New Statistic! PNOPS!
Dale Stephenson
[...Position Normaliztion...]
>Below you will see every major league starter ranked by how his adjusted
>OPS (OBP*1.2+SLG) compares to the average starter at that position. PNOPS
>can be read as a percentage of average for the position. This means that
>Frank Thomas is compared not to everybody, but only to other AL first
>basemen -- a much tougher group, as befits the lesser defensive demands
>of first base. I also include a ranking of the OPS* by position in each
>league, and a wrap-up of the best and worst OPS* and PNOPS. Note that as
>expected, the OPS* leaders are loaded with first basemen and outfielders,
>while the PNOPS leaders include all positions.
[...]
>OPS* = OBP*1.2+SLG
>PNOPS = OPS* / average of all starters at that position
One flaw you didn't mention -- what if the lesser defensive demands don't
correlate into increased offensive. As usual, the most egregious offender
is in the AL:
>AMERICAN LEAGUE
>===============
[...]
>By Position OPS* PNOPS
>============= ===== ======
>First Base .954 111
>Right Field .948 111
>Designated Hit .935 109
>Center Field .854 100
>Third Base .853 100
>Left Field .845 99
>Second Base .816 95
>Catcher .766 89
>Shortstop .745 87
>
>Average: .857
First basemen and rightfielders outhit designated hitters -- not too
uncommon, since DHs are rarely the best hitters in the league. But
*any* first baseman and *any* right fielder can play DH as well as any
current DH. Under this particular adjustment, any right fielder or
first basemen could improve their "value" by playing DH -- that's not
right. Likewise, if Reggie Jefferson started playing RF (assuming average
defense) his value would go down -- that's not right. I think the only
good way to deal with the (abomination of) DH is to compare them to the
best group of hitters.
[Side note -- some firstbasemen may be more valuable as DHs, due to
poor first base defense. But I don't think their offense should be
considered more valuable if produced from the DH slot]
A related consideration is outfielders -- how different are the outfield
spots? Traditionally, centerfield is considered the most difficult, and
I would guess that anyone who can play center can play left. If true,
a .854 OPS* centerfielder *should* be more valuable than a .845 leftfielder,
since we know he *could* be an .854 OPS* leftfielder. (We don't know he
could be a .854 OPS* infielder or catcher, though.)
>NATIONAL LEAGUE
>===============
>
>By Position OPS* PNOPS
>============ ===== =====
>Left Field .927 109
>Center Field .926 109 Note -- this is for regulars only,
>Third Base .887 105 not all players.
>Right Field .868 102
>First Base .837 99
>Catcher .814 96
>Second Base .812 96
>Shortstop .706 83
>Pitcher not available
>
>Average: .847
>
Again the same considerations apply. Could the centerfielder play right?
Can the outfielder or third baseman play first? I have no problem with
letting the skill positions (CF, SS, 2B, 3B, C) seeking positional
considerations, but when firstbasemen get a slight plus for being able
to handle first I get a little wary.
--
Dale J. Stephenson |*| (st...@cs.uiuc.edu) |*| Baseball fanatic
"All the rivers run into the sea; yet the sea is not full"
-- Ecclesiastes 1:7
RVES...@vma.cc.nd.edu
smart. i'm not completely convinced that the defense gained by playing
a shortstop at shortstop outweighs the offense gained by playing a
first baseman at shortstop.
bob vesterman.
Steven Thornton
stat with grotesque errors. In fact, all AL shortstops are NOT below
average, which should be a relief to Boston fans among others. Here's the
fixed version.
To recap: PNOPS, "Positionally Normalized On-base Plus Slugging"
(pronounced "Pee-Nopps") is a measure of how a player's corrected OPS*
(OBP*1.2 + SLG) stacks up against the league average _for his position_.
No more comparing Frank Thomas to Felix Fermin directly -- they play
different positions with different defensive and offensive expectations
and norms. (Frank still wins, duh). This is one way I have concocted to
include some defensive adjustments in ranking batters, because a CF with
the same OPS* as a DH is a more valuable player. No, it doesn't adjust
for how _well_ you play the position. Sorry.
The formula is (player OPS*) / (average OPS* at his position) * 100, and
can be read as "percent of average offensive ability at the position".
The average is taken to be the average of all regulars, not everybody
who's played the position; my data source (McWeekly) doesn't provide
this. So the averages are a little high, since most non-regulars hit
worse than most regulars. I can live with it.
I included every player who has the most innings at each position for his
team, AND played at least once in the preceding week -- no injured guys,
until they get better. There is also a comparison of the positions
against each other, and a summary of the best and worst OPS* and PNOPS in
each league. As expected, the best OPS* is loaded with 1B, while the best
PNOPS is evenly divided amongst the positions. Similarly, the worst OPS*
has lots of SS, while the worst PNOPS has every position.
AMERICAN LEAGUE through May 8
===============
Catcher OPS* PNOPS First Base OPS* PNOPS
============= ===== ===== ============ ===== =====
Karkovice, CHI 1.086 142 Thomas, CHI 1.263 132
Steinbach, OAK .983 128 Palmiero, BAL 1.237 130
Macfarlane, KC .958 125 Vaughn, BOS 1.230 129
Tettleton, DET .931 122 Clark, TEX 1.187 124
Hoiles, BAL .860 112 Olerud, TOR .991 104
Rodriguez, TEX .834 109 Fielder, DET .976 102
Nillson, MIL .776 101 Neel, OAK .925 97
Stanley, NY .775 101 Mattingly, NY .924 97
Alomar, CLE .753 98 Joyner, KC .920 96
Borders, TOR .684 89 Jaha, MIL .835 88
Valle, BOS .601 78 McCarty, MIN .754 79
Wilson, SEA .585 76 Perez, CAL .732 77
Walbeck, MIN .556 73 Sorrento, CLE .731 77
Turner, CAL .348 45 T.Martinez, SEA .654 69
Average C: .766 Average 1B: .954
Second Base OPS* PNOPS Third Base OPS* PNOPS
============ ===== ===== ============ ===== =====
Whitaker, DET 1.074 132 Cooper, BOS 1.083 127
Naehring, BOS 1.045 128 Ventura, CHI 1.022 120
