Description of VORP: a new method of player valuation
Keith....@us.oracle.com
I've been developing my own methods of player value estimate on the Red Sox
mailing list (where I'm a frequent poster) over the past couple of months,
usually referred to by the tongue-twister acronym PkRPosMLV (ParK-adjusted
Replacement-POSitional Marginal Lineup Value). I'll present some of the
results in separate messages to r.s.bb.a as points of interest in the
ongoing MVP debate.
I've also decided that to market and popularize a stat like this, you need a
catchy name that (unlike PkRPosMLV) can be at least pronounced in English. :-)
Here, then, is the first pass at each player's 1995 "VORP" (Value Over a
Replacement at a Position).
The rest of this message will be a description of the VORP method, in brief:
For all players:
----------------
Each player is rated only for the time spent with a particular team -- there
are no combined entries. For example, there are two lines for David Cone --
DCone/TOR and DCone/NYY. Furthermore, each player is considered to play
only the position he most frequently appeared at (sorry, Tony Phillips). :-)
Park factors are the three year park factors from the 1995 STATS Major League
Handbook, which covers stats through the 1994 season. I have not yet gone
back and updated the park factors with Clay Davenport's park factors from
his EqA work.
For hitters:
------------
Compute the player's park-adjusted MLV relative to league average, and compute
the positional replacement level's MLV in the same number of plate appearances
relative to league average. The difference is PkRPosMLV.
MLV, or Marginal Lineup Value, computes the difference in team runs scored
by replacing an league (or position) average player with the player you
want to evaluate, by using the Runs Created approximation R ~= OBP * TB
Replacement level is set .035 points below the positional average in each of
AVG, OBP, and SLG (based on a study I did on 1994 stats of regulars vs.
backups)
Compute stolen base runs (SBR) as .3 * (SB - 2*CS)
Note: no park or positional adjustments are made to SBR, and a replacement
level hitter is considered to be a break-even base-stealer (0 SBR).
A player's VORP is SBR + PkRPosMLV. This does not include the value of his
defense relative to positional average. Designated hitters are compared
to first baseman for positional considerations.
To go beyond VORP to total player value, you'd need to adjust for above or
below average defense at a position using something like DA-based fielding
runs. I'd also knock off 10 batting runs from fulltime DH's as a defensive
adjustment (equating them to a bad fielding first-baseman).
For pitchers:
-------------
The pitcher's VORP = Pit% * IP/9 * ( (LgRA+RepLvl) - PkRA),
where
Pit% is the percentage of the defensive effort attributable to the pitcher
(assumed here to be 80%),
LgRA is the league average Run Average
PkRA is the pitcher's own park-adjusted Run Average.
RepLvl is the incremental number of runs per 9 innings a replacement level
pitcher is above the league average (here assumed to be +1.00)
Pitcher's defense (AL & NL) and hitting (NL) are not included. In the complete listing
of VORP's, I have separate entries for individual NL pitcher hitting, but have not
yet consolidated them with the pitching entries.
Relief pitchers are currently not treated differently from starting pitchers,
though future refinements may do so.
I'll post separate articles for each league, including the top 30 or so
players, and some of the worst players. I have complete results for each
league, but didn't want to post a 50K+ message to the group if there wasn't
any interest. If there's sufficient interest, I can post it on
rec.sport.baseball.analysis (or .data) and/or upload it to an FTP-site.
Feel free to post here, or email me privately if you have any questions or
comments.
NOTE: The stats I'm working on were derived from box-scores collected for
the fantasy league that I run, and do not represent the final official
league statistics. There are a few discrepancies, both in stats and name
recognition, particularly for minor players, but for the most part the totals
are accurate enough to be useful.
This is still a work in progress, so any constructive comments are welcomed.
Thanks!
---
Keith Woolner
kwoo...@oracle.com
--
* Gary Huckabay * Sports Decision Sciences, Davis, California *
* Moderator, rec.sport.baseball.analysis * base...@cedar.cic.net *
Christopher Hill
On 19 Oct 1995, kwoo...@oracle.com wrote:
> I've been developing my own methods of player value estimate
Thanks for the interesting post, Kraig. Here are a couple of questions
and comments for you about your methods.
