
      case. Generally speaking, bad teams win more close games than any other kind
      and good teams win fewer close games than any other kind.

      The reason is that bad teams don't win too many games that aren't close. Bad
      teams, pretty much by definition, aren't going to blow out better teams.
      They're going to play over their heads, while the better team plays below
      expectations resulting in a close win for the bad team.

      Good teams, on the other hand, routinely blow out their opponents, and
      engage in many mor eblot-outs throughout the course of a season than close
      games.

      This is not to say that bad teams have better winning percentages in close
      games than good teams, although there are a number of instances where this
      is the case. Last season in the WNBA, the Detroit Shock was 10-22 and 4-4 in
      games decided by three points or less. Sacramento was 20-12 and 3-3 in games
      decided by three points or less. Drawing the conclusion that both Detroit
      and Sacramento are both equally good teams "in the clutch" is inapproriate,
      in my opinion. Or, if it does mean that both teams are equally good "in the
      clutch," then perhaps having a good clutch team isn't all that important.

      Also in the WNBA last season, the Portland Fire were 10-22 on the year, but
      4-1 in overtime games. I would imagine that overtime is a decent barometer
      of the "pucker factor" in the regular season, but if Portland is such a
      clutch team, why are they only 10-22? On the other hand, Charlotte finished
      18-14 on the season and advanced to the WNBA Finals but was 0-3 in overtime
      games during the regular season.

      I'm obviously cherry-picking here, but to say that "clutch" play by a team
      is often reflected in the percent of close games it wins isn't supported by
      the facts.

      Here's a link to a baseball study on the issue
      http://www.baseballstuff.com/btf/scholars/ruane/articles/onerun.htm

      With regards to players, I looked at Yolanda Griffith and Lisa Leslie last
      year and how they performed "in the clutch" to back up my opinion that
      Griffith, and not Leslie, should have been the league MVP. I defined "in the
      clutch" as being any time in the last 5 minutes of a game where the teams
      were separated by no more than 5 points. Admittedly it was pretty arbitrary,
      but as you all have discussed, defining "clutch" performance is one of the
      larger stumbling block to determining if the ability to perform "in the
      clutch" exists.

      The first item of note from my study was that out of a possible 185 minutes
      for Leslie and 190 for Griffith each played just shy of 80 minutes worth of
      "clutch" time. That's two full games in the WNBA. Is that enough of a sample
      size, 80 minutes, to be able to determine a player's ability "in the
      clutch?" I don't know.

      Anyway, Leslie's field goal percentage dropped 140 points "in the clutch"
      while Griffith dropped 174. Leslie's rebound average dropped by two, her
      assist and blocked shots averages were down slightly, while she increased
      her steals average by half a point and decreased her turnover rate from 3.1
      to 1.7. Her scoring average decreased by two. She doubled her trips to the
      free throw line, but her percentage fell 100 points.

      Griffith's rebound average dropped half a board, her assists remained
      constant, she blocked no shots during this time (blocking 37 during the rest
      of the season) while she picked up an extra half of a steal and decreased
      her turnover rate from 2.34 to 1.28. Her scoring average decreased by more
      than five points. Her trips to the free throw line decreased slightly, but
      she hit essentially the same percentage.

      So what does that all mean? I haven't a clue. My gut tells me that the
      sample size is just too small to mean anything with regards to most of these
      numbers. And while the drop in field goal percentage is alarming, it may
      simply have to do with a difference in the way the opposition is defending
      these two players. Then again, it might be because these two, as go-to
      players, expect to take the shots at the end of the game and tend to force
      them as a result.

      I forget if it was Bill James, Rob Neyer or another Sabermetrician who did a
      study into "clutch" hitting in baseball players using whatever the
      definition in is for close and late situations -- something like after the
      7th inning down two runs or less. What they found during the years they
      studied (1980s) was players like Dane Iorg throughout the top-ten in batting
      average in these situations. They also found that there was no consistency
      with regards to these batting averages from year to year, leading them to
      conclude that "clutch" hitting was not an actual ability. While "clutch"
      performances exist, the idea that players have the ability to consistently
      perform above expectations "in the clutch" has yet to be proven.

      Just found the Rob Neyer article about which I was thinking.
      http://www.diamond-mind.com/articles/neyerclutch.htm

      John Maxwell
      Director of Public Relations
      Charlotte Sting
    • Ed Weiland
      ... Not only that, baseball teams with a good winning pct. in one-run games generally decline the following season (as myself and several other White Sox fans
      Message 2 of 19 , Dec 1, 2001
      • 0 Attachment
        --- John Maxwell <John.Maxwell@...>
        wrote:
        > > "Clutch" play by a team is often reflected in the
        > percent of close
        > > games they win.
        >
        > Actually, at least in baseball and the WNBA, this
        > has not proven to be the
        > case. Generally speaking, bad teams win more close
        > games than any other kind
        > and good teams win fewer close games than any other
        > kind.
        >
        Not only that, baseball teams with a good winning pct.
        in one-run games generally decline the following
        season (as myself and several other White Sox fans
        found out this past summer). I suspect the same is
        true in the NBA, though I have never looked at the
        subject, nor am I aware of anyone who has.