Alomar, TOR .928 114 Seitzer, MIL .964 113
Amaral, SEA .862 106 Leius, MIN .954 112
Kelly, NY .856 105 Gaetti, KC .942 110
Knoblauch, MIN .835 102 Sprague, TOR .870 102
Reynolds, CAL .811 99 Boggs, NY .851 100
Baerga, CLE .808 99 Thome, CLE .814 95
Strange, TEX .773 95 Easley, CAL .798 94
McLemore, BAL .731 90 Ripken, TEX .791 93
Reed, MIL .722 88 Fryman, DET .770 90
Cora, CHI .711 87 Sabo, BAL .767 90
Sax, OAK .633 78 Brosius, OAK .718 84
Lind, KC .631 77 Blowers, SEA .603 71
Average 2B: .816 Average 3B: .853
Shortstop OPS* PNOPS Left Field OPS* PNOPS
============ ===== ===== ============ ===== =====
JnValentin, BOS .889 119 Belle, CLE 1.091 129
Trammell, DET .880 118 Ward, MIL 1.021 121
Gagne, KC .835 112 Raines, CHI .956 113
Bordick, OAK .785 105 Phillips, DET .935 111
Ripken, BAL .780 105 Delgado, TOR .885 105
Schofield, TOR .776 104 Gonzalez, TEX .861 102
DiSarcina, CAL .769 103 Anthony, SEA .860 102
Lee, TEX .741 99 Greenwell, BOS .845 100
Fermin, SEA .741 99 Polonia, NY .823 97
Meares, MIN .719 96 Anderson, BAL .809 96
Lewis, CLE .701 94 Henderson, OAK .791 94
Gallego, NY .656 88 Munoz, MIN .721 85
Guillen, CHI .635 85 Smith, CAL .704 83
JsValentin, MIL .525 70 Coleman, KC .534 63
Average SS: .745 Average LF: .845
Center Field OPS* PNOPS Right Field OPS* PNOPS
============ ===== ===== ============ ===== =====
Griffey, SEA 1.180 138 O'Neill, NY 1.477 156
Lofton, CLE 1.071 125 Carter, TOR 1.149 121
White, TOR 1.029 120 Buhner, SEA 1.093 115
Cole, MIN 1.023 120 Ramirez, CLE 1.045 110
McRae, KC .978 115 Hammonds, BAL 1.002 106
Javier, OAK .943 110 Jackson, CHI .968 102
Devereaux, BAL .813 95 Gibson, DET .890 94
Hamilton, MIL .770 90 James, TEX .886 93
Hulse, TEX .756 89 Salmon, CAL .883 93
B.Williams, NY .696 82 Puckett, MIN .879 93
Johnson, CHI .695 81 Jose, KC .798 84
Nixon, BOS .687 80 Sierra, OAK .770 81
Curtis, CAL .678 79 Hatcher, BOS .750 79
Davis, DET .635 74 Mieske, MIL .682 72
Average CF: .854 Average RF: .948
Designated Hit OPS* PNOPS By Position OPS* PNOPS
============= ===== ===== ============= ===== =====
Jefferson, SEA 1.415 151 First Base .954 111
Davis, CAL 1.126 120 Right Field .948 111
Berroa, OAK 1.075 115 Designated Hit .935 109
Hamelin, KC 1.032 110 Center Field .854 100
Canseco, TEX 1.029 110 Third Base .853 100
Baines, BAL 1.009 108 Left Field .845 99
Molitor, TOR .956 102 Second Base .816 95
Franco, CHI .913 98 Catcher .766 89
Tartabull, NY .870 93 Shortstop .745 87
Murray, CLE .851 91
Dawson, BOS .850 91 Average: .857
Winfield, MIN .777 83
Harper, MIL .624 67 Note -- this is for regulars
Livingstone,DET .561 60 only, not all players.
Average DH: .935
Best PNOPS - AL Pos PNOPS Best OPS* - AL Pos OPS*
=============== ===== ===== ============== ===== =====
O'Neill, NY RF 156 O'Neill, NY RF 1.477
Jefferson, SEA DH 151 Jefferson, SEA DH 1.415
Karkovice, CHI C 142 Thomas, CHI 1B 1.263
Griffey, SEA CF 138 Palmiero, BAL 1B 1.237
Thomas, CHI 1B 132 Vaughn, BOS 1B 1.230
Whitaker, DET 2B 132 Clark, TEX 1B 1.187
Palmiero, TEX 1B 130 Griffey, SEA CF 1.180
Belle, CLE LF 129 Carter, TOR RF 1.149
Vaughn, BOS 1B 129 Davis, CAL DH 1.126
Naehring, BOS 2B 128 Buhner, SEA RF 1.093
Steinbach, OAK C 128 Belle, CLE LF 1.091
Cooper, BOS 3B 127 Karkovice, CHI C 1.086
Lofton, CLE CF 125 Cooper, BOS 3B 1.083
Macfarlane, KC C 125 Berroa, OAK DH 1.075
Clark, TEX 1B 124 Whitaker, DET 2B 1.074
Worst PNOPS-AL Pos PNOPS Worst OPS*-AL Pos OPS*
============= ===== ====== ============= ===== ======
Turner, CAL C 45 Turner, CAL C .348
Livingstone,DET DH 60 JsValentin, MIL SS .525
Coleman, KC LF 63 Coleman, KC LF .534
Harper, MIL DH 67 Walbeck, MIN C .556
T.Martinez, SEA 1B 69 Livingstone,DET DH .561
JsValentin, MIL SS 70 Wilson, SEA C .585
Blowers, SEA 3B 71 Valle, BOS C .601
Mieske, MIL RF 72 Blowers, SEA 3B .603
Walbeck, MIN C 73 Harper, MIL DH .624
Davis, DET CF 74 Lind, KC 2B .631
Wilson, SEA C 76 Sax, OAK 2B .633
Lind, KC 2B 77 Davis, DET CF .635
Perez, CAL 1B 77 Guillen, CHI SS .635
Sorrento, CLE 1B 77 T.Martinez, SEA 1B .654
Sax, OAK 2B 78 Gallego, NY SS .656
Valle, BOS C 78
NATIONAL LEAGUE through May 8
===============
Catcher OPS* PNOPS First Base OPS* PNOPS
============ ===== ===== ============ ===== =====
Daulton, PHI 1.121 138 Bagwell, HOU 1.078 129
Hundley, NY 1.087 134 Galarraga, COL 1.011 121
Piazza, LA .900 110 Jefferies, STL .991 118
Fletcher, MON .892 110 McGriff, ATL .961 115
Lopez, ATL .870 107 Kruk, PHI .909 109
Dorsett, CIN .856 105 Karros, LA .827 99
Ausmus, SD .793 97 Morris, CIN .824 98
McGriff, STL .775 95 Destrade, FLA .818 98
Slaught, PIT .745 92 Floyd, MON .793 95
Santiago, FLA .730 90 Staton, SD .786 94
Wilkins, CHI .688 85 Segui, NY .760 91
Girardi, COL .670 82 Grace, CHI .732 87
Servais, HOU .640 79 Benzinger, SF .691 82
Manwaring, SF .632 78 Young, PIT .541 65
Average C: .814 Average 1B: .837
Second Base OPS* PNOPS Third Base OPS* PNOPS
============ ===== ===== ============ ===== =====
Kent, NY 1.146 141 Wallach, LA 1.087 123
Biggio, HOU 1.021 126 Williams, SF 1.079 122
Boone, CIN .960 118 Bonilla, NY 1.022 115
Duncan, PHI .900 111 Berry, MON 1.006 114
Mejia, COL .900 111 Fernandez, CIN .936 106
Sandberg, CHI .884 109 Buechele, CHI .917 103
Garcia, PIT .801 99 Zeile, STL .877 99
Lemke, ATL .768 95 Magadan, FLA .871 98
Lansing, MON .732 90 Pendleton, ATL .844 95
Barberie, FLA .731 90 King, PIT .801 90
Roberts, SD .658 81 Hollins, PHI .796 90
Thompson, SF .658 81 Caminiti, HOU .744 84
Alicea, STL .636 78 Cianfrocco, SD .723 82
DeShields, LA .570 70 Hayes, COL .709 80
Average 2B: .812 Average 3B: .887
Shortstop OPS* PNOPS Left Field OPS* PNOPS
============ ===== ===== ============ ===== =====
Cedeno, HOU .980 139 Mitchell, CIN 1.110 120
Belliard, ATL .815 115 Klesko, ATL 1.107 119
Vizcaino, NY .783 111 Alou, MON 1.058 114
Abbott, FLA .783 111 Rodriguez, LA .978 105
Clayton, SF .772 109 Bonds, SF .973 105
Bell, PIT .753 107 Plantier, SD .962 104
Cordero, MON .748 106 Conine, FLA .961 104
Weiss, COL .746 106 May, CHI .940 101
Offerman, LA .723 102 Martin, PIT .932 100
Larkin, CIN .717 102 Thompson, PHI .896 97
