kaosfere
09-13-2008, 03:57 PM
Background: As I mentioned here (http://www.operationsports.com/fofc/showpost.php?p=1832319&postcount=3), I was looking at this seasons' results thinking about what factors might go into a team's performance in a season that would cause it to differ from the Pythagorean expectation.
Toronto was notable, this year, for having a very high difference from expectation of +11 games. When I was looking at my team stats to try to figure out why this might be, I noticed something else interesting: my record in "clutch" games -- one-run and extra inning -- was also notably high (.769 and .739, respectively).
After a bit of thought, this started to make sense to me. Wins or loses in very close games will have a disproportionate effect in pulling your expectation away from the mean: a one-run win will not raise your expected average (EXP%) much, but it will raise your actual average just as much as a 10-run blowout. A team that plays in a lot of close games, then, can be expected to be likely to have a greater differential from their EXP% in reality.
I started to think about this a slightly different way. The typical thought about the Pythagorean model is that it measures a team's "luck". A team that is lucky and performing above its EXP% can be expected to regress towards the mean, the same as a team having an unlucky streak. However, this expectation may break down for teams with a tendency to close games, in which the clutch-factor may make for a bigger win/loss swing in the EXP% than one would expect.
Looking at it this way, I theorize that we can break down a team's EXP% into two different factors: "clutch", the ability to perform well (or poorly!) in tight ball games, and "luck" -- all the other nebulous factors that introduce fuzz to the statistics.
In the table below, I have columns for several statistics that I have invented -- and which all might need more thought and refinement! -- but the most relevant here are the two I would like to submit to you which I have termed (creatively) "CLUTCH" and "LUCK".
I have figured that if you take a team's W/L in one-run games, and subtract from it the team's overall W/L record, you can isolate the portion of their performance that comes from clutch situations. In the tables below, I have taken this one-run W/L minus overall W/L and multiplied it by 1000 to remove the usual percentage figures and give us an easier number to compare and manipulate.
Conversely, if you assume that EXP% is made of two components, then once you remove the clutch component from it you are left with pure luck. (Note that this isn't necessarily "luck" per se, just another, smaller sub-category of things which can alter the outcome of games). The LUCK statistic, then, is what we gate if we take a team's EXP% and subtract the clutch percentage (one run W/L - overall W/L, or CLUTCH/1000) -- then change the exponent on that just like we do for CLUTCH.
(This *1000 may not be necessarily, and may be a hindrance to further calculations, but it made it nicer for me to look at and compare the numbers on a general level.)
By breaking these two figures out, we can take any team's difference from expectations, and determine how much of that was a quantifiable ability, and how much was luck. In particular, if we know compare any two teams' CLUTCH standings, you may have a good idea how a close game between them would come out.
(And you can figure out, with LUCK, who really was the luckiest, and who might have fared even better -- or worse! -- if everyone's luck was the same).
So, anyway... here's a table I have constructed with each team's 1975 performance. Below it, you will find the definitions for all the non-standard columns:
W L AVG R RA EXP% DOE ORW ORL ORGA PORG CLUTCH LUCK
Chicago 89 65 0.578 756 616 0.601 (0.023) 20 20 0.500 0.260 -78 55
Columbus 85 69 0.552 678 646 0.524 0.028 35 22 0.614 0.370 62 (34)
Boston 80 74 0.519 726 695 0.522 (0.002) 21 26 0.447 0.305 -73 70
Compton 77 77 0.500 621 655 0.473 0.027 23 19 0.548 0.273 48 (21)
Colorado 75 79 0.487 619 620 0.499 (0.012) 26 24 0.520 0.325 33 (45)
Hartford 73 81 0.474 641 685 0.467 0.007 23 29 0.442 0.338 -32 39
Ann Arbor 73 81 0.474 683 711 0.480 (0.006) 19 21 0.475 0.260 1 (7)
New Orleans 64 90 0.416 628 724 0.429 (0.014) 17 23 0.425 0.260 9 (23)
Toronto 93 61 0.604 580 544 0.532 0.072 40 12 0.769 0.338 165 (93)
Valdosta 91 63 0.591 629 490 0.622 (0.031) 24 25 0.490 0.318 -101 70
Rio Grande 80 74 0.519 632 627 0.504 0.016 26 24 0.520 0.325 1 15
Atlanta 75 79 0.487 649 625 0.519 (0.032) 23 29 0.442 0.338 -45 13
Texas 72 82 0.468 574 620 0.462 0.006 21 26 0.447 0.305 -21 27
New York 72 82 0.468 577 622 0.463 0.005 29 27 0.518 0.364 50 (45)
Baltimore 68 86 0.442 558 567 0.492 (0.050) 23 36 0.390 0.383 -52 1
Brooklyn 65 89 0.422 553 657 0.415 0.007 24 31 0.436 0.357 14 (7)
DOE: Difference Over Expected (EXP% - AVG)
ORW: One Run Wins
ORL: One Run Losses
ORGA: One Run Game Average (ORW / (ORW + ORL))
PORG : Percentage of One Run Games ((ORW + ORL) / (W + L))
CLUTCH: (ORGA - AVG) * 1000
LUCK: (DOE - (ORGA - AVG)) * 1000
Toronto was notable, this year, for having a very high difference from expectation of +11 games. When I was looking at my team stats to try to figure out why this might be, I noticed something else interesting: my record in "clutch" games -- one-run and extra inning -- was also notably high (.769 and .739, respectively).