First a minor point. I may just misunderstand your description here. As
I read it, in your method you calculate the park-adjusted Runs Created
(thank you Bill James) for a batter, and then compare his production to
what you could reasonably expect from the hypothetical player who would
replace him if he were gone. You figure the value of the replacement
player by taking making an adjustment downwards from the league average
production for all players at that position. OK, so far so good. But
you only use the league average to figure the production of the
replacement level player, not the actual player being rated, right? So
why do you talk about computing "the player's park-adjusted MLV relative
to the league average?" ^^^^^^^^
^^^^^^^^^^^^^^^^^^^^^
I feel like I'm missing something here.
I'll quote the original passage, then get back to more sustantial comments.
> For hitters:
> ------------
> Compute the player's park-adjusted MLV relative to league average, and compute
> the positional replacement level's MLV in the same number of plate appearances
> relative to league average. The difference is PkRPosMLV.
>
> MLV, or Marginal Lineup Value, computes the difference in team runs scored
> by replacing an league (or position) average player with the player you
> want to evaluate, by using the Runs Created approximation R ~= OBP * TB
>
> Replacement level is set .035 points below the positional average in each of
> AVG, OBP, and SLG (based on a study I did on 1994 stats of regulars vs.
> backups)
OK, me again. Here's a disclaimer before I go any further. I live in
Seattle, and I may have predictable bias towards the Mariners players.
I don't think that materially affects my subsequent comments (other than
that the AL MVP race interests me). You'll have to decide.
Big question number one: did you account for outs made? This is certainly
relevant to an offensive rating. It looks to me like you left outs out,
and that would make your statistic much less useful than it could be.
Consider, as an example, the two batters most often mentioned as AL MVP
candidates, Albert Belle and Edgar Martinez. As I figure it (crudely,
details below), Albert created about 151 runs, and Edgar about 154.
They're indistinguishable on that basis. But Belle used up 373 outs while
creating his runs, while Edgar made only 329 outs, a big difference. You
only get 27 outs a game, and Edgar left 44 more opportunities than Albert
for his teammates to produce runs. You can see the magnitude of that
difference by figuring runs created per 25 outs (batters make about 25
outs/game; the other two are consumed on the basepaths). A team of nine
Martinez's, figured this way, would score 12.0 runs per "game." A team of
Belles would score 10.1 per game (with the Big Hurt right behind him at
9.9). So those outs make a clear separation between the two players that
simple runs created totals would not. You could normalize your statistic
to outs instead of plate appearances to get rid of that problem.
[note: runs created figured from OBAxTB. Outs figured as AB-Hits. No
park adjustments made. I didn't have stolen base, grounded into double
play, or park adjustment data.]
Big question number two: if you are comparing value at each position, why
not be consistent and compare DH's with other DH's? What is the reason
for comparing offensive production of actual DH's with projections for
replacement first basemen? Because DH's usually look like first basemen?
Because they often are former first basemen? I think this is another
manifestation of the widespread confusion about the DH
position. Designated Hitters are not first basemen, and if you have the
data to figure what a replacement DH would hit, use it. Like it or not (I
would prefer to be rid of it), DH is a part of the game. It is a real
position, not some sort of quasi-bastard-first-base-like pseudo-position.
There is no logical reason that designated hitters's offensive production
should be lumped with that of first basemen in a statistic like yours.
And if the whole point of your rating is to compare players to their peers
at each position, you should not make any adjustment to the DH's if and
when you include defense. Their defensive responsibility is zero, and
they are all equally good at it :) Just as most first basemen would play
bad defense at third base, DH's would probably play badly if they played
at first. That is irrelevant to the discussion of how valuable they are
to their team.
Thanks again for the posts. In particular, I liked seeing how Chuck
Knoblauch stacked up, having heard so much about what a great year he had.
Sincerely,
Chris Hill
Seattle, WA
ceh...@u.washington.edu
> A player's VORP is SBR + PkRPosMLV. This does not include the value of his
> defense relative to positional average. Designated hitters are compared
> to first baseman for positional considerations.
>
> To go beyond VORP to total player value, you'd need to adjust for above or
> below average defense at a position using something like DA-based fielding
> runs. I'd also knock off 10 batting runs from fulltime DH's as a defensive
> adjustment (equating them to a bad fielding first-baseman).