        I would have no idea how to analyze which players are
        clutch and which ones aren't. Basketball isn't like
        baseball where you can just look at what each player
        does in each AB and go from there. In basketball
        there's defense, rebounding and passing going on in
        addition to shooting. Those things would have to be
        looked at also, once clutch situations were defined.

        I've always felt "clutch" was one of those terms
        people used to describe players they wanted to like.
        Jerry West was called Mr. Clutch, despite being on the
        losing team in eight NBA finals and winning only once.
        This isn't to say West wasn't a clutch player. I just
        wonder why West got tagged with Mr. Clutch, when it
        was Bill Russell who was the biggest winner of that
        time. Probably a racial thing.


        Ed


        __________________________________________________
        Do You Yahoo!?
        Buy the perfect holiday gifts at Yahoo! Shopping.
        http://shopping.yahoo.com
      • Michael K. Tamada
        ... I agree 100% with the statements that it is a mistake to look at teams records in close games, and to try to call the ones with good records clutch . A
        Message 3 of 19 , Dec 1, 2001
        • 0 Attachment
          On Sat, 1 Dec 2001, Ed Weiland wrote:

          > Not only that, baseball teams with a good winning pct.
          > in one-run games generally decline the following
          > season (as myself and several other White Sox fans
          > found out this past summer). I suspect the same is
          > true in the NBA, though I have never looked at the
          > subject, nor am I aware of anyone who has.

          I agree 100% with the statements that it is a mistake to look at teams'
          records in close games, and to try to call the ones with good records
          "clutch".

          A minor quibble with the argument above however: while it is indeed true
          that baseball teams with a good winning pct. in one-run games can be
          expected to decline the following season, the same is true of ANY team in
          ANY sport in ANY sort of games. Bill James many years ago thought he'd
          discovered some profound truth in this and dreamt up some corny name for
          it -- "The Law of Elastic Reboound" or something -- but it's been known
          for about a century in statistics as "regression to the mean".

          A team which wins 72 games in an NBA season is EXTREMELY likely to have
          fewer wins the following season. A team which wins 90% of its 1-point
          games in a season is extremely likely to win a lower percentage the
          following season. A team which wins 70% of its 1-point games is very
          likely to win a lower pct. the following season. Etc.

          So while I agree 100% with both of the statements (that teams' 1-run
          records have little meaning, except of course to contribute to their
          win-loss record; and that teams with good 1-run records are likely to see
          a decline in those records the following season), it is not the case that
          the latter statement is evidence in favor of the former statement.

          > I would have no idea how to analyze which players are
          > clutch and which ones aren't. Basketball isn't like
          > baseball where you can just look at what each player
          > does in each AB and go from there. In basketball
          > there's defense, rebounding and passing going on in
          > addition to shooting. Those things would have to be
          > looked at also, once clutch situations were defined.

          True enough if we're looking for "total clutchness" but most of the NBA
          players who are known as clutch are known for being clutch as shooters
          during crunch time. Maybe once in a very long while they'll get a
          reputation for good D in crunch time (Havlicek steals the ball, Bird
          steals the ball), and I can't think of a single player who had a
          reputation as a clutch rebounder. Maybe, say, Wilt, Russell, Silas, et
          al -- but they were simply known as great rebounders period, it's not as
          if people thought they only grabbed rebounds during crunch time and
          lollygagged the rest of the game.

          So to look for clutch players, I think it's an easy step to limit the
          search to being a search for clutch *shooters*, and that is a more
          limited, specific, easy-to-define concept.

          > I've always felt "clutch" was one of those terms
          > people used to describe players they wanted to like.
          > Jerry West was called Mr. Clutch, despite being on the
          > losing team in eight NBA finals and winning only once.
          > This isn't to say West wasn't a clutch player. I just
          > wonder why West got tagged with Mr. Clutch, when it
          > was Bill Russell who was the biggest winner of that
          > time. Probably a racial thing.

          I agree with this also, although I would add the following hypothesis:
          some players are given (or demand) the ball a lot in clutch situations.
          And they thus shoot a lot of those crucial shots. I have no idea if some
          players have a systematically higher probability of making those shots,
          but if they take enough of them, some of them will go in. And people will
          remember those, and tend to forget the shots that they missed. And that
          will lead to the player getting a clutch reputation.

          E.g. maybe Jerry West shot in his career 100 clutch shots, and made 47 of
          them. That'd be identical to his career shooting percentage (both regular
          season and playoff). So unless there's a tendency for clutch shots to
          have a lower percentage overall (which actually might be the case), Jerry
          West shot no better in clutch situations than in non-clutch. But
          sportswriters, fans, and coaches would remember those 47 clutch shots
          made, whereas maybe Wilt only made 15 and Gail Goodrich only made 8, and
          thus Jerry West would get the reputation as Mr. Clutch.

          I would add that the notion that Mike Goodman and others have advocated,
          of looking at playoff games as clutch situations, is I think a good one,
          and the fact that West's FG% was as high in the playoffs as it was in the
          regular season is in itself a fairly remarkable, one might even say
          clutch, performance. Especially given that his scoring per game INCRASED.


