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

Clutch FT shooting...

Expand Messages
  • Dean Oliver
    Roland --- It s always entertaining listening to guys broadcast the NCAA Tournament. One radio broadcaster was blatantly making up stats yesterday, saying
    Message 1 of 9 , Mar 19, 2004
      Roland ---

      It's always entertaining listening to guys broadcast the NCAA
      Tournament. One radio broadcaster was blatantly making up stats
      yesterday, saying that 65% of 3pt shots are rebounded by the offense
      and that Jason Kidd shoots 95% from the line at clutch time and 60%
      other times. In looking at Jason Kidd and Kobe Bryant and thinking
      through a few things, I actually came to wonder whether any 15+ ppg
      scorer shoots better than their average under clutch conditions...

      DeanO

      Dean Oliver
      Author, Basketball on Paper
      http://www.basketballonpaper.com
      "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
    • Kevin Pelton
      ... Nice quote, DeanO. Did James reply to you again after finishing _Basketball on Paper_?
      Message 2 of 9 , Mar 19, 2004
        > "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

        Nice quote, DeanO. Did James reply to you again after finishing
        _Basketball on Paper_?
      • roland_beech
        ... offense ... Well, the shooting percentages on a league wide basis drop down considerably (FG and eFG at least) in the clutch moments of a game. The problem
        Message 3 of 9 , Mar 19, 2004
          > It's always entertaining listening to guys broadcast the NCAA
          > Tournament. One radio broadcaster was blatantly making up stats
          > yesterday, saying that 65% of 3pt shots are rebounded by the
          offense
          > and that Jason Kidd shoots 95% from the line at clutch time and 60%
          > other times. In looking at Jason Kidd and Kobe Bryant and thinking
          > through a few things, I actually came to wonder whether any 15+ ppg
          > scorer shoots better than their average under clutch conditions...
          >
          > DeanO


          Well, the shooting percentages on a league wide basis drop down
          considerably (FG and eFG at least) in the clutch moments of a game.
          The problem for individuals is sample size of course, and even using
          the looser "last five minutes, neither team ahead by more then 5"
          definition it's still pretty small pickings. And if you move the
          definition one way or the other it can change things quick.

          That said, Ray Allen's got a nifty .530 effective FG%, Gasol's
          at .550, Kirilenko's .571...

          and Yao? Yao can dance -- he's at .744 (29 of 39)

          I'll see if I can figure out how to upload a simple player "clutch
          stats" file to the group
        • igor eduardo küpfer
          ... From: Dean Oliver To: Sent: Friday, March 19, 2004 6:57 AM Subject: [APBR_analysis] Clutch FT
          Message 4 of 9 , Mar 19, 2004
            ----- Original Message -----
            From: "Dean Oliver" <deano@...>
            To: <APBR_analysis@yahoogroups.com>
            Sent: Friday, March 19, 2004 6:57 AM
            Subject: [APBR_analysis] Clutch FT shooting...


            >
            > Roland ---
            >
            > It's always entertaining listening to guys broadcast the NCAA
            > Tournament. One radio broadcaster was blatantly making up stats
            > yesterday, saying that 65% of 3pt shots are rebounded by the offense
            > and that Jason Kidd shoots 95% from the line at clutch time and 60%
            > other times. In looking at Jason Kidd and Kobe Bryant and thinking
            > through a few things, I actually came to wonder whether any 15+ ppg
            > scorer shoots better than their average under clutch conditions...
            >
            > DeanO
            >

            According to Roland's spreadsheet, only four +15ppg players out of 68 shoot
            FTs at a significantly different percentage in clutch time, two better and
            two worse. Top 10 lowest p-values:

            Team Player nonClutchFTM-FTA ClutchFTM-FTA Diff p
            MIN Sprewell 209-260 19-30 -17.1% 0.022
            BOS Pierce 446-534 54-72 -8.5% 0.042
            ATL Abdur-Rahim 343-393 23-23 12.7% 0.044
            WAS Hughes 203-251 22-23 14.8% 0.049
            CHI Crawford 160-193 12-18 -16.2% 0.072
            HOU Francis 261-342 42-49 9.4% 0.079
            DAL Nowitzki 291-332 26-33 -8.9% 0.105
            ATL Jackson 159-210 13-21 -13.8% 0.114
            UTA Kirilenko 319-400 20-29 -10.8% 0.115
            SEA Allen 182-203 15-19 -10.7% 0.126


            ed
          • Dean Oliver
            ... Yes, Bill James did initially respond on the book after the first few chapters with how pleased he was, then this came a few days ago. So, yeah, that
            Message 5 of 9 , Mar 19, 2004
              --- In APBR_analysis@yahoogroups.com, "Kevin Pelton" <kpelton08@h...>
              wrote:
              > > "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
              >
              > Nice quote, DeanO. Did James reply to you again after finishing
              > _Basketball on Paper_?

              Yes, Bill James did initially respond on the book after the first few
              chapters with how pleased he was, then this came a few days ago. So,
              yeah, that should help.

              DeanO

              Dean Oliver
              Author, Basketball on Paper
              http://www.basketballonpaper.com
              When basketball teams start playing Moneyball, this is the book
              they'll use!
            • Michael Tamada
              ... From: igor eduardo küpfer [mailto:edkupfer@rogers.com] Sent: Friday, March 19, 2004 12:49 PM ... From: Dean Oliver To:
              Message 6 of 9 , Mar 19, 2004
                -----Original Message-----
                From: igor eduardo küpfer [mailto:edkupfer@...]
                Sent: Friday, March 19, 2004 12:49 PM

                ----- Original Message -----
                From: "Dean Oliver" <deano@...>
                To: <APBR_analysis@yahoogroups.com>

                [...]

                >> through a few things, I actually came to wonder whether any 15+ ppg
                >> scorer shoots better than their average under clutch conditions...
                >>
                >> DeanO

                [...]

                >According to Roland's spreadsheet, only four +15ppg players out of 68 shoot
                >FTs at a significantly different percentage in clutch time, two better and
                >two worse. Top 10 lowest p-values:
                >
                >Team Player nonClutchFTM-FTA ClutchFTM-FTA Diff p
                >MIN Sprewell 209-260 19-30 -17.1% 0.022
                >BOS Pierce 446-534 54-72 -8.5% 0.042
                >ATL Abdur-Rahim 343-393 23-23 12.7% 0.044
                >WAS Hughes 203-251 22-23 14.8% 0.049

                [...]

                >ed

                If we choose a 5% significant level, (5% probability of a Type I error),
                then out of a sample of 68, under the null hypothesis of no difference
                between clutch and non-clutch, we'd expect exactly 3.4 observations to
                show differences with a p-value of less than .05.

                This sample of 68 has 4 observations with p-values < .05.

                In other words, these numbers look almost exactly like the ones that
                random chance would produce, under the null hypothesis.

                So not only do 64 of the 68 show no significant difference, the 4 who
                do appear significant quite possibly may simply be reflecting pure chance,
                and not any true clutch or non-clutch shooting ability.

                Some contributors to baseballprimer.com have started using Stein
                estimators, which "shrink" or "regress" estimates to the overall mean,
                which is a way of correcting for the odd random 3.4 or 4 observations
                that pop up with high or low values.

                BUT: small sample sizes. It's only, what, 3/4 of a season? So it would
                be prudent to re-examine this issue periodically.

                In statistical terms, our small sample sizes mean that our tests have
                low power, i.e. a high probability of a Type II error. There may
                in fact be clutch vs non-clutch differences there, but we can't detect
                them yet. Let's look again at the end of the year, then next year,
                and the year after that. My informal judgement is that it'd be around
                that point that, if we're still seeing a lack of significant results,
                that we would be able to conclusively say there appears to be no
                clutch FT shooting.


                --MKT
              • dan_t_rosenbaum
                This is very nice work, Michael - just the kind of thing that makes coming to APBR Analysis great. Thanks! ... ppg ... conditions... ... of 68 shoot ...
                Message 7 of 9 , Mar 20, 2004
                  This is very nice work, Michael - just the kind of thing that makes
                  coming to APBR Analysis great. Thanks!