--
David M. Tate
In article <Pine.OSF.3.91j.9510...@saul2.u.washington.edu>,
Christopher Hill <ceh...@u.washington.edu> wrote:
>On 19 Oct 1995, kwoo...@oracle.com wrote:
>> I've been developing my own methods of player value estimate
>Thanks for the interesting post, Kraig. Here are a couple of questions
>and comments for you about your methods.
I'm not Kraig, but I did invent MLV methods, so I'll throw in my $0.02...
>First a minor point. I may just misunderstand your description here. As
>I read it, in your method you calculate the park-adjusted Runs Created
>(thank you Bill James) for a batter, and then compare his production to
>what you could reasonably expect from the hypothetical player who would
>replace him if he were gone. You figure the value of the replacement
>player by taking making an adjustment downwards from the league average
>production for all players at that position. OK, so far so good. But
>you only use the league average to figure the production of the
>replacement level player, not the actual player being rated, right? So
>why do you talk about computing "the player's park-adjusted MLV relative
>to the league average?"
League averages are necessary to provide a context not only for the value
of a given level of production, but also for the value of that rate of
production on a team.
If I understand Kraig's article correctly, he's asking the following
question: "How many fewer runs would an average team score if I were
to replace this player with a replacement-level player at that position?".
This isn't quite the same as the original MLV definition, which considered
replacing an average player on an average team with the player under
consideration. There may be problems there; I haven't checked through
the math carefully.
>> For hitters:
>> ------------
>> Compute the player's park-adjusted MLV relative to league average, and
>> compute
>> the positional replacement level's MLV in the same number of plate
>> appearances
>> relative to league average. The difference is PkRPosMLV.
OK, there *is* a problem here. The problem is in the phrase "same number
of plate appearances". The lesser hitter won't *have* the same number of
plate appearances, because he'll use up more outs. Similarly, his teammates
won't get as many plate appearances, either.
What you need to do is compute the MLV for the player in question, over a
fixed number of games (or a fixed *proportion* of team PAs), then do the
same for the replacement-level player (who will get fewer PAs, as will his
teammates), and *then* do the subtraction to figure out the marginal value
of this player with respect to replacement level.
>> MLV, or Marginal Lineup Value, computes the difference in team runs scored
>> by replacing an league (or position) average player with the player you
>> want to evaluate, by using the Runs Created approximation R ~= OBP * TB
You can generalize this to replacing any player in any lineup with any
other player. The problem with trying to replace a replacement-level
player in an average lineup is that there don't tend to be replacement-
level players in average lineups; you have to assume that the rest of the
team is pretty damn good to bring it up to average. That's why I originally
worked only with average lineups. But, as described above, you can still
do the comparison by comparing both the player of interest and his
replacement-level sub against the league-average performance at that
position (in marginal runs).
>Big question number one: did you account for outs made? This is certainly
>relevant to an offensive rating. It looks to me like you left outs out,
>and that would make your statistic much less useful than it could be.
I agree. The point of MLV is to account explicitly for the affect of
the given player's OBP on the total team PAs, without the distortion
that plagues RC. If you assume equal plate appearances, you lose that
accounting.
>Consider, as an example, the two batters most often mentioned as AL MVP
>candidates, Albert Belle and Edgar Martinez. [...]
>A team of 9
>Martinez's, figured this way, would score 12.0 runs per "game." A team of
>Belles would score 10.1 per game (with the Big Hurt right behind him at
>9.9). So those outs make a clear separation between the two players that
>simple runs created totals would not. You could normalize your statistic
>to outs instead of plate appearances to get rid of that problem.
Not really. The problem is that the additional outs not consumed by
Edgar are (mostly) not used by Edgar; you have to look at the effect
on *team* production of those additional outs, which is a weighted
combination of what an average team would do with them and what Edgar
would do with them. Again, that's the point of MLV.
>Big question number two: if you are comparing value at each position, why
>not be consistent and compare DH's with other DH's? What is the reason
>for comparing offensive production of actual DH's with projections for
>replacement first basemen? Because DH's usually look like first basemen?
>Because they often are former first basemen?
No, because anybody can play DH. Replacement level for DH is, by
definition, the batting level of the best hitter still available after
you've put your 9 defensive players out on the field. That could be a
first baseman, or a left fielder, or a shortstop. It makes sense to
use the *best* offensive replacement level (which is first base) as an
indication of what is available at essentially no cost as a replacement
DH.