          --MKT
        • Ed Weiland
          ... Increased shooting could also be a case of a player trying to shoulder too much of the load. It s interesting that in West s case the season his team
          Message 4 of 19 , Dec 2, 2001
          • 0 Attachment
            --- "Michael K. Tamada" <tamada@...> wrote:
            >
            > I would add that the notion that Mike Goodman and
            > others have advocated,
            > of looking at playoff games as clutch situations, is
            > I think a good one,
            > and the fact that West's FG% was as high in the
            > playoffs as it was in the
            > regular season is in itself a fairly remarkable, one
            > might even say
            > clutch, performance. Especially given that his
            > scoring per game INCRASED.


            Increased shooting could also be a case of a player
            trying to shoulder too much of the load. It's
            interesting that in West's case the season his team
            finally broke through and won the championship, 1972,
            was the only year he averaged fewer points in the
            playoffs than the regular season. West also shot only
            .376 during the 1972 playoffs, by far the worst
            showing of his career. He did post a career playoff
            high in assists per game during the '72 playoffs.

            Here are some other championship performances:

            Wilt in '67 averaged a then career-low 21.7 ppg in the
            playoffs, shot 104 points below his regular season FG
            pct. (albeit a more-than-adequate .579), but posted a
            career high with 9.0 assists per game.

            Hakeem in '94 and '95 had FG pct. similar to his
            regular season and career totals, but posted two of
            his three highest playoff assist per game totals, 4.5
            and 4.3 apg, both well above his career playoff
            average of 3.3. Hakeem scored 33.0 ppg in the '95
            playoffs, so it's not like he was sacrificing his
            shots.

            I'm not sure if the spike in assists is most
            responsible for the championships, but I don't think
            it can be ignored. Especially considering that star
            players who aren't point guards, but possessed
            good-to-great passing skills like Russell, Barry,
            Walton, Bird and Jordan tended to win championships.
            Sometimes the the most clutch thing for a player to do
            is to get his teammates involved.

            btw, I don't mean to knock West as non-clutch. HIs
            Laker teams lost three game sevens to the Celtics by a
            total of seven points. There had to be some bad luck
            involved in all that.

            Ed Weiland

            __________________________________________________
            Do You Yahoo!?
            Buy the perfect holiday gifts at Yahoo! Shopping.
            http://shopping.yahoo.com
          • harlanzo@yahoo.com
            When considering clutch it seems weird to think about players actually improving over how they would in normal (nonpressure) situations. Rather, it seems to
            Message 5 of 19 , Dec 2, 2001
            • 0 Attachment
              When considering clutch it seems weird to think about players
              actually improving over how they would in normal (nonpressure)
              situations. Rather, it seems to me that we might better define
              clutch by looking at who did not become worse in clutch situations.
              How you define clutch situations, incidentally, is a question I can't
              really answer.

              --- In APBR_analysis@y..., Ed Weiland <weiland1029@y...> wrote:
              >
              > --- "Michael K. Tamada" <tamada@o...> wrote:
              > >
              > > I would add that the notion that Mike Goodman and
              > > others have advocated,
              > > of looking at playoff games as clutch situations, is
              > > I think a good one,
              > > and the fact that West's FG% was as high in the
              > > playoffs as it was in the
              > > regular season is in itself a fairly remarkable, one
              > > might even say
              > > clutch, performance. Especially given that his
              > > scoring per game INCRASED.
              >
              >
              > Increased shooting could also be a case of a player
              > trying to shoulder too much of the load. It's
              > interesting that in West's case the season his team
              > finally broke through and won the championship, 1972,
              > was the only year he averaged fewer points in the
              > playoffs than the regular season. West also shot only
              > .376 during the 1972 playoffs, by far the worst
              > showing of his career. He did post a career playoff
              > high in assists per game during the '72 playoffs.
              >
              > Here are some other championship performances:
              >
              > Wilt in '67 averaged a then career-low 21.7 ppg in the
              > playoffs, shot 104 points below his regular season FG
              > pct. (albeit a more-than-adequate .579), but posted a
              > career high with 9.0 assists per game.
              >
              > Hakeem in '94 and '95 had FG pct. similar to his
              > regular season and career totals, but posted two of
              > his three highest playoff assist per game totals, 4.5
              > and 4.3 apg, both well above his career playoff
              > average of 3.3. Hakeem scored 33.0 ppg in the '95
              > playoffs, so it's not like he was sacrificing his
              > shots.
              >
              > I'm not sure if the spike in assists is most
              > responsible for the championships, but I don't think
              > it can be ignored. Especially considering that star
              > players who aren't point guards, but possessed
              > good-to-great passing skills like Russell, Barry,
              > Walton, Bird and Jordan tended to win championships.
              > Sometimes the the most clutch thing for a player to do
              > is to get his teammates involved.
              >
              > btw, I don't mean to knock West as non-clutch. HIs
              > Laker teams lost three game sevens to the Celtics by a
              > total of seven points. There had to be some bad luck
              > involved in all that.
              >
              > Ed Weiland
              >
              > __________________________________________________
              > Do You Yahoo!?
              > Buy the perfect holiday gifts at Yahoo! Shopping.
              > http://shopping.yahoo.com
            • igor eduardo küpfer
              ... Here s the correlation matrix. (I hope it formats ok.) Days Dist Home MatchupP Dist 0.076 0.000 Home 0.173 -0.249 0.000 0.000
              Message 6 of 19 , May 30, 2004
              • 0 Attachment
                Dean Oliver wrote:
                > Ed --
                >
                > Nice. Is there correlation between variables? One that is key to
                > understand is whether distance from previous game and days off are
                > correlated. A home stand could be hiding some aspect of time off
                > between games.