                  --- In APBR_analysis@yahoogroups.com, "Michael Tamada" <tamada@o...>
                  wrote:
                  > -----Original Message-----
                  > From: igor eduardo küpfer [mailto:edkupfer@r...]
                  > Sent: Friday, March 19, 2004 12:49 PM
                  >
                  > ----- Original Message -----
                  > From: "Dean Oliver" <deano@r...>
                  > To: <APBR_analysis@yahoogroups.com>
                  >
                  > [...]
                  >
                  > >> through a few things, I actually came to wonder whether any 15+
                  ppg
                  > >> scorer shoots better than their average under clutch
                  conditions...
                  > >>
                  > >> DeanO
                  >
                  > [...]
                  >
                  > >According to Roland's spreadsheet, only four +15ppg players out
                  of 68 shoot
                  > >FTs at a significantly different percentage in clutch time, two
                  better and
                  > >two worse. Top 10 lowest p-values:
                  > >
                  > >Team Player nonClutchFTM-FTA ClutchFTM-FTA Diff p
                  > >MIN Sprewell 209-260 19-30 -17.1% 0.022
                  > >BOS Pierce 446-534 54-72 -8.5% 0.042
                  > >ATL Abdur-Rahim 343-393 23-23 12.7% 0.044
                  > >WAS Hughes 203-251 22-23 14.8% 0.049
                  >
                  > [...]
                  >
                  > >ed
                  >
                  > If we choose a 5% significant level, (5% probability of a Type I
                  error),
                  > then out of a sample of 68, under the null hypothesis of no
                  difference
                  > between clutch and non-clutch, we'd expect exactly 3.4
                  observations to
                  > show differences with a p-value of less than .05.
                  >
                  > This sample of 68 has 4 observations with p-values < .05.
                  >
                  > In other words, these numbers look almost exactly like the ones
                  that
                  > random chance would produce, under the null hypothesis.
                  >
                  > So not only do 64 of the 68 show no significant difference, the 4
                  who
                  > do appear significant quite possibly may simply be reflecting pure
                  chance,
                  > and not any true clutch or non-clutch shooting ability.
                  >
                  > Some contributors to baseballprimer.com have started using Stein
                  > estimators, which "shrink" or "regress" estimates to the overall
                  mean,
                  > which is a way of correcting for the odd random 3.4 or 4
                  observations
                  > that pop up with high or low values.
                  >
                  > BUT: small sample sizes. It's only, what, 3/4 of a season? So it
                  would
                  > be prudent to re-examine this issue periodically.
                  >
                  > In statistical terms, our small sample sizes mean that our tests
                  have
                  > low power, i.e. a high probability of a Type II error. There may
                  > in fact be clutch vs non-clutch differences there, but we can't
                  detect
                  > them yet. Let's look again at the end of the year, then next year,
                  > and the year after that. My informal judgement is that it'd be
                  around
                  > that point that, if we're still seeing a lack of significant
                  results,
                  > that we would be able to conclusively say there appears to be no
                  > clutch FT shooting.
                  >
                  >
                  > --MKT
                • igor eduardo küpfer
                  ... shoot ... and ... Thanks Mike, that s what I was getting at. ... As it happens, I ran a similar test on last season s 82games clutch stats. The
                  Message 8 of 9 , Mar 21, 2004
                    DeanO:

                    > >> through a few things, I actually came to wonder whether any 15+ ppg
                    > >> scorer shoots better than their average under clutch conditions...
                    > >>
                    > >> DeanO
                    >
                    > [...]
                    >

                    Ed:

                    > >According to Roland's spreadsheet, only four +15ppg players out of 68
                    shoot
                    > >FTs at a significantly different percentage in clutch time, two better
                    and
                    > >two worse. Top 10 lowest p-values:
                    > >
                    > >Team Player nonClutchFTM-FTA ClutchFTM-FTA Diff p
                    > >MIN Sprewell 209-260 19-30 -17.1% 0.022
                    > >BOS Pierce 446-534 54-72 -8.5% 0.042
                    > >ATL Abdur-Rahim 343-393 23-23 12.7% 0.044
                    > >WAS Hughes 203-251 22-23 14.8% 0.049
                    >
                    > [...]