Dunston, CHI .619 88 Gonzalez, HOU .841 91
Smith, STL .593 84 Gilkey, STL .794 86
Gutierrez, SD .508 72 McReynolds, NY .785 85
Batiste, PHI .343 49 Johnson, COL .646 70
Average SS: .706 Average LF: .927
Center Field OPS* PNOPS Right Field OPS* PNOPS
============ ===== ===== ============ ===== =====
Burks, COL 1.370 148 Sheffield, FLA 1.274 147
Lankford, STL 1.205 130 Bichette, COL 1.138 131
Butler, LA 1.037 112 Gwynn, SD 1.119 129
Finley, HOU .981 106 Justice, ATL .887 102
Thompson, NY .953 103 Walker, MON .861 99
Rhodes, CHI .948 102 Sanders, CIN .858 99
Dykstra, PHI .905 98 Mondesi, LA .853 98
Sanders, ATL .896 97 Merced, PIT .815 94
Carr, FLA .849 92 Sosa, CHI .779 90
Van Slyke, PIT .827 89 Jordan, STL .763 88
Bell, SD .809 87 Mouton, HOU .710 82
Lewis, SF .766 83 McGee, SF .707 81
Kelly, CIN .735 79 Burnitz, NY .696 80
Grissom, MON .687 74 Eisenreich, PHI .696 80
Average CF: .926 Average RF: .868
By Position OPS* PNOPS
============ ===== =====
Left Field .927 109
Center Field .926 109
Third Base .887 105
Right Field .868 102
First Base .837 99
Catcher .814 96
Second Base .812 96
Shortstop .706 83
Pitcher not available
Average: .847
Note -- this is for regulars
only, not all players.
Best PNOPS-NL Pos PNOPS Best OPS*-NL Pos OPS*
============== ===== ===== ============= ===== =====
Burks, COL CF 148 Burks, COL CF 1.37
Sheffield, FLA RF 147 Sheffield, FLA RF 1.274
Kent, NY 2B 141 Lankford, STL CF 1.205
Cedeno, HOU SS 139 Kent, NY 2B 1.146
Daulton, PHI C 138 Bichette, COL RF 1.138
Hundley, NY C 134 Daulton, PHI C 1.121
Bichette, COL RF 131 Gwynn, SD RF 1.119
Lankford, STL CF 130 Mitchell, CIN LF 1.11
Bagwell, HOU 1B 129 Klesko, ATL LF 1.107
Gwynn, SD RF 129 Hundley, NY C 1.087
Biggio, HOU 2B 126 Wallach, LA 3B 1.087
Wallach, LA 3B 123 Williams, SF 3B 1.079
Williams, SF 3B 122 Bagwell, HOU 1B 1.078
Galarraga, COL 1B 121 Alou, MON LF 1.058
Mitchell, CIN LF 120 Butler, LA CF 1.037
Worst PNOPS-NL Pos PNOPS Worst OPS*-NL Pos OPS*
============= ===== ====== ============= ===== ======
Batiste, PHI SS 49 Batiste, PHI SS .343
Young, PIT 1B 65 Gutierrez, SD SS .508
DeShields, LA 2B 70 Young, PIT 1B .541
Johnson, COL LF 70 Deshields, LA 2B .570
Gutierrez, SD SS 72 Smith, STL SS .593
Grissom, MON CF 74 Dunston, CHI SS .619
Alicea, STL 2B 78 Manwaring, SF C .632
Manwaring, SF C 78 Alicea, STL 2B .636
Kelly, CIN CF 79 Servais, HOU C .640
Servais, HOU C 79 Johnson, COL LF .646
Burnitz, NY RF 80 Roberts, SD 2B .658
Eisenreich, PHI RF 80 Thompson, SF 2B .658
Hayes, COL 3B 80 Girardi, COL C .670
McGee, SF RF 81 Grissom, MON CF .687
Roberts, SD 2B 81 Wilkins, CHI C .688
Thompson, SF 2B 81
--
__________________________________________________________________________
Steve Thornton ste...@eskimo.com Seattle, Washington
__________________________________________________________________________
Cory Ridgway
>bob vesterman.
Have you ever *played* shortstop? It's not easy. Most first basemen
would make you re-think this opinion in short order. And those with
the physical abilities to handle more demanding positions are often at
first base because they are left-handed throwers.
Can you imagine Frank "de Milo" Thomas at short? He can barely handle
first.
Cory Ridgway.
Cory Ridgway
have compared a player's offensive performance measure vs. the league
average at his position. Including his own performance.
The main problem with this sort of "Position Normalizing" is just a
failing in the sample size. A single player can contribute more than
100/N% of the total offense produced at a given position in a league
of N teams. Last year Alomar and Baerga accounted for 14.4% of the
AL's 2b PA.
Cory Ridgway
unknown
What I've been working on is a position normalization based on
X runs games. So what I've done is look at how 1) teams and
2) players have done in games, for example, in which the team
was shut-out. This would be the bottom line for scoring runs
(the baseline, basically). Thus, we find out that teams have
(on average) a .375 ops in shutouts, .520 ops when they score
one run, etc. I'd like to break it down by position, but I
have a very limited number of box scores (just some old BW's).
This would be a weighted heavier to teams that get shut out
a lot, but if I were to use a sufficiently large base (many
years) I think it would average out. Any suggestions?
paul
RVES...@vma.cc.nd.edu
i'm sure it's hard. i'm also sure that joe dude off the street could
make at least some of the plays that a shortstop makes. i'm not sure
that the number of balls that a shortstop gets to that joe dude wouldn't
makes up for something like a one hundred and fifty point drop in OPS.
even if you don't believe it for first basemen, my point still stands.
how about third basemen? do you honestly believe that your typical
third baseman couldn't handle playing shortstop? sure, there's going
to be a drop in defensive ability, but at the same time there's going
to be a gain in offensive ability. i am pretty confident that the
gain more than makes up for the drop. witness cal ripken, howard
johnson, and travis fryman.
anyway, the idea that i'm getting at is that a stat like this assumes
that a shortstop who hits exactly as well as the average shortstop is
exactly as valuable as a first baseman who hits exactly as well as the
average first baseman. that is, the gain in offense by putting a typical
first baseman at short exactly counters the decrease in defense. i have
a hard time believing that this is true. couldn't it be possible that
they wouldn't exactly counter each other? if you agree that yes, it is
possible that they wouldn't exactly counter, wouldn't you agree that
it is also possible that the positive value of the gained offense and
the negative value of the lost defense might not even be close to each
other? after all, the only reason to believe that they are close is if
you believe that managers and scouts and such didn't make the decision
of how much offense is worth how much defense arbitrarily. if you
believe that decision is not arbitrary, i'd like you to point out
a coach who ever tried to test his decision.
bob vesterman.