                Here's the correlation matrix. (I hope it formats ok.)

                Days Dist Home MatchupP
                Dist 0.076
                0.000

                Home 0.173 -0.249
                0.000 0.000

                MatchupP -0.014 0.021 0.001
                0.509 0.315 0.975

                PtsDiff 0.060 -0.041 0.226 0.465
                0.003 0.048 0.000 0.000

                Cell Contents: Pearson correlation
                P-Value

                The correlations are generally pretty low.

                > (Something also irks me about the p_win variable being
                > endogenous.)
                >

                I'm not quite sure what endogenous means. If it means being related to the
                other variables, I'm not quite sure if that's true: the matchup probability
                calculation uses only team winning percentage and opponent winning
                percentage, neither of which have any relationship to the other variables.
                Maybe I misunderstood.


                > I know I did a study of time off between games and saw that there is
                > an optimal period of time off (more than 2 wasn't good, but neither
                > was 0). That would imply a squared term in days off. But I didn't do
                > it as rigorously as you did.
                >

                I did something like that, too. I can't remember which season I used, but I
                found that most wins came on 2 day rests (I think). However, I didn't
                include any other variables, so I could have just been looking at a
                scheduling quirk for that season. I'll probably rerun this study on another
                season to see if the results hold. If anyone else wants to give it a shot,
                here's a table showing travel distances between NBA cities:

                http://members.rogers.com/brothered/junk/TravelDistances.htm

                ed
              • Dean Oliver
                ... Yeah, pretty low. Probably not much to worry about. ... to the ... probability ... variables. ... Basically, I assume you use 2004 win-loss records to
                Message 7 of 19 , May 30, 2004
                • 0 Attachment
                  --- In APBR_analysis@yahoogroups.com, igor eduardo küpfer
                  <edkupfer@r...> wrote:
                  > Dean Oliver wrote:
                  > > Ed --
                  > >
                  > > Nice. Is there correlation between variables? One that is key to
                  > > understand is whether distance from previous game and days off are
                  > > correlated. A home stand could be hiding some aspect of time off
                  > > between games.
                  >
                  > Here's the correlation matrix. (I hope it formats ok.)
                  >
                  > Days Dist Home MatchupP
                  > Dist 0.076
                  > 0.000
                  >
                  > Home 0.173 -0.249
                  > 0.000 0.000
                  >
                  > MatchupP -0.014 0.021 0.001
                  > 0.509 0.315 0.975
                  >
                  > PtsDiff 0.060 -0.041 0.226 0.465
                  > 0.003 0.048 0.000 0.000
                  >
                  > Cell Contents: Pearson correlation
                  > P-Value
                  >
                  > The correlations are generally pretty low.
                  >

                  Yeah, pretty low. Probably not much to worry about.

                  > > (Something also irks me about the p_win variable being
                  > > endogenous.)
                  > >
                  >
                  > I'm not quite sure what endogenous means. If it means being related
                  to the
                  > other variables, I'm not quite sure if that's true: the matchup
                  probability
                  > calculation uses only team winning percentage and opponent winning
                  > percentage, neither of which have any relationship to the other
                  variables.
                  > Maybe I misunderstood.

                  Basically, I assume you use 2004 win-loss records to evaluate p_win.
                  Well, those win-loss records are built from the things you are looking
                  at -- whether a team is at home or on the road, how many days off,
                  their whole schedule. Maybe the win-loss records of teams prior to
                  the matchup of the game you're looking at is exogenous (known a
                  priori). i.e., San Antonio faces the Lakers when one team is 12-5 and
                  the other is 10-7 -- use those records rather than their end of season
                  records. Maybe that's what you're doing, I dunno. I have doubt that
                  it would make a significant difference.


                  >
                  >
                  > > I know I did a study of time off between games and saw that there is
                  > > an optimal period of time off (more than 2 wasn't good, but neither
                  > > was 0). That would imply a squared term in days off. But I didn't do
                  > > it as rigorously as you did.
                  > >
                  >
                  > I did something like that, too. I can't remember which season I
                  used, but I
                  > found that most wins came on 2 day rests (I think). However, I didn't
                  > include any other variables, so I could have just been looking at a
                  > scheduling quirk for that season. I'll probably rerun this study on
                  another
                  > season to see if the results hold.

                  Just include the variable Days^2 in your regression and rerun that.
                  See what comes out significant.