                    MikeT:

                    > If we choose a 5% significant level, (5% probability of a Type I error),
                    > then out of a sample of 68, under the null hypothesis of no difference
                    > between clutch and non-clutch, we'd expect exactly 3.4 observations to
                    > show differences with a p-value of less than .05.
                    >
                    > This sample of 68 has 4 observations with p-values < .05.
                    >
                    > In other words, these numbers look almost exactly like the ones that
                    > random chance would produce, under the null hypothesis.
                    >
                    > So not only do 64 of the 68 show no significant difference, the 4 who
                    > do appear significant quite possibly may simply be reflecting pure chance,
                    > and not any true clutch or non-clutch shooting ability.
                    >

                    Thanks Mike, that's what I was getting at.

                    <snip>

                    >
                    > BUT: small sample sizes. It's only, what, 3/4 of a season? So it would
                    > be prudent to re-examine this issue periodically.
                    >
                    > In statistical terms, our small sample sizes mean that our tests have
                    > low power, i.e. a high probability of a Type II error. There may
                    > in fact be clutch vs non-clutch differences there, but we can't detect
                    > them yet. Let's look again at the end of the year, then next year,
                    > and the year after that. My informal judgement is that it'd be around
                    > that point that, if we're still seeing a lack of significant results,
                    > that we would be able to conclusively say there appears to be no
                    > clutch FT shooting.
                    >

                    As it happens, I ran a similar test on last season's 82games clutch stats.
                    The pattern is the same: among 15+ppg players, 5 out of 64 (8%) had clutch
                    FT%s significantly different from non-clutch FT%s (that is, significant at
                    5%). Interestingly, Sprewell makes the list for both this year and last.

                    Player FTm-FTA cFTm-cFTA Diff p
                    Zydrunas Ilgauskas 354-440 46-72 -16.6% 0.0008
                    Tracy McGrady 519-645 57-81 -10.1% 0.0193
                    Shawn Marion 203-244 48-51 10.9% 0.0198
                    Allan Houston 317-341 46-54 -7.8% 0.0343
                    Latrell Sprewell 142-174 31-44 -11.2% 0.0490

                    What else can we do? There are other situations which fall under the
                    category of "clutch" -- playoff games when one's team is facing elimination,
                    for example. I made a list of elimination games since 1996, and came up with
                    173 team-games (game sevens count as two team-games, as both teams are
                    facing elimination). In those games, 39 different players attempted thirty
                    or more free throws (spread among more than 1 game, obviously). Four of
                    those thirty-nine players (10%) posted FT%s significantly different from
                    their career averages -- I used career FT% in this case for ease. This is
                    not what I used above, where non-clutch FT% in that single season was used
                    for the probability calculations.

                    Player FTm-FTA CareerFT% Diff p
                    Gary Payton 71- 85 72.7% 10.8% 0.0137
                    Michael Finley 33- 35 80.2% 14.0% 0.0206
                    John Stockton 50- 54 82.6% 10.0% 0.0307
                    Karl Malone 83-124 74.2% -7.3% 0.0429

                    The whole thing still smells to me like random variation. I am comfortable
                    with the idea that clutch ability doesn't exist, but I realize there hardly
                    has been enough testing to make that conclusion. I think that we can
                    provisionally accept that if clutch FT shooting ability does exist, it is a
                    small effect.


                    ed
                  • Michael Tamada
                    Nice work, I would modify your provisional conclusion just a little bit: provisionally, it appears that clutch FT shooting is a small or non-existent effect
                    Message 9 of 9 , Mar 24, 2004
                      Nice work, I would modify your provisional conclusion just a little bit:
                      provisionally, it appears that clutch FT shooting is a small or non-existent
                      effect for almost all players. But there might be a few who have non-trivial
                      clutch impacts ... Sprewell, Payton, etc. However, it's too early to conclude
                      that they truly have clutch effects, given the possibility of Type I errors.


                      --MKT


                      -----Original Message-----
                      From: igor eduardo küpfer [mailto:edkupfer@...]
                      Sent: Sunday, March 21, 2004 1:35 AM
                      To: APBR_analysis@yahoogroups.com
                      Subject: Re: [APBR_analysis] Clutch FT shooting...