>I think this is another
>manifestation of the widespread confusion about the DH
>position. Designated Hitters are not first basemen, and if you have the
>data to figure what a replacement DH would hit, use it. Like it or not (I
>would prefer to be rid of it), DH is a part of the game. It is a real
>position, not some sort of quasi-bastard-first-base-like pseudo-position.
You've missed the point. Other offensive positions are constrained by the
need for the player to field a given position. Only people who can field
as shortstops get to bat in the shortstop's spot in the batting order.
This reduces the pool of candidates, which lowers the average level of
performance (and the replacement level). This effect occurs, in varying
degrees, at every position. It does *not* occur at DH, because nothing
is reducing the pool of candidates.
(This raises the $64000 question: why don't DHs hit better than all other
positions, on average? It's a good question, with a complex answer.)
>There is no logical reason that designated hitters's offensive production
>should be lumped with that of first basemen in a statistic like yours.
It isn't; it's lumped with *all* other positions, of which first base
happens to be the highest. Since your replacement will be the best
hitter available, it makes sense to assume that he'll be a first baseman
(on average).
>And if the whole point of your rating is to compare players to their peers
>at each position, you should not make any adjustment to the DH's if and
>when you include defense. Their defensive responsibility is zero, and
>they are all equally good at it :) Just as most first basemen would play
>bad defense at third base, DH's would probably play badly if they played
>at first. That is irrelevant to the discussion of how valuable they are
>to their team.
Not really. As Jonathan Bernstein has pointed out, the cost of a DH is
that you have to find 9 other guys who *can* field, and can't play any of
your other non-fielding sluggers. It's an opportunity cost; if I put
Howard Johnson at shortstop, I can DH Kevin Mitchell. If I DH HoJo, I
have to replace Mitchell's bat with Kevin Elster's, or some such. It's
a real cost.
--
David M. Tate, Senior Operations Research Analyst (dt...@dsava.com)
Decision-Science Applications, Arlington, Virginia. (703) 243-2500
Founding Member, Archdruid, and Cantor: Rob Deer Fan Club
Inquiries regarding the RDFC will be promptly and cheerfully ignored.
Christopher Hill
on 25 October, KEITH Woolner wrote:
> Thanks, and its Keith, not Kraig, BTW. :-)
Sorry about that, Keith.
[CH]>Big question number two: if you are comparing value at each position,
> >why not be consistent and compare DH's with other DH's?
[KW]>Using positional-offense adjustments is a proxy for directly measuring
> the difference in difficulty among the various positions on the field.
> We estimate the difference by observing the amount of offense a team is
> willing to trade for the defense.
The "positional offense reflects defensive value" argument works very
well in general. I am interested in why it does not work in the
specific case of the DH.
And it doesn't work. Or if it does, then one is forced to conclude that
AL first basemen have negative defensive responsibility, since they hit
better than DH's who have zero responsibility. (thanks for supplying the
data, Keith).
Here's the crux of the argument for *not* judging DH production by first
base standards. Value comes from scarcity. It is harder to find a
terrific hitting DH than a terrific hitting first baseman. That means
that the team with exceptional DH production will gain more than team with
a first baseman who hits equally well. I'm not saying that's the way it
should be, I'm just looking at the data and saying that's the way it is.
There is not an unlimited pool of players out there who can hit like
starting first basemen. Those who can hit like Frank Thomas seem to end
up playing first, *regardless of defensive ability*. The leftovers often
end up playing DH. I think it's useful to judge DH's on that basis,
rather than on assumed defensive inferiority.
Well, I'll wrap it up. Some day maybe I'll have the time to go look up
and analyze just why DH production doesn't match up to first basemen.
Thanks again to Dave and KEITH for the clarification of the MLV and VORP
methods, and KEITH for better describing your study of replacement level
performance.
Chris Hill
Seattle, WA
ceh...@u.washington.edu
KEITH supplied this:
> FWIW, DH's hit .277/.363/.459, or better than any other position in
> the AL or NL, except AL 1B (which is probably more due to the seemingly
> unusual number of outstanding first basemen in the AL this year). For
> comparison, NL 1B hit .278/.343/.460 (admittedly in a lower offense
> league).
Keith....@us.oracle.com
Christopher Hill <ceh...@u.washington.edu> writes:
>On 19 Oct 1995, kwoo...@oracle.com wrote:
>> I've been developing my own methods of player value estimate
>Thanks for the interesting post, Kraig. Here are a couple of questions
>and comments for you about your methods.