                  DeanO

                  Dean Oliver
                  Author, Basketball on Paper
                  http://www.basketballonpaper.com
                  "Oliver goes beyond stats to dissect what it takes to win. His breezy
                  style makes for enjoyable reading, but there are plenty of points of
                  wisdom as well. This book can be appreciated by fans, players,
                  coaches and executives, but more importantly it can be used as a text
                  book for all these groups. You are sure to learn something you didn't
                  know about basketball here." Pete Palmer, co-author, Hidden Game of
                  Baseball and Hidden Game of Football
                • igor eduardo küpfer
                  Dean Oliver wrote: ... Ah. I will try to use contemporary win/loss records in my next analysis. ... You ll have to help me out here, as I don t
                  Message 8 of 19 , May 30, 2004
                  • 0 Attachment
                    Dean Oliver wrote:
                    <snip>

                    >>>
                    >>
                    >> I'm not quite sure what endogenous means. If it means being related
                    > to the
                    >> other variables, I'm not quite sure if that's true: the matchup
                    >> probability calculation uses only team winning percentage and
                    >> opponent winning percentage, neither of which have any relationship
                    >> to the other variables. Maybe I misunderstood.
                    >
                    > Basically, I assume you use 2004 win-loss records to evaluate p_win.
                    > Well, those win-loss records are built from the things you are looking
                    > at -- whether a team is at home or on the road, how many days off,
                    > their whole schedule. Maybe the win-loss records of teams prior to
                    > the matchup of the game you're looking at is exogenous (known a
                    > priori). i.e., San Antonio faces the Lakers when one team is 12-5 and
                    > the other is 10-7 -- use those records rather than their end of season
                    > records. Maybe that's what you're doing, I dunno. I have doubt that
                    > it would make a significant difference.

                    Ah. I will try to use contemporary win/loss records in my next analysis.

                    <snip>

                    > Just include the variable Days^2 in your regression and rerun that.
                    > See what comes out significant.
                    >

                    You'll have to help me out here, as I don't know anything about transforming
                    data. Do you mean include Days^2 in addition to Days or instead of Days? I
                    did both, and here's how they turned out:

                    PtsDiff = - 21.2 + 2.59 Days +0.000021 Dist + 5.59 Home + 30.0 MatchupP -
                    0.394 Days_2

                    Predictor Coef SE Coef T P
                    Constant -21.169 1.195 -17.71 0.000
                    Days 2.5924 0.7956 3.26 0.001
                    Dist 0.0000214 0.0003555 0.06 0.952
                    Home 5.5937 0.4858 11.51 0.000
                    MatchupP 29.991 1.134 26.45 0.000
                    Days_2 -0.3938 0.1350 -2.92 0.004

                    PtsDiff = - 18.1 +0.000099 Dist + 5.87 Home + 29.9 MatchupP + 0.0236 Days_2

                    Predictor Coef SE Coef T P
                    Constant -18.0811 0.7304 -24.76 0.000
                    Dist 0.0000990 0.0003555 0.28 0.781
                    Home 5.8677 0.4794 12.24 0.000
                    MatchupP 29.890 1.136 26.32 0.000
                    Days_2 0.02362 0.04275 0.55 0.581


                    ed
                  • Dean Oliver
                    ... I should note that, not being an economist, I like throwing this word around without as great an appreciation or understanding for it as I should. ...
                    Message 9 of 19 , May 30, 2004
                    • 0 Attachment
                      --- In APBR_analysis@yahoogroups.com, igor eduardo küpfer
                      <edkupfer@r...> wrote:
                      > >> I'm not quite sure what endogenous means. If it means being

                      I should note that, not being an economist, I like throwing this word
                      around without as great an appreciation or understanding for it as I
                      should.


                      > You'll have to help me out here, as I don't know anything about
                      transforming
                      > data. Do you mean include Days^2 in addition to Days or instead of
                      Days? I
                      > did both, and here's how they turned out:

                      Include both, which you did in the first set below. Looks like it got
                      you significant on both days and days^2. And the signs are as
                      expected. It suggests optimal rest at about 3 days, longer than the 2
                      days we saw before. (Potentially important for the talk about rust vs
                      rest, esp if the Lakers wrap up on M.) Let me also ask -- is Days = 0
                      if a team plays back to back nights or is that Days = 1?

                      I'm sure there are other ways to manipulate things, but this looks
                      like a pretty good thing. I'm saving it.

                      Home is a binary 1/0 indicator for home/road, resp?

                      >
                      > PtsDiff = - 21.2 + 2.59 Days +0.000021 Dist + 5.59 Home + 30.0
                      MatchupP -
                      > 0.394 Days_2
                      >
                      > Predictor Coef SE Coef T P
                      > Constant -21.169 1.195 -17.71 0.000
                      > Days 2.5924 0.7956 3.26 0.001
                      > Dist 0.0000214 0.0003555 0.06 0.952
                      > Home 5.5937 0.4858 11.51 0.000
                      > MatchupP 29.991 1.134 26.45 0.000
                      > Days_2 -0.3938 0.1350 -2.92 0.004
                      >
                      >

                      DeanO

                      Dean Oliver
                      Author, Basketball on Paper
                      http://www.basketballonpaper.com
                      "Dean Oliver looks at basketball with a fresh perspective. If you
                      want a new way to analyze the game, this book is for you. You'll
                      never watch a game the same way again. We use his stuff and it helps
                      us." Yvan Kelly, Scout, Seattle Sonics
                    • igor eduardo küpfer
                      Okay, I ran the test again, this time using 03-04 results. Before I show you what I got, let me address a couple of things. ... Hell, that s nothing. Once
                      Message 10 of 19 , May 31, 2004
                      • 0 Attachment
                        Okay, I ran the test again, this time using 03-04 results. Before I show you
                        what I got, let me address a couple of things.