                      DeanO:

                      > >> through a few things, I actually came to wonder whether any 15+ ppg
                      > >> scorer shoots better than their average under clutch conditions...
                      > >>
                      > >> DeanO
                      >
                      > [...]
                      >

                      Ed:

                      > >According to Roland's spreadsheet, only four +15ppg players out of 68
                      shoot
                      > >FTs at a significantly different percentage in clutch time, two better
                      and
                      > >two worse. Top 10 lowest p-values:
                      > >
                      > >Team Player nonClutchFTM-FTA ClutchFTM-FTA Diff p
                      > >MIN Sprewell 209-260 19-30 -17.1% 0.022
                      > >BOS Pierce 446-534 54-72 -8.5% 0.042
                      > >ATL Abdur-Rahim 343-393 23-23 12.7% 0.044
                      > >WAS Hughes 203-251 22-23 14.8% 0.049
                      >
                      > [...]

                      MikeT:

                      > If we choose a 5% significant level, (5% probability of a Type I error),
                      > then out of a sample of 68, under the null hypothesis of no difference
                      > between clutch and non-clutch, we'd expect exactly 3.4 observations to
                      > show differences with a p-value of less than .05.
                      >
                      > This sample of 68 has 4 observations with p-values < .05.
                      >
                      > In other words, these numbers look almost exactly like the ones that
                      > random chance would produce, under the null hypothesis.
                      >
                      > So not only do 64 of the 68 show no significant difference, the 4 who
                      > do appear significant quite possibly may simply be reflecting pure chance,
                      > and not any true clutch or non-clutch shooting ability.
                      >

                      Thanks Mike, that's what I was getting at.

                      <snip>

                      >
                      > BUT: small sample sizes. It's only, what, 3/4 of a season? So it would
                      > be prudent to re-examine this issue periodically.
                      >
                      > In statistical terms, our small sample sizes mean that our tests have
                      > low power, i.e. a high probability of a Type II error. There may
                      > in fact be clutch vs non-clutch differences there, but we can't detect
                      > them yet. Let's look again at the end of the year, then next year,
                      > and the year after that. My informal judgement is that it'd be around
                      > that point that, if we're still seeing a lack of significant results,
                      > that we would be able to conclusively say there appears to be no
                      > clutch FT shooting.
                      >

                      As it happens, I ran a similar test on last season's 82games clutch stats.
                      The pattern is the same: among 15+ppg players, 5 out of 64 (8%) had clutch
                      FT%s significantly different from non-clutch FT%s (that is, significant at
                      5%). Interestingly, Sprewell makes the list for both this year and last.

                      Player FTm-FTA cFTm-cFTA Diff p
                      Zydrunas Ilgauskas 354-440 46-72 -16.6% 0.0008
                      Tracy McGrady 519-645 57-81 -10.1% 0.0193
                      Shawn Marion 203-244 48-51 10.9% 0.0198
                      Allan Houston 317-341 46-54 -7.8% 0.0343
                      Latrell Sprewell 142-174 31-44 -11.2% 0.0490

                      What else can we do? There are other situations which fall under the
                      category of "clutch" -- playoff games when one's team is facing elimination,
                      for example. I made a list of elimination games since 1996, and came up with
                      173 team-games (game sevens count as two team-games, as both teams are
                      facing elimination). In those games, 39 different players attempted thirty
                      or more free throws (spread among more than 1 game, obviously). Four of
                      those thirty-nine players (10%) posted FT%s significantly different from
                      their career averages -- I used career FT% in this case for ease. This is
                      not what I used above, where non-clutch FT% in that single season was used
                      for the probability calculations.

                      Player FTm-FTA CareerFT% Diff p
                      Gary Payton 71- 85 72.7% 10.8% 0.0137
                      Michael Finley 33- 35 80.2% 14.0% 0.0206
                      John Stockton 50- 54 82.6% 10.0% 0.0307
                      Karl Malone 83-124 74.2% -7.3% 0.0429

                      The whole thing still smells to me like random variation. I am comfortable
                      with the idea that clutch ability doesn't exist, but I realize there hardly
                      has been enough testing to make that conclusion. I think that we can
                      provisionally accept that if clutch FT shooting ability does exist, it is a
                      small effect.


                      ed





                      Yahoo! Groups Links
                    Your message has been successfully submitted and would be delivered to recipients shortly.