Steven Thornton
> First basemen and rightfielders outhit designated hitters -- not too
> uncommon, since DHs are rarely the best hitters in the league. But
> *any* first baseman and *any* right fielder can play DH as well as any
> current DH. Under this particular adjustment, any right fielder or
> first basemen could improve their "value" by playing DH -- that's not
> right. Likewise, if Reggie Jefferson started playing RF (assuming average
> defense) his value would go down -- that's not right. I think the only
> good way to deal with the (abomination of) DH is to compare them to the
> best group of hitters.
Yeah, well I don't know what the heck to do with DH numbers. If I was
seriously trying to put this forth as some sort of "MVP evaluation tool",
I'd have to handicap them somehow -- a DH would have to be Babe Ruth to
be seriously considered. What I did was, just treat them like any other
position and let the chips fall. We could of course start up a thread
about how managers have rarely used the DH slot the way it "logically"
should be used. I'm totally uninterested in whether the DH is a "good"
thing or not; it exists, and the point should be to make use of the rules
to one's advantage. But few people do.
> A related consideration is outfielders -- how different are the outfield
> spots? Traditionally, centerfield is considered the most difficult, and
> I would guess that anyone who can play center can play left. If true,
> a .854 OPS* centerfielder *should* be more valuable than a .845 leftfielder,
> since we know he *could* be an .854 OPS* leftfielder. (We don't know he
> could be a .854 OPS* infielder or catcher, though.)
I'm just sort of presuming that the positions will sort themselves out in
order of defensive difficulty. Whether they do or not is a different
question. But I don't understand this -- a .854 CF should be, and is,
more "valuable", i.e, has a higher PNOPS, than a .854 RF, most of the
time. Another thing to consider is that my listings include a fair number
of bench players, in cases where the regular is hurt. Since a lot of
teams end up subbing a total stiff, a few of those can really screw the
positional average. This should presumably shake out over the year, since
a lot of the hurt guys will return, and teams will at least be trying to
solve the problem cases where he can't.
If I understand your general drift, you're right, though -- a team could
maximize their PNOPS by playing their nine best hitters, even if they are
all DH-types, which would look nice on my list but suck pretty bad on the
field. Frank Thomas would be an interesting shortstop....but I sort of
allow the natural course of things to take care of this -- no team is
going to allow that to happen. Most players are at least plausible at
their position.
> considerations, but when firstbasemen get a slight plus for being able
> to handle first I get a little wary.
I don't see how 1B get a "plus" for "being able to handle
first". As mentioned, any temporary blip that causes 1B to have a lower
average OPS* than some other position will presumably shake out over
time. And if it doesn't, that suggests to me that some managers are
playing people out of position. Logically, comparing 1B to other 1B makes
perfect sense to me. If the numbers don't agree with the true defensive
spectrum, it's still interesting.
Thanks for your comments, though. My original remarks about "the be-all
and end-all of statistics" was facetious; I was just interested in a
quick look at an intuitive notion.
Steven Thornton
> i'm sure it's hard. i'm also sure that joe dude off the street could
> make at least some of the plays that a shortstop makes. i'm not sure
> that the number of balls that a shortstop gets to that joe dude wouldn't
> makes up for something like a one hundred and fifty point drop in OPS.
Speaking from personal experience, I am absolutely certain that I
couldn't handle 0.01% of the plays at MLB-level SS. I can't handle any of
the plays at office-party softball, either. I'd have a real good fielding
percentage, though -- wouldn't get close enough to anything to get
charged with an error.
> anyway, the idea that i'm getting at is that a stat like this assumes
> that a shortstop who hits exactly as well as the average shortstop is
> exactly as valuable as a first baseman who hits exactly as well as the
> average first baseman. that is, the gain in offense by putting a typical
> first baseman at short exactly counters the decrease in defense. i have
> a hard time believing that this is true. couldn't it be possible that
> they wouldn't exactly counter each other? if you agree that yes, it is
> possible that they wouldn't exactly counter, wouldn't you agree that
> it is also possible that the positive value of the gained offense and
> the negative value of the lost defense might not even be close to each
> other? after all, the only reason to believe that they are close is if
My intuitive response is that the defensive edge is much _greater_ than
the offensive decrease -- the opposite of yours. I don't have them in
front of me, but I recall from the DA reports that a good SS saves enough
hits over a bad SS to make up for more offense than any player in the
league. And the worst SS in MLB history is a lot better than a bad
defensive 1B would be at short. So I don't buy your argument.
On the other hand, I absolutely agree that my PNOPS is illogically
assuming that managers are smart. No manager is dumb enough to try Frank
Thomas at short, but there are lots of players who might be better off
somewhere other than where they're playing. I can't really help that,
though; all I can do is assume that the problem will be at least
partially self-correcting. It may take a while -- see how long it took
Minnesota to move Kirby Puckett out of center -- but overall I don't
think it's grotesquely wrong to compare players this way.
Steven Thornton
Well, I'm not a real statistician. I used "average" as you describe
because it makes sense in layman's terms, which is what I am. This also
means I don't quite get what you're saying (through no fault of yours).
let me see if I can figure it out.
"A single player contributing more than 100/N% of the total offense" is
exactly what I was looking for. Do you mean that the stat, being a rate
stat and not a cumulative one, isn't giving enough weight to players with
a lot of playing time? That's true. It's that way for a couple of
reasons, not least of which is plain laziness -- I was looking for a
quick-look kind of thing. More importantly, I wanted a reasonable way to
include part-time substitutes for injured players. If a sub comes in and
tears up the league for a week or two until Joe Superstar comes off the
DL, I wanted to give him credit for it, temporarily. When Joe comes back,
the sub falls off the list. I could have done this by just using stats
from the past week, but I don't have them (I'm not going to collate all
the box scores). So I used a rule which selects the team leader in
innings at the position, who was also active that week. by the end of the
season, most teams will have a player fitting that description at each
position with a substantial number of innings. Not an equal number,
you're right. Eventually, a number that incorporates PA would be more
appropriate, I suppose, if that doesn't mean too much work. For now you
get "freak" results, like Reggie Jefferson, and admittedly platoon
players are handled badly (Paul O'Neill, several Tigers, etc.) It's not
the end of the world.
Maybe I should just plug in a cumulative stat like Linear Weights or Runs
Created into a positional table. I was looking for something quicker, though.
RVES...@vma.cc.nd.edu
>
>Speaking from personal experience, I am absolutely certain that I
>couldn't handle 0.01% of the plays at MLB-level SS.
oh, bull. that's a play every fifteen or twenty years, man.
i freely admit that i may be underestimating how hard it is to make
a routine play at short, but you are GROSSLY overestimating. you're
telling me that you can't grab a weak grounder that's hit near you
and then throw it? frank thomas can certainly do it, and look at all
the flak that he's getting in this thread.
bob vesterman.