                        Dean Oliver wrote:
                        > --- In APBR_analysis@yahoogroups.com, igor eduardo küpfer
                        > <edkupfer@r...> wrote:
                        >>>> I'm not quite sure what endogenous means. If it means being
                        >
                        > I should note that, not being an economist, I like throwing this word
                        > around without as great an appreciation or understanding for it as I
                        > should.

                        Hell, that's nothing. Once during the course of an argument with an
                        ex-girlfriend I used the word "heretofore." I still don't know what it
                        means.

                        >
                        >> You'll have to help me out here, as I don't know anything about
                        >> transforming data. Do you mean include Days^2 in addition to Days or
                        >> instead of Days? I did both, and here's how they turned out:
                        >
                        > Include both, which you did in the first set below. Looks like it got
                        > you significant on both days and days^2. And the signs are as
                        > expected. It suggests optimal rest at about 3 days, longer than the 2
                        > days we saw before. (Potentially important for the talk about rust vs
                        > rest, esp if the Lakers wrap up on M.)

                        Questions: I don't understand a couple of things about the squared term. How
                        did you know that squaring the Days variable would give a better fit? And,
                        just exactly how does it suggest the optimal 3 day rest?

                        > Let me also ask -- is Days = 0
                        > if a team plays back to back nights or is that Days = 1?
                        >

                        The latter. I am subtracting game dates from each other.

                        > I'm sure there are other ways to manipulate things, but this looks
                        > like a pretty good thing. I'm saving it.
                        >
                        > Home is a binary 1/0 indicator for home/road, resp?

                        Yes.

                        Okay. Here are the results for 03-04. For the Matchup Probability, I used
                        the team records heading into the game. For example, for two teams playing
                        their first games of the season, I would use 0-0 records for each team in my
                        probability calculation. Interestingly, this doesn't seem to affect the
                        regression results too much. The effect of Days between games is reduced in
                        this sample. Weird.


                        PtsDiff = - 13.6 + 7.31 Home +0.000027 Distance + 18.1 WinProb + 0.722
                        Days - 0.122 Days^2

                        Predictor Coef SE Coef T P
                        Constant -13.582 1.173 -11.58 0.000
                        Home 7.3056 0.5010 14.58 0.000
                        Distance 0.0000269 0.0003734 0.07 0.943
                        WinProb 18.054 1.163 15.53 0.000
                        Days 0.7221 0.7202 1.00 0.316
                        Days^2 -0.1216 0.1138 -1.07 0.286

                        S = 11.48 R-Sq = 16.7% R-Sq(adj) = 16.6%

                        Analysis of Variance

                        Source DF SS MS F P
                        Regression 5 62072 12414 94.19 0.000
                        Residual Error 2343 308806 132
                        Total 2348 370877

                        ed
                      • Dean Oliver
                        ... show you ... I ve had those moments, often inspired by arguments with soon-to-be ex-girlfriends. What the hell is vis-a-vis ? ... term. How ... fit? And,
                        Message 11 of 19 , May 31, 2004
                        • 0 Attachment
                          --- In APBR_analysis@yahoogroups.com, igor eduardo küpfer
                          <edkupfer@r...> wrote:
                          > Okay, I ran the test again, this time using 03-04 results. Before I
                          show you
                          > what I got, let me address a couple of things.
                          >
                          > Dean Oliver wrote:
                          > > --- In APBR_analysis@yahoogroups.com, igor eduardo küpfer
                          > > <edkupfer@r...> wrote:
                          > >>>> I'm not quite sure what endogenous means. If it means being
                          > >
                          > > I should note that, not being an economist, I like throwing this word
                          > > around without as great an appreciation or understanding for it as I
                          > > should.
                          >
                          > Hell, that's nothing. Once during the course of an argument with an
                          > ex-girlfriend I used the word "heretofore." I still don't know what it
                          > means.

                          I've had those moments, often inspired by arguments with soon-to-be
                          ex-girlfriends. What the hell is "vis-a-vis"?

                          > >
                          > >> You'll have to help me out here, as I don't know anything about
                          > >> transforming data. Do you mean include Days^2 in addition to Days or
                          > >> instead of Days? I did both, and here's how they turned out:
                          > >
                          > > Include both, which you did in the first set below. Looks like it got
                          > > you significant on both days and days^2. And the signs are as
                          > > expected. It suggests optimal rest at about 3 days, longer than the 2
                          > > days we saw before. (Potentially important for the talk about rust vs
                          > > rest, esp if the Lakers wrap up on M.)
                          >
                          > Questions: I don't understand a couple of things about the squared
                          term. How
                          > did you know that squaring the Days variable would give a better
                          fit? And,
                          > just exactly how does it suggest the optimal 3 day rest?