Thanks, and its Keith, not Kraig, BTW. :-)
>First a minor point. I may just misunderstand your description here.
>[...] why do you talk about computing "the player's park-adjusted MLV
>relative to the league average"
>^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
David Tate (whom I originally neglected to properly credit as the
inventor of MLV), already explained MLV pretty thoroughly. I'll address
some of his concerns later in this message.
>Big question number one: did you account for outs made?
Yes (with one caveat on % of PA's). Since OBP is a principal component
of the MLV calculation, a player's impact on the overall team OBP is
considered in VORP.
>Big question number two: if you are comparing value at each position,
>why not be consistent and compare DH's with other DH's?
Dave Tate addressed this one, too, but here's my take on it.
It relates back to what the purpose of a positional-adjustment is.
You have a spectrum of defensive difficulty and leverage, ranging from
the very high (catcher, shortstop) to the very low (DH, first base).
We have decent measures of intrapositional defense (DA, Range Factors,
Fielding Runs, etc) which compares players who play the same position,
and of offense, which compares all players (regardless of position) as
hitters. Since major-league quality defensive catchers and shortstops
are rare (and valuable), we are willing to trade some offense to get
better defense at these positions. What we don't have is a way of
directly measuring the difference in defensive value among the
different positions (*inter*positional comparisons vs *intra*positional
comparisons).
Using positional-offense adjustments is a proxy for directly measuring
the difference in difficulty among the various positions on the field.
We estimate the difference by observing the amount of offense a team is
willing to trade for the defense.
Back to DH's: Since DH's have no defensive responsibilities, there
is no tradeoff to be made. It therefore makes sense to compare them
against the highest offensive position, at least for starters. The
fact that the DH is often times used at the manager's option as a
"resting place" for slightly injured players to keep their bats in
the lineup doesn't change the fact that the defensive responsibilities
are lessened. A full-time DH (who plays no other position) needs to
have at least as good a bat as a typical first baseman to justify
having his job, and should not be compared to the "watered down" stats
accumulated by regulars getting an off-day from the field.
In a later message ceh...@u.washington.edu goes on to write:
>David, in Kraig's original description he said he had done a study of
^^^^^ Keith
>what the actual replacement level is at each position. That is, he
>looked at regulars who got hurt and the men who replaced them while they
>were out.
Not exactly. The methodology was more than I removed the top N players
at each position by number of plate appearances (N=14 for a 14 team
league), and calculated the averages of the remaining players, who
are a combination of backups, platooned players, midseason acquisitions,
injured regulars, and so on. On average, this "lesser" group of players
hit about 70 points worse in OPS than the league average, leading to
an proposed adjustment for replacement level of 35 points off OBP and
SLG (and AVG just for consistency). It wasn't a direct examination of
injured regulars and their replacements.
>What he found was that replacement players stats were was well
>estimated by knocking .035 off of the regular's batting average, on base
>average, and slugging percentage (as I understood Kraig. It would be nice
>to see the actual results of that study).
I only did the study for the 1994 AL and NL, and didn't think it
worthwhile enough to post. I want to examine some others years of data,
to see if its a consistent spread. Additionally, I'm interested in
the question of whether replacment level is relatively lower for
high-offense positions or vice versa. The one year totals taken
position by position were too variable to be of much good in
answering that, but when I get a chance to look across multiple years,
there may be enough data to draw some conclusions.
[in regards to whether DH's should be compared to DH's or 1B]
>What I am said is that it is always preferable to use real data when
>you have it. It better to look at what replacement DH's actually hit, if
>you have the data, than to make assumptions about what you think they
>should hit. [...]
>I still maintain that this discussion shows that DH
>still not treated as a real position. If you are happy with the VORP
>comparing first basemen with what other first basemen actually *do* hit,
>why do you suggest that DHs should be rated by what you *think* they
>should hit instead of what other DHs actually *do* hit.
>Give me data over opinion any time.
As long as the data doesn't cloud understanding. In order to decide how to
treat DH's with regard to positional considerations, you have to understand
why positional considerations enter into it to begin with. The reason,
as I see it, is what I stated above -- to act as a proxy for
interpositional defensive difficulty. Since DH's have no defensive
difficulty, you should compare them to the entire replacement population,
which means, as a first step, assessing them against the highest
offensive position that has any defensive responsibilities (first base).