Cory Ridgway
>In article <csr.76...@crash.cts.com>, c...@crash.cts.com (Cory Ridgway) says:
>>
>>In <94133.1618...@vma.cc.nd.edu> <RVES...@vma.cc.nd.edu> writes:
>i'm sure it's hard. i'm also sure that joe dude off the street could
>make at least some of the plays that a shortstop makes. i'm not sure
>that the number of balls that a shortstop gets to that joe dude wouldn't
>makes up for something like a one hundred and fifty point drop in OPS.
It would, and then some. The consensus is that the worst major leaguer
at short generally costs his team about 30 runs a year. If you think
some schmoe off the street can play short as well as the worst major
league regular your nuts.
If you doubt me. Find out if there is an amatuer league near you.
And go watch some of the lower level games. See how many unearned runs
are scored per game. And remember these are the Joe Dudes that *can*
play. If there isn't any amatuer league around check out a community
college game. Even at Princenton it was quite rare to see a game that
didn't have a couple of unearned runs scoring. And that's just the
effect of the bad plays that are called errors.
>even if you don't believe it for first basemen, my point still stands.
>how about third basemen? do you honestly believe that your typical
>third baseman couldn't handle playing shortstop? sure, there's going
>to be a drop in defensive ability, but at the same time there's going
>to be a gain in offensive ability. i am pretty confident that the
>gain more than makes up for the drop. witness cal ripken, howard
>johnson, and travis fryman.
Typical third basemen?
Fryman had played a total of 0 games at 3b before reaching the majors.
Johnson was the only one of these 3 who was originally a 3b. And he
has never played more games at short than at 3b in a season. He had
997 G at 3b and 272 at ss coming into this year.
OBP/SLG by position, league, and year (from '92-94 GABS):
'91 '92 '93
AL NL AL NL AL NL
P/DH n/a n/a 354/409 166/170 332/426 182/185
C 307/374 305/352 309/369 307/357 324/396 317/400
1B 344/417 341/411 344/417 342/422 368/469 352/449
2B 338/365 334/371 336/361 326/361 339/370 334/386
3B 328/388 324/408 328/385 329/397 335/399 336/432
SS 310/352 314/361 313/335 315/340 318/358 338/365
LF 331/398 330/410 339/401 320/400 344/430 344/452
CF 332/405 330/378 321/395 341/398 339/410 341/408
RF 325/436 334/422 328/400 324/399 335/417 338/441
PH n/a n/a 300/285 296/314 326/349 304/330
Lge 329/395 317/373 328/385 315/368 337/408 327/399
>anyway, the idea that i'm getting at is that a stat like this assumes
>that a shortstop who hits exactly as well as the average shortstop is
>exactly as valuable as a first baseman who hits exactly as well as the
>average first baseman. that is, the gain in offense by putting a typical
A stat doesn't assume anything. Apparently the reader assumed it.
Comparing "position normalized" stats for 2 different positions is
similar to comparing stats for players from different ERAs. Normalizing
for the league can help to illustrate the player's performance in
context. As can normalize for position.
Cory Ridgway
Cory Ridgway
You lost me. This sounds like a pretty dicey idea. How about a
little more detail. A line-up of .375 OPS hitters WILL score. A
line-up of .520 OPS hitters will average more than 1 run per game
(2.0-2.5). In fact a line-up of .375 OPS hitters should average
about 1 run per game. Just watch the Padres for a while and you'll
see what I mean :)
What are you doing?
Cory Ridgway
Ira K. Blum
|> In <CpoyG...@eskimo.com> ste...@eskimo.com (Steven Thornton) writes:
|> [...Position Normaliztion...]
|>
There is one thing about DH's different managers use them in different
ways. If the manager uses the DH to rest players who would otherwise be
playing in the field, then they would produce the same numbers as
otherwise. Making a player a full-time DH is difficult. Does anyone
remember Jorge Bell? Most of the best offensive players in the league
take their pride in their defense (no matter how good or bad it really
is.) Griffey, Bonds, Thomas, Fielder. It matters not whether they are
good fielders or bad, they still want to play the field. If Lou Pinella
came up to Ken Griffey and said, "I don't think your defense is good
enough, why don't you DH for the rest of your career." What do you think
Griffey would do? What would Bonds or Thomas, etc do? It is a rare thing
for a player to produce well at DH since most full time DH's are on the
down sides of their careers and most manager would rather get rid of a man
one year to late than one year to early.
--
Ira
ib...@utdallas.edu
Go Rangers and Phillies (and Cowboys and Mavericks and Speed Racer Go!)
Will Clark for MVP!!!!!
A tisket, a tasket
You'll all end up in a casket,
I pull the thread
And cut off your head
It rolls into a basket.
Please direct all flames to /dev/null
Neal Traven
> In article <Cpzon...@eskimo.com>, ste...@eskimo.com (Steven Thornton) says:
> >
> >Speaking from personal experience, I am absolutely certain that I
> >couldn't handle 0.01% of the plays at MLB-level SS.
> oh, bull. that's a play every fifteen or twenty years, man.
Sounds about right to me. I've met Steve. ;^)
> i freely admit that i may be underestimating how hard it is to make
> a routine play at short, but you are GROSSLY overestimating. you're
> telling me that you can't grab a weak grounder that's hit near you
> and then throw it? frank thomas can certainly do it, and look at all
> the flak that he's getting in this thread.
That's exactly what Steve is saying. If he catches the grounder without
fumbling it, he won't be able to throw it accurately to 1B. Or as far
as 1B.
Now, on a softball field, he might get up to a .100 fielding percentage.
And a 0.5 range factor. And a .200 DA.
Maybe...
--
-----------------------------------------------------------------------------
neal tra...@pitt.edu You're only young once, but you can be
tra...@vms.cis.pitt.edu immature forever. -- Larry Andersen
unknown
I deleted my first post. All relevent details are below.
Sorry, I know it was sketchy. i'll put in a few more details
below.
> You lost me. This sounds like a pretty dicey idea. How about a
>little more detail. A line-up of .375 OPS hitters WILL score.
On what do you base this claim? That was my initial question: How
bad do you have to be to be shut out. I looked at 60 shutouts
(small sample size, sorry. About 2000 PA total). The average OPS
was .375. Thus, teams with a .375 OPS didn't score _at all_. Zip.
Thus, .375 is what I consider the team baseline (I didn't check it
against RC...I should do that).
A
>line-up of .520 OPS hitters will average more than 1 run per game
>(2.0-2.5).
And when I looked at games in which the teams scored 1 run, the
ops was .520. Yes, a .520 OPS gave them 1 rpg.
One of the reasons I did this was to test an OPS/RC model I've been
working on. What I proposed was a model where Runs = A*PA*(OPS - C),
where A = runs/change_in_ops and C is the baseline in which no runs
result. When I ran a linear regression against my team data base
I got A=.28 and C = .300. This indicates that a team with a .300 OPS
would not score, a team with a .400 OPS would score about 0.9 runs/game
(using game = 32 PA for a low ops team). Similarly, a .520 OPS would be
2-2.5 runs/game as you say.
So the model I set forth was a reasonable approximation of RC. I liked
it because 1) I did it myself, and 2) I don't remember the RC formula
(what is it? obp*TB?). 3) I also like the finding of a baseline ops.
Thus, I can say that a .800 ops is twice as good as a .550 ops.