                          I didn't _know_ it would give a better fit. I hoped it would because
                          of what we were observing -- that there was an optimal number of days
                          off. The only way to get an optimum out of a regression is to throw
                          in higher order terms. Usually a squared term is plenty. It doesn't
                          answer the bigger question of whether teams get rusty, though. It
                          suggests an answer (another lesson in how to lie with statistics), one
                          that I wouldn't trust from this study.

                          Look at the results of your regression. Take just the Days and Days^2
                          coefficients and calculate the marginal net points those terms
                          contribute for Days = 1, 2, 3, 4, etc. You'll see a max at 3.

                          >
                          > > Let me also ask -- is Days = 0
                          > > if a team plays back to back nights or is that Days = 1?
                          > >
                          >
                          > The latter. I am subtracting game dates from each other.
                          >

                          So 2 days of rest is optimal.

                          > > I'm sure there are other ways to manipulate things, but this looks
                          > > like a pretty good thing. I'm saving it.
                          > >
                          > > Home is a binary 1/0 indicator for home/road, resp?
                          >
                          > Yes.
                          >
                          > Okay. Here are the results for 03-04. For the Matchup Probability, I
                          used
                          > the team records heading into the game. For example, for two teams
                          playing
                          > their first games of the season, I would use 0-0 records for each
                          team in my
                          > probability calculation.

                          I was curious to see how you handled the early games of the season,
                          especially the times where one team was undefeated. It looks like you
                          used Pythagorean projections, rather than real records anyway. That
                          helps. But 0-0 usually requires some other assumption, like a
                          Bayesian prior that carries through the first few games.

                          >Interestingly, this doesn't seem to affect the
                          > regression results too much. The effect of Days between games is
                          reduced in
                          > this sample. Weird.

                          Not sure what to make of that weakening of the Days. What was the R2
                          of the previous version? We may have to improve the prior matchup P
                          to get back a reasonable estimate of the value of Days. If you just
                          look at games beyond the first 20 in the season, does r2 get better
                          and does Days become more significant?

                          >
                          >
                          > PtsDiff = - 13.6 + 7.31 Home +0.000027 Distance + 18.1 WinProb + 0.722
                          > Days - 0.122 Days^2
                          >
                          > Predictor Coef SE Coef T P
                          > Constant -13.582 1.173 -11.58 0.000
                          > Home 7.3056 0.5010 14.58 0.000
                          > Distance 0.0000269 0.0003734 0.07 0.943
                          > WinProb 18.054 1.163 15.53 0.000
                          > Days 0.7221 0.7202 1.00 0.316
                          > Days^2 -0.1216 0.1138 -1.07 0.286
                          >
                          > S = 11.48 R-Sq = 16.7% R-Sq(adj) = 16.6%
                          >
                          > Analysis of Variance
                          >
                          > Source DF SS MS F P
                          > Regression 5 62072 12414 94.19 0.000
                          > Residual Error 2343 308806 132
                          > Total 2348 370877
                          >

                          DeanO

                          Dean Oliver
                          Author, Basketball on Paper
                          http://www.basketballonpaper.com
                          "Excellent writing. There are a lot of math guys who just rush from
                          the numbers to the conclusion. . .they'll tell you that Shaq is a real
                          good player but his team would win a couple more games a year if he
                          could hit a free throw. Dean is more than that; he's really
                          struggling to understand the actual problem, rather than the
                          statistical after-image of it. I learn a lot by reading him." Bill
                          James, author Baseball Abstract
                        • igor eduardo küpfer
                          Replies to DanR and DeanO ... I m sorry I didn t make it clear. For the second analysis (on the 03-04 regular season results) I didn;t use Pythagorean records.
                          Message 12 of 19 , Jun 2, 2004
                          • 0 Attachment
                            Replies to DanR and DeanO

                            Dean Oliver wrote:

                            > I was curious to see how you handled the early games of the season,
                            > especially the times where one team was undefeated. It looks like you
                            > used Pythagorean projections, rather than real records anyway. That
                            > helps. But 0-0 usually requires some other assumption, like a
                            > Bayesian prior that carries through the first few games.

                            I'm sorry I didn't make it clear. For the second analysis (on the 03-04
                            regular season results) I didn;t use Pythagorean records. I instead used
                            each team's record to date. Two teams facing each other on the first game of
                            the season each had a 0.5 chance of winning that game, since they had
                            identical 0-0 records.

                            The results don't deviate much from my first analysis, which used season's
                            end Pythagorean Win%. I supposed this is because after the first part of the
                            season, each team's Pyth is relatively stable. I must admit to being a
                            little surprised by this, though.

                            > Not sure what to make of that weakening of the Days. What was the R2
                            > of the previous version?

                            r = 0.06 for 00-01, r = 0.03 for this season.