FWIW, DH's hit .277/.363/.459, or better than any other position in
the AL or NL, except AL 1B (which is probably more due to the seemingly
unusual number of outstanding first basemen in the AL this year). For
comparison, NL 1B hit .278/.343/.460 (admittedly in a lower offense
league).
>What I think Kraig and David should do (sorry to pick on you guys for
>voicing what is probably a majority opinion)
No problem, as long as you promise to get my name right in the future.
:-) :-) :-)
>is get away from what they think *should* be the case with designated
>hitters, and recognize that it is apparently *harder* to fill the DH
>position, than to fill first base, with a productive hitter.
It's not *harder* to fill the DH than first base, but rather that the
manager *chooses* to use the position's unique role differently --
namely to rest slightly injured regulars from the grinds of defensive
play. Teams that have full-time DHes (Chili Davis, Jose Canseco, Paul
Molitor [pre-95], etc) tend to play hitters than are comparable to
first basemen in offensive prowess.
In another posting on the same thread dta...@pitt.edu (David M. Tate)
writes:
>If I understand Kraig's article correctly, he's asking the following
>question: "How many fewer runs would an average team score if I were
>to replace this player with a replacement-level player at that position?".
That's accurate. Furthermore, I'm treating replacement level as
meaning replacement-level offense at that position, break-even
basestealing, and positionally average defense.
>>> For hitters:
>>> ------------
>>> Compute the player's park-adjusted MLV relative to league average, and
>>> compute
>>> the positional replacement level's MLV in the same number of plate
>>> appearances
>>> relative to league average. The difference is PkRPosMLV.
>OK, there *is* a problem here. The problem is in the phrase "same
>number of plate appearances". The lesser hitter won't *have* the same
>number of plate appearances, because he'll use up more outs. Similarly,
>his teammates won't get as many plate appearances, either.
Good point, and one I am aware of. What I intend to do is give the
player and his hypothetical replacement the same percentage of
overall team PA's as the player had on his actual team. It was an
additional layer of computation I didn't have the heart to tackle just
yet, so I used the PA adjustment directly. It's not as accurate, but
not too far off for most players.
>What you need to do is compute the MLV for the player in question, over a
>fixed number of games (or a fixed *proportion* of team PAs), then do the
>same for the replacement-level player (who will get fewer PAs, as will his
>teammates), and *then* do the subtraction to figure out the marginal value
>of this player with respect to replacement level.
Basically, this means that a RepLvlMLV is a function of 10 variables:
the players's own AVG, OBP, and SLG, the league-average AVG, OBP, and
SLG, the positional replacement level AVG, OBP, and SLG, and the
player's fraction of the team's plate apperances. Yes?
[Re: problem accounting for outs made]
>I agree. The point of MLV is to account explicitly for the affect of
>the given player's OBP on the total team PAs, without the distortion
>that plagues RC. If you assume equal plate appearances, you lose that
>accounting.
Hmmm... what I had been doing is computing the MLV for an entire season
and then prorated the MLV by (Actual PA)/(Hypothetical full season PA). I can see why you'd want to prorate
I understand why you'd want to change it to the same percentage of PA's
relative to the team [scaling MLV down by (%ofTeamPA / 11.1%)], but I
don't know how large an error this simpler approximation introduces.
A few quickie examples in my MLV spreadsheet suggests its in the single
digit run area for the top players. It's worth correcting (and I'll
try to do so in future iterations), but I don't think it skews the
overall ordering too much.
---
Keith Woolner
kwoo...@oracle.com
Keith....@us.oracle.com
>well in general. I am interested in why it does not work in the
>specific case of the DH.
I still think that, in fact, it does work, when you think about the tactical
use of the DH position. I'll try to convince you of that below.
>And it doesn't work. Or if it does, then one is forced to conclude that
>AL first basemen have negative defensive responsibility, since they hit
>better than DH's who have zero responsibility. (thanks for supplying the
>data, Keith).
Alternatively, it means that a first baseman's defensive responsibilities
are so light that its nearly the same as a DH, who, by definition, has none,
and that there are other, non-defensive factors at work that explain
the differences we measure.