So after finding the model, I thought I'd test it. I started with 0 runs
/game. Where i'd expect to see the .300 that I got from the linear
regression, the actual ops I found was .375 (obp = .200, slg = .175).
A .375 OPS according to my model should give .6 rpg.
So then I looked at 1 run games. The expected OPS would be .402 - .419.
I found .520.
Now, I think I have figured out what's going on. Basically, run scoring
isn't linear when you are talking very little scoring. The linear model
holds well when you are in the region of 'normal' scoring, ie > 3.5 rpg
or so, but the linearity breaks down when you get too low. I will continue
to work on this question, as to when does linearity begin. I will post
what I find when I find it.
Now, a problem here is that I just don't have enough box scores to
do a really good study. Thus, results should be evaluated in that
light. However, 60 games is an OK set to look at for teams (2000 PA).
For individual players, however, it's not near enough. What I want
to do is the same analysis using runs = sum(9 positions + ph'ers).
I'd like to try to get 'baseline' values for each position. However,
this will take _lots_ of boxscores and way too much time if I
keep doing it the way I am now (laying on the floor with old BW).
paul
In fact a line-up of .375 OPS hitters should average
>about 1 run per game. Just watch the Padres for a while and you'll
>see what I mean :)
> What are you doing?
Hope this helps.
>
>Cory Ridgway
David Grabiner
>> You lost me. This sounds like a pretty dicey idea. How about a
>> little more detail. A line-up of .375 OPS hitters WILL score.
> On what do you base this claim? That was my initial question: How
> bad do you have to be to be shut out. I looked at 60 shutouts
> (small sample size, sorry. About 2000 PA total). The average OPS
> was .375. Thus, teams with a .375 OPS didn't score _at all_. Zip.
> Thus, .375 is what I consider the team baseline (I didn't check it
> against RC...I should do that).
There are two questions you could be asking here. You might ask, "How
many runs will a team with a .375 OPS score?" or "What is the average
OPS of teams when they score no runs?" These two will not give the same
answer, because there is an inherent bias. I think what you want is the
first question.
If you look at all the games in which one team had an OPS near .375, you
will find some shutouts, and some non-shutouts such as three-hitters in
which one of the hits was a home run. And a team with an OPS of .375
would do still better, because they will score more runs in a .575 game
and a .175 game than in two .375 games.
If your model for runs scored is OBP*SLG, then a team with a .375 OPS
will score 1/4 as many runs as a team with a .750 OPS. That's more than
one run per game.
> One of the reasons I did this was to test an OPS/RC model I've been
> working on. What I proposed was a model where Runs = A*PA*(OPS - C),
> where A = runs/change_in_ops and C is the baseline in which no runs
> result. When I ran a linear regression against my team data base
> I got A=.28 and C = .300. This indicates that a team with a .300 OPS
> would not score, a team with a .400 OPS would score about 0.9 runs/game
> (using game = 32 PA for a low ops team). Similarly, a .520 OPS would be
> 2-2.5 runs/game as you say.
This also predicts that a team with a .200 OPS will score a negative
number of runs. As you realized, there is a problem in trying to apply
a linear regression to a system which is non-linear. It does make sense
to use a linear approximation when the OPS's are in a small range such
as the typical .700-.750
> So the model I set forth was a reasonable approximation of RC. I liked
> it because 1) I did it myself, and 2) I don't remember the RC formula
> (what is it? obp*TB?).
That's the basic RC formula.
> 3) I also like the finding of a baseline ops.
> Thus, I can say that a .800 ops is twice as good as a .550 ops.
For individual players, I think a better baseline is replacement level.
If replacement level at first base is .700, so that a first baseman at
that level has no value, then a first baseman who is .900 in 300 PA is
just as valuable as a first baseman who is .800 in 600 PA.
--
David Grabiner, grab...@zariski.harvard.edu
"We are sorry, but the number you have dialed is imaginary."
"Please rotate your phone 90 degrees and try again."
Disclaimer: I speak for no one and no one speaks for me.
unknown
>
>There are two questions you could be asking here. You might ask, "How
>many runs will a team with a .375 OPS score?" or "What is the average
>OPS of teams when they score no runs?" These two will not give the same
>answer, because there is an inherent bias. I think what you want is the
>first question.
>
Actually, I want both. And I want to rationalize why they aren't
the same (the inherent bias). I have an idea, I just haven't sat
and tried to think it through.
>If you look at all the games in which one team had an OPS near .375, you
>will find some shutouts, and some non-shutouts such as three-hitters in
>which one of the hits was a home run. And a team with an OPS of .375
>would do still better, because they will score more runs in a .575 game
>and a .175 game than in two .375 games.
>
>If your model for runs scored is OBP*SLG, then a team with a .375 OPS
>will score 1/4 as many runs as a team with a .750 OPS. That's more than
>one run per game.
Assuming it stays linear in that region. Which is not necessarily
true.
>
>> One of the reasons I did this was to test an OPS/RC model I've been
>> working on. What I proposed was a model where Runs = A*PA*(OPS - C),
>> where A = runs/change_in_ops and C is the baseline in which no runs
>> result. When I ran a linear regression against my team data base
>> I got A=.28 and C = .300. This indicates that a team with a .300 OPS
>> would not score, a team with a .400 OPS would score about 0.9 runs/game
>> (using game = 32 PA for a low ops team). Similarly, a .520 OPS would be
>> 2-2.5 runs/game as you say.
>
>This also predicts that a team with a .200 OPS will score a negative
>number of runs. As you realized, there is a problem in trying to apply
>a linear regression to a system which is non-linear. It does make sense
>to use a linear approximation when the OPS's are in a small range such
>as the typical .700-.750
>
This is true for the standard RC model as well, No?
>> So the model I set forth was a reasonable approximation of RC. I liked
>> it because 1) I did it myself, and 2) I don't remember the RC formula
>> (what is it? obp*TB?).
>
>That's the basic RC formula.
>
>> 3) I also like the finding of a baseline ops.
>> Thus, I can say that a .800 ops is twice as good as a .550 ops.
>
>For individual players, I think a better baseline is replacement level.
>If replacement level at first base is .700, so that a first baseman at
>that level has no value, then a first baseman who is .900 in 300 PA is
>just as valuable as a first baseman who is .800 in 600 PA.
>
But even a team of replacement level players will score some runs.
A 'zero-point' baseline would give the 'absolute ' value of even
replacement level. I think it's worthy to think about it for that
reason.
p
Cory Ridgway
>On Tue, 17 May 1994 07:14:10 GMT, Cory Ridgway (c...@crash.cts.com) wrote:
>> I didn't check to see but based upon your description you appear to
>> have compared a player's offensive performance measure vs. the league
>> average at his position. Including his own performance.
>> The main problem with this sort of "Position Normalizing" is just a
>> failing in the sample size. A single player can contribute more than
>> 100/N% of the total offense produced at a given position in a league
>> of N teams. Last year Alomar and Baerga accounted for 14.4% of the
>> AL's 2b PA.
>Well, I'm not a real statistician. I used "average" as you describe
>because it makes sense in layman's terms, which is what I am. This also
>means I don't quite get what you're saying (through no fault of yours).
>let me see if I can figure it out.