After a bit of thought, this started to make sense to me. Wins or loses in very close games will have a disproportionate effect in pulling your expectation away from the mean: a one-run win will not raise your expected average (EXP%) much, but it will raise your actual average just as much as a 10-run blowout. A team that plays in a lot of close games, then, can be expected to be likely to have a greater differential from their EXP% in reality.
I started to think about this a slightly different way. The typical thought about the Pythagorean model is that it measures a team's "luck". A team that is lucky and performing above its EXP% can be expected to regress towards the mean, the same as a team having an unlucky streak. However, this expectation may break down for teams with a tendency to close games, in which the clutch-factor may make for a bigger win/loss swing in the EXP% than one would expect.
Looking at it this way, I theorize that we can break down a team's EXP% into two different factors: "clutch", the ability to perform well (or poorly!) in tight ball games, and "luck" -- all the other nebulous factors that introduce fuzz to the statistics.
In the table below, I have columns for several statistics that I have invented -- and which all might need more thought and refinement! -- but the most relevant here are the two I would like to submit to you which I have termed (creatively) "CLUTCH" and "LUCK".
I have figured that if you take a team's W/L in one-run games, and subtract from it the team's overall W/L record, you can isolate the portion of their performance that comes from clutch situations. In the tables below, I have taken this one-run W/L minus overall W/L and multiplied it by 1000 to remove the usual percentage figures and give us an easier number to compare and manipulate.
Conversely, if you assume that EXP% is made of two components, then once you remove the clutch component from it you are left with pure luck. (Note that this isn't necessarily "luck" per se, just another, smaller sub-category of things which can alter the outcome of games). The LUCK statistic, then, is what we gate if we take a team's EXP% and subtract the clutch percentage (one run W/L - overall W/L, or CLUTCH/1000) -- then change the exponent on that just like we do for CLUTCH.
(This *1000 may not be necessarily, and may be a hindrance to further calculations, but it made it nicer for me to look at and compare the numbers on a general level.)
By breaking these two figures out, we can take any team's difference from expectations, and determine how much of that was a quantifiable ability, and how much was luck. In particular, if we know compare any two teams' CLUTCH standings, you may have a good idea how a close game between them would come out.
(And you can figure out, with LUCK, who really was the luckiest, and who might have fared even better -- or worse! -- if everyone's luck was the same).
So, anyway... here's a table I have constructed with each team's 1975 performance. Below it, you will find the definitions for all the non-standard columns:
W L AVG R RA EXP% DOE ORW ORL ORGA PORG CLUTCH LUCK
Chicago 89 65 0.578 756 616 0.601 (0.023) 20 20 0.500 0.260 -78 55
Columbus 85 69 0.552 678 646 0.524 0.028 35 22 0.614 0.370 62 (34)
Boston 80 74 0.519 726 695 0.522 (0.002) 21 26 0.447 0.305 -73 70
Compton 77 77 0.500 621 655 0.473 0.027 23 19 0.548 0.273 48 (21)
Colorado 75 79 0.487 619 620 0.499 (0.012) 26 24 0.520 0.325 33 (45)
Hartford 73 81 0.474 641 685 0.467 0.007 23 29 0.442 0.338 -32 39
Ann Arbor 73 81 0.474 683 711 0.480 (0.006) 19 21 0.475 0.260 1 (7)
New Orleans 64 90 0.416 628 724 0.429 (0.014) 17 23 0.425 0.260 9 (23)
Toronto 93 61 0.604 580 544 0.532 0.072 40 12 0.769 0.338 165 (93)
Valdosta 91 63 0.591 629 490 0.622 (0.031) 24 25 0.490 0.318 -101 70
Rio Grande 80 74 0.519 632 627 0.504 0.016 26 24 0.520 0.325 1 15
Atlanta 75 79 0.487 649 625 0.519 (0.032) 23 29 0.442 0.338 -45 13
Texas 72 82 0.468 574 620 0.462 0.006 21 26 0.447 0.305 -21 27
New York 72 82 0.468 577 622 0.463 0.005 29 27 0.518 0.364 50 (45)
Baltimore 68 86 0.442 558 567 0.492 (0.050) 23 36 0.390 0.383 -52 1
Brooklyn 65 89 0.422 553 657 0.415 0.007 24 31 0.436 0.357 14 (7)
DOE: Difference Over Expected (EXP% - AVG)
ORW: One Run Wins
ORL: One Run Losses
ORGA: One Run Game Average (ORW / (ORW + ORL))
PORG : Percentage of One Run Games ((ORW + ORL) / (W + L))
CLUTCH: (ORGA - AVG) * 1000
LUCK: (DOE - (ORGA - AVG)) * 1000