Consider a team that has two equally poor defensive players Thomas and Baines,
that are both exceptional hitters, with Thomas noticeably better than Baines.
The manager wants to work them both into the lineup as much as possible.
In order to avoid too much damage on defense, he wants to make one the
regular first baseman to ensure that he gets enough work to stay sharp
in the field, and generally to give his players the stability of assigned
roles. The other player will be the regular DH. Based on fielding
ability alone, he has no preference between Thomas and Baines.
We also know that the DH slot often gets used as a sort of "easing back
in" for injured players, or a partial day-off for fielding regulars to
get a break from the rigors of fielding without losing their bat in the
lineup. Therefore, we know that the regular DH will get fewer plate
appearances over the course of a season than the regular first baseman,
because of the manager's tactical use of that lineup position.
Assuming the manager is interested in maximizing offensive production,
which player is going to be the regular first baseman, and which will be
the DH? Since you get more offense out of Thomas than Baines, Thomas
will play first regularly, and Baines will be the DH (except when the
manager needs to use that slot as outlined above). The team will get
more absolute production out of first than out of DH, but not because
the DH position was harder to fill.
>Here's the crux of the argument for *not* judging DH production by first
>base standards. Value comes from scarcity.
I agree with you about value coming from scarcity.
But note that this has nothing to do with how scarce good DH's are. The
hypothetical team above had at least two, for example, equally capable of
playing DH, with no defensive tradeoffs to be considered. Other factors,
not related to the defensive responsibilities account for the small
difference between the offensive levels of the two positions.
>It is harder to find a terrific hitting DH than a terrific hitting first
>baseman.
Since all first baseman (and in fact all other players) are equally capable
of playing DH, this is not a true statement.
>That means that the team with exceptional DH production will gain more
>than team with a first baseman who hits equally well. I'm not saying that's
>the way it should be, I'm just looking at the data and saying that's the way
>it is.
But there are reasons underlying that data too, namely that a large portion
of the cumulative DH plate appearances come from regular players at other
positions who are known not to be at the top of their game. This bias is
important in understanding why the DH level comes out lower. The
difference is much less pronounced when you compare first basemen to
the genuinely fulltime DH'es:
293/.376/.505 -- AL first basemen
289/.370/.489 -- AL regular fulltime DH'es
I'd bet that the 22 points of OPS difference between those figures are
due more to Frank Thomas than anything else. :-)
[ For those who are curious, I did calculate it... :-) ]
[ .292/.370/.497 -- AL first basemen without Frank Thomas ]
It would seem that the genuine fulltime DH'es are drawn from a population
whose offensive profile is similar to those selected to be first basemen.
>There is not an unlimited pool of players out there who can hit like
>starting first basemen. Those who can hit like Frank Thomas seem to end
>up playing first, *regardless of defensive ability*.
Exactly -- it has nothing to do with the defensive requirements of the DH
(or 1B) position. It has to do with other factors influencing their patterns
of use. And since the adjustment I am making is designed to account for
the defensive differences among positions, it doesn't make sense to judge
designated hitters against a lower standard when they have no defensive
responsibilities to adjust for.
>The leftovers often end up playing DH.
To some extent, you could think of a DH as the second best first baseman
on the team (except where there are clear defensive differences among the
candidates). This doesn't mean that DH is harder to play, though. In
fact, it means its easier, because if that position wasn't available, the DH
wouldn't be playing at all (as opposed to pushing a less-productive first
baseman onto the bench).
>I think it's useful to judge DH's on that basis, rather than on assumed
>defensive inferiority.
A DH has no defensive responsibilities at all. A first baseman has minimal,
but certainly nonzero responsibilities. It's a logical conclusion, then,
that DH-es have less defensive importance than first baseman, rather than
an assumption.
While there may be valid reasons why you see differences in the production
rates of DH and 1B, they aren't related to the relative defensive demands of
the positions (a conclusion we make because we know something about how
the game works). Since the positional adjustment I'm making in VORP
is trying to account for defensive differences, rather than other factors
in managerial use, I still think comparing DH to 1B is the best way to go.
carl brenner
In article <471da1$d...@news.cic.net>, m...@Radix.Net (Ted Frank) wrote:
> Perhaps the proper solution is to take a set of (first basemen plus dh's)
> and determine replacement level for that -- but that's unfair to people
> who *can* play the field. Edgar Martinez clogs up the dh spot because
> he can't field; the same is true for Harold Baines and, this season,
> Juan Gonzalez and Chili Davis and Jose Canseco.