>"A single player contributing more than 100/N% of the total offense" is
>exactly what I was looking for. Do you mean that the stat, being a rate
>stat and not a cumulative one, isn't giving enough weight to players with
>a lot of playing time? That's true.
I used the term "offensive performance" not "offense". You cut the
sentence in the wrong place. I should have just written PA but I wanted
to include performance measures like EQA which have SB and CS built into
their denominators for rate of performance.
I simply meant that the mean of the performance measure of a sample of
14 players is NOT a very good estimate of what that performance measure
would be for an average player at that position. What I was pointing out
is that it is somewhat unfair to compare an AL second baseman to a group
that is comprised in a large part (14.4%) of Roberto Alomar and Carlos
Baerga.
This "everybody's good" problem of normalizing will occur quite freq.
because the sample is too small for random distribution of talent by
position to even out. Consider the quality of 2bs in general right now;
Alomar, Baerga, DeShields, Sandberg, Biggio, Kent, Boone, ... There are
a load of good hitting second basemen right now. And second base hasn't
become easier to play in the last 5 years. Or at least I don't think it
has. A similar problem of position normalization can be found in the
defense of AL cfs right now. Look at the quality of the gloves; White,
Lofton, McRae, Griffey, Johnson, Hulse, Williams, ...
Cory Ridgway
Steven Thornton
> I simply meant that the mean of the performance measure of a sample of
> 14 players is NOT a very good estimate of what that performance measure
> would be for an average player at that position. What I was pointing out
> is that it is somewhat unfair to compare an AL second baseman to a group
> that is comprised in a large part (14.4%) of Roberto Alomar and Carlos
> Baerga.
True. However, I don't have stats for _everyone_ at second base. Also, it
would be equally unfair to include players who have a little 2B but
normally play somewhere else, say 3B. There's no way for me to get
numbers for every 2B accumulated only while they were actually playing
2B. Third, I hope I included in my caveats that I am only looking at this
as "average REGULAR second baseman" (etc), which is not a hopelessly
insignificant thing to look at. And lastly, there are some pretty stiff
regular players in my lists, too, to balance against Alomar and Baerga.
In general, I'm afraid I have to dispute your contention that "the mean
of the performance measure of a sample of 14 players" isn't very good --
these aren't just any 14, they're the 14 regulars, and most of the time,
they will account for the vast majority of total league innings at that
position.
> This "everybody's good" problem of normalizing will occur quite freq.
> because the sample is too small for random distribution of talent by
> position to even out. Consider the quality of 2bs in general right now;
> Alomar, Baerga, DeShields, Sandberg, Biggio, Kent, Boone, ... There are
> a load of good hitting second basemen right now. And second base hasn't
> become easier to play in the last 5 years. Or at least I don't think it
It's not easier to play. But if there are a lot of good-hitting 2B around
right now, it _has_ gotten harder to crack a lineup. If the standard for
2B hitting goes up, that means that what was good enough two years ago
isn't good enough anymore. I don't see what's wrong with that, and I
think my listings show that. One of the questions I was trying to answer
is, "is MY team's 2B above-average as a hitter, or is he replaceable?" If
the standard for 2B hitting goes up, and my guy doesn't, he _is_ in fact
less valuable relative to the others. After all, that's the competition.
If they're getting better, you'd be well-advised to start getting better too.
Cory Ridgway
>In article <csr.76...@crash.cts.com> c...@crash.cts.com (Cory Ridgway) writes:
>> You lost me. This sounds like a pretty dicey idea. How about a
>>little more detail. A line-up of .375 OPS hitters WILL score.
>On what do you base this claim? That was my initial question: How
>bad do you have to be to be shut out. I looked at 60 shutouts
>(small sample size, sorry. About 2000 PA total). The average OPS
>was .375. Thus, teams with a .375 OPS didn't score _at all_. Zip.
>Thus, .375 is what I consider the team baseline (I didn't check it
>against RC...I should do that).
Very sloppy logic. First; a line-up of X OPS hitters will not
always produce an OPS of X for a given game. Second; you looked at
shutouts and found that the average OPS of those teams that were shut
out was .375. That's not the same as saying that no team that produced
an OPS of .375 or less for a game avoided getting shut out. In fact
just last week I watched a game in which the Dodgers had an OPS of
.269 (a Raul Modesi solo shot followed by a Jose Offerman single in
26 PA). Needless to say they were not shut out. In fact Pedro Astacio
shut out the Astros and the Dodgers won 1-0. I believe in Matt Young's
no-hitter loss he allowed an OPS of less than .200 and 2 runs.
A team with a .375 OPS would score about 1 run per game. If the
correlation between RC and EQR and Runs holds up at such an extreme
low level of production.
> A
>>line-up of .520 OPS hitters will average more than 1 run per game
>>(2.0-2.5).
>And when I looked at games in which the teams scored 1 run, the
>ops was .520. Yes, a .520 OPS gave them 1 rpg.
To find out how many RPG a .520 OPS produces you need to take the
average (or whatever) of the runs scored in games in which a team had
a .520 OPS (or say .470-570 OPS). What you've calculated is the
expected OPS for games in which N runs were scored. The expected runs
for games in which a team had a given OPS cannot be easily calculated
from this.
>One of the reasons I did this was to test an OPS/RC model I've been
>working on. What I proposed was a model where Runs = A*PA*(OPS - C),
>where A = runs/change_in_ops and C is the baseline in which no runs
>result. When I ran a linear regression against my team data base
>I got A=.28 and C = .300. This indicates that a team with a .300 OPS
>would not score, a team with a .400 OPS would score about 0.9 runs/game
>(using game = 32 PA for a low ops team). Similarly, a .520 OPS would be
>2-2.5 runs/game as you say.
OPS does have a strong linear correlation with runs scoring. However
you must remember that it is only a fitted curve. There is no magic
involved in mathematical interpolation. Here's a question to illustrate
my point; how many runs would a team with a .200 OPS be expected to score?
About -0.8 R/G according to your linear equation. How the heck are they
going to do that?
I once thought that R/G would be distributed in something approaching
a Poisson distribution with a lambda of 4. I was dead wrong. This would
have an average team scoring 10+ runs only once a season. And getting
shut-out about 3 times a year. A friend of mine once said he fitted R/G
to a negative binomial distribution. But I never got the p and r values
from him.
>So the model I set forth was a reasonable approximation of RC. I liked
>it because 1) I did it myself, and 2) I don't remember the RC formula
>(what is it? obp*TB?). 3) I also like the finding of a baseline ops.
>Thus, I can say that a .800 ops is twice as good as a .550 ops.
But it isn't. A .550 OPS is easy to replace and has very little value.
Even attached to a glove like Jose Lind's. Would you say Ryne Sandberg
is twice as valuable a hitter as Jose Lind is? You've also left out the
fact that players with higher OPSs tend to have higher OBPs and so will
come to the plate more times per game. In fact you must divide the
OPS-baseOPS by (1-OBP) to set a level above baseline. Since a team has
a given number of outs, not PA.
>So after finding the model, I thought I'd test it. I started with 0 runs
>/game. Where i'd expect to see the .300 that I got from the linear
>regression, the actual ops I found was .375 (obp = .200, slg = .175).