                            > We may have to improve the prior matchup P
                            > to get back a reasonable estimate of the value of Days. If you just
                            > look at games beyond the first 20 in the season, does r2 get better
                            > and does Days become more significant?
                            >

                            Games 2-20: r = 0.048 (p = 0.261)
                            Games 21-82: r = 0.024 (p = 0.314

                            dan_t_rosenbaum wrote:

                            > Interesting results. Here are a couple of suggestions.
                            >
                            > I would leave out the MatchupP variable, since it is a lot like the
                            > dependent variable. Including it probably increases R-squared a
                            > lot, but probably doesn't do much else. (All in all, it probably is
                            > pretty harmless, since it unlikely to be correlated with your
                            > independent variables.)
                            >
                            > Another option with your day variable is to enter it as a series of
                            > dummy variables.
                            >
                            > DAY0 - equals 1 if 0 days of rest, 0 otherwise
                            > DAY1 - equals 1 if 1 day of rest , 0 otherwise
                            > DAY2 - equals 1 if 2 days of rest, 0 otherwise
                            > DAY3 - equals 1 if 3 days of rest, 0 otherwise
                            > DAY4+ - equals 1 if 4 days or more of rest, 0 otherwise
                            >
                            > Then run the regression leaving one of those variables out.
                            >
                            > If, for example, you left DAY0 out of the regression, the DAY1
                            > coefficient would give you the effect of playing on one day's rest
                            > versus playing in a back-to-back.
                            >
                            > The DAY2 coefficent would give you the effect of playing on two
                            > days' rest versus playing in a back-to-back.
                            >
                            > The DAY3 coefficent would give you the effect of playing on three
                            > days' rest versus playing in a back-to-back.
                            >
                            > The DAY4+ coefficent would give you the effect of playing on four or
                            > more days' rest versus playing in a back-to-back.
                            >

                            Okay, I tried this. The regression outputs follow. I'm afraid that I don't
                            know how to interpret the results -- very few of the coefficients are
                            significant. (Note that I use Day1 to mean 1 day between games, ie back to
                            back -- the 1 does not mean "rest days.")

                            Ommitting Days1

                            Predictor Coef SE Coef T P
                            Constant -3.8515 0.5971 -6.45 0.000
                            Home 7.1257 0.5267 13.53 0.000
                            Distance -0.0000105 0.0003920 -0.03 0.979
                            Days2 0.5837 0.6157 0.95 0.343
                            Days3 0.3022 0.7996 0.38 0.706
                            Days4 -2.363 1.394 -1.70 0.090
                            Days5+ 0.935 1.801 0.52 0.604


                            Omitting Days2

                            Predictor Coef SE Coef T P
                            Constant -3.2678 0.5624 -5.81 0.000
                            Home 7.1257 0.5267 13.53 0.000
                            Distance -0.0000105 0.0003920 -0.03 0.979
                            Days1 -0.5837 0.6157 -0.95 0.343
                            Days3 -0.2815 0.6981 -0.40 0.687
                            Days4 -2.947 1.338 -2.20 0.028
                            Days5+ 0.351 1.758 0.20 0.842


                            Omitting Days3

                            Predictor Coef SE Coef T P
                            Constant -3.5494 0.7833 -4.53 0.000
                            Home 7.1257 0.5267 13.53 0.000
                            Distance -0.0000105 0.0003920 -0.03 0.979
                            Days1 -0.3022 0.7996 -0.38 0.706
                            Days2 0.2815 0.6981 0.40 0.687
                            Days4 -2.665 1.426 -1.87 0.062
                            Days5+ 0.633 1.824 0.35 0.729


                            Omitting Days4

                            Predictor Coef SE Coef T P
                            Constant -6.215 1.385 -4.49 0.000
                            Home 7.1257 0.5267 13.53 0.000
                            Distance -0.0000105 0.0003920 -0.03 0.979
                            Days1 2.363 1.394 1.70 0.090
                            Days2 2.947 1.338 2.20 0.028
                            Days3 2.665 1.426 1.87 0.062
                            Days5+ 3.298 2.152 1.53 0.126

                            Omitting Days5+

                            Predictor Coef SE Coef T P
                            Constant -2.917 1.799 -1.62 0.105
                            Home 7.1257 0.5267 13.53 0.000
                            Distance -0.0000105 0.0003920 -0.03 0.979
                            Days1 -0.935 1.801 -0.52 0.604
                            Days2 -0.351 1.758 -0.20 0.842
                            Days3 -0.633 1.824 -0.35 0.729
                            Days4 -3.298 2.152 -1.53 0.126


                            ed
                          • igor eduardo küpfer
                            ... http://www.shrpsports.com/nba/stand/2002.htm -- ed
                            Message 13 of 19 , Nov 2, 2004
                            • 0 Attachment
                              ivan ivan wrote:
                              > this is a simple question
                              > but i can't find it anywhere....
                              >
                              >
                              > I'm doing analysis on how a history of winning or losing affects your
                              > chances of winning at the end of close games... so does anyone know
                              > where i can standings for the 2001-2002 NBA season?
                              > i want the home and away records?
                              >

                              http://www.shrpsports.com/nba/stand/2002.htm

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