>A .375 OPS according to my model should give .6 rpg.
>So then I looked at 1 run games. The expected OPS would be .402 - .419.
>I found .520.
See above.
>Now, I think I have figured out what's going on. Basically, run scoring
>isn't linear when you are talking very little scoring. The linear model
>holds well when you are in the region of 'normal' scoring, ie > 3.5 rpg
>or so, but the linearity breaks down when you get too low. I will continue
>to work on this question, as to when does linearity begin. I will post
>what I find when I find it.
A linear curve fit breaking down for extreme values? Unheard of :-)
Actually I think more of the error comes from the fact that it is not
possible to score negative runs.
Cory Ridgway
Cory Ridgway
>On 20 May 94 22:41:20 GMT, Cory Ridgway (c...@crash.cts.com) wrote:
>> I simply meant that the mean of the performance measure of a sample of
>> 14 players is NOT a very good estimate of what that performance measure
>> would be for an average player at that position. What I was pointing out
>> is that it is somewhat unfair to compare an AL second baseman to a group
>> that is comprised in a large part (14.4%) of Roberto Alomar and Carlos
>> Baerga.
>insignificant thing to look at. And lastly, there are some pretty stiff
>regular players in my lists, too, to balance against Alomar and Baerga.
>In general, I'm afraid I have to dispute your contention that "the mean
>of the performance measure of a sample of 14 players" isn't very good --
>these aren't just any 14, they're the 14 regulars, and most of the time,
>they will account for the vast majority of total league innings at that
>position.
The problem isn't the number of PA. You are weighting each player's
OPS by PA/AB (I assume your calculation the OPS of the total stats for
the players). The problem is that there are only 14 players. In which
case it is possible, and in fact quite common, for a single outlier
value (weighted by PA) to skew the curve. In the case of the AL the
best two players had the most PA as well. In fact, this is generally
true; the best players get the most playing time. A significant bias.
Consider this, the expected extreme value for a distribution with 50
samples is about 2.24 standard deviations from the mean (using Blom's
formula). With about 2% of the weight of the population it doesn't
skew the curve very much. With 14 samples the expected extreme value
will be about 1.7 std. dev. from the mean and carry over 7% of the
weight of the population. This is why robust estimators were developed.
The mean performance at a position in the majors seems to me to be
less significant to relative player value than the mean performance at
that position in AAA.
That the measured mean league and park adjusted offensive performance
of the top 28 players at a position has gone up does not imply that the
measured performance for all players (and most significantly, the
replacement players) at that position has similarly increased.
Cory Ridgway
Steven Thornton
> The problem isn't the number of PA. You are weighting each player's
> OPS by PA/AB (I assume your calculation the OPS of the total stats for
> the players). The problem is that there are only 14 players. In which
> case it is possible, and in fact quite common, for a single outlier
> value (weighted by PA) to skew the curve. In the case of the AL the
> best two players had the most PA as well. In fact, this is generally
> true; the best players get the most playing time. A significant bias.
> Consider this, the expected extreme value for a distribution with 50
> samples is about 2.24 standard deviations from the mean (using Blom's
> formula). With about 2% of the weight of the population it doesn't
> skew the curve very much. With 14 samples the expected extreme value
> will be about 1.7 std. dev. from the mean and carry over 7% of the
> weight of the population. This is why robust estimators were developed.
OK, you're bustin' me with this expected extreme value stuff. Your second
paragraph above has me utterly flummoxed. I told you
I'm not a statistician, I'm just a crank, so I'm likely to misinterpretate
most of what you've said here. Let me try.
I hadn't thought I was weighting each player's OPS by anything; if fact I
thought that was probably a flaw in the study. The best OPS* wins, even if
the player has 1/2 as many innings. If we imagine a league with no
substitution, where every team has one and only one player who plays
every inning at a position, so that all 14 SS, for example, would have
very similar amounts of playing time, would that make PNOPS better for you?
The calculation for the position totals was just a plain average of the 14
players listed, no others. If you're saying that this screws up (inflates,
in fact) the average (which is then divided into each player's individual
OPS* to get a deflated PNOPS), I agree. If you're saying that the players
who _aren't_ included are likely to be worse than the ones who are, making
the average "too high", I agree. I never intended that figure to be taken
as a true average of all players. It is an accurate statement of the
average rate of production, i.e. OPS*, of the 14 regular players, though.
That's all it is -- it's not "replacement value" or anything else. It's
just a marker, a base point.
I had to draw the line somewhere, since I don't _have_ the numbers for
_every_ player at a position. Actually, I sort of do, but you start
running into the problem of multi-position players. If a 3B plays a game
or an inning at SS, for whatever reason, that doesn't mean it's fair to
count all his stats in with the SS, too. I guess it counts in rotisserie
ball, but not for me.
As far as the 14 samples goes, it's not just any 14 -- it encompasses a
very high percentage of all Innings Played at the position.
> The mean performance at a position in the majors seems to me to be
> less significant to relative player value than the mean performance at
> that position in AAA.
I'd argue for the _top_ performance at AAA, not the mean, but basically I
agree with you. I think you're trying to get more out of PNOPS than it
was intended to give. I very much like the idea of measuring players
against some kind of positionally-adjusted fixed point. I picked the
point I did (simple average of the regulars) for convenience. To get a
better idea of "replacement value" or whatever you want, a different
point might be better, the way the RC/27 lists that Nelson Lu posted uses
RC/27 - 1, for example, but I was trying to get it a little easier and
quicker. Anything more accurate than that, and I just plain don't know
how to do it, 'cause I don't know the math. If you're going to help me
fix it, you're going to have put it in layman's terms, or go a little
slower than you probably have the time or inclination for....
> That the measured mean league and park adjusted offensive performance
> of the top 28 players at a position has gone up does not imply that the
> measured performance for all players (and most significantly, the
> replacement players) at that position has similarly increased.
True. But it does mean that a) any player who hasn't gone up has fallen
behind and hence is less valuable, and b) any potential replacement
players had better step up if they want to crack the majors. But I'm not
really looking at replacement value precisely, beyond a general
assumption that the worst regulars are probably replaceable. I wouldn't
draw a line based on PNOPS -- 90 percent of average OPS* certainly
doesn't mean that the player is truly 90 percent of a major leaguer. It's
just interesting, that's all. And it certainly does mean that he's
producing _at a rate_ less than that of a guy with a PNOPS of 91, or 104,
or whatever.
I think the real problem here is that we're reading this at different
levels. Criticisms based on actual knowledge of statistical methods are
unfortunately going to go over my head. Rsbb probably isn't the best
place to conduct a basic tutorial in statistics, but I am interested.
I'm admittedly using a layman's approach. I hadn't originally intended to
achieve perfection, mind you; I just wanted to answer the following sorts
of questions: Is Alomar producing OPS at a rate higher than most second
basemen? By how much, over the average? What about Griffey, Thomas,
Bonds, DiSarcina, Wallach, Hatcher, etc. etc. against their position?
Who's highest overall? Is Griffey producing at a rate more over average
for AL CF than Thomas is for AL 1B? I still think that's pretty much what
I did, considering the limitations of my knowledge, the facts that were
readily available to me, and the OPS stat itself.
I appreciate your comments, though; I wish I could understand them better!