I wonder if someone could help me understand why Edgar is considered to
be such a lousy fielder. I don't live in Seattle and so never got to see
him on a regular basis, but in my copy of _Total Baseball_ (the 1993
version, not the newest one), he is rated as an average fielder (in fact,
even a little better than Blowers). But this book has many fielding
ratings that go against the conventional wisdom. Don Mattingly, whom
Yankeee fans are convinced is the greatest fielder on the planet, is rated
as below average (just for example).
Could someone give me some background on whether or not Thorn &
Palmer's fielding ratings are really useful, and if not, why not?
Thanks,
Carl Brenner
David Nieporent
In article <4736u9$m...@news.cic.net>,
carl brenner <cbre...@postoffice.ptd.net> wrote:
>In article <471da1$d...@news.cic.net>, m...@Radix.Net (Ted Frank) wrote:
>> Perhaps the proper solution is to take a set of (first basemen plus dh's)
>> and determine replacement level for that -- but that's unfair to people
>> who *can* play the field. Edgar Martinez clogs up the dh spot because
>> he can't field; the same is true for Harold Baines and, this season,
>> Juan Gonzalez and Chili Davis and Jose Canseco.
> I wonder if someone could help me understand why Edgar is considered to
>be such a lousy fielder. I don't live in Seattle and so never got to see
>him on a regular basis, but in my copy of _Total Baseball_ (the 1993
>version, not the newest one), he is rated as an average fielder (in fact,
>even a little better than Blowers).
This is just slightly misleading phrasing on the part of the original
poster. It's not that Edgar can't field because he's got no glove, it's
that Edgar can't field because it's likely that he'll get hurt if he
does. He's not like Sam Horn, who had as much use for a glove as a
quadruple amputee. He's more like Paul Molitor, who can't field because
he'd be on the disabled list faster than you can say "Today's players go
on the DL for a hangnail."
>But this book has many fielding
>ratings that go against the conventional wisdom. Don Mattingly, whom
>Yankeee fans are convinced is the greatest fielder on the planet, is rated
>as below average (just for example).
> Could someone give me some background on whether or not Thorn &
>Palmer's fielding ratings are really useful, and if not, why not?
They're not very useful. They're better than flipping coins, and they're
better than fielding percentage, (which might be redundant), but that's
about it. Essentially, they're range factor, with a bunch of (as far as
I can tell) arbitrary modifications.
I'd believe T&P before I'd believe "Peter Gammons said he was the best
shortstop in the game."
Has anybody tried to see how well DAs or ZRs and FRs correlate?
--
David M. Nieporent "We don't want 'rational' decisions made in
Niep...@pluto.njcc.com sport. We don't need your notion of 'rational.'
Deer Creek/Plainsboro, NJ We have our own." -- RLM in r.s.b, 10/27/95
DAVEY JOHNSON & ORIOLES 1996!!!!!!!!
Eric Roush
In article <4736u9$m...@news.cic.net>, cbre...@postoffice.ptd.net (carl
brenner) wrote:
> In article <471da1$d...@news.cic.net>, m...@Radix.Net (Ted Frank) wrote:
> > Perhaps the proper solution is to take a set of (first basemen plus dh's)
> > and determine replacement level for that -- but that's unfair to people
> > who *can* play the field. Edgar Martinez clogs up the dh spot because
> > he can't field; the same is true for Harold Baines and, this season,
> > Juan Gonzalez and Chili Davis and Jose Canseco.
> I wonder if someone could help me understand why Edgar is considered to
> be such a lousy fielder. I don't live in Seattle and so never got to see
> him on a regular basis, but in my copy of _Total Baseball_ (the 1993
> version, not the newest one), he is rated as an average fielder (in fact,
> even a little better than Blowers).
I think that in the case of Edgar, it's not that he *can't* field,
it's that the team is better off if he doesn't field, because if he
plays 3rd, he could get hurt, and then the Mariner offense goes down
the proverbial toilet. He was supposedly an average to slightly-below
average 3B in the field before the injuries began to strike.
(Sorta like the Wade Boggs rep; great hitter, little power, must
be so-so in the field...could this be a 3B corollary to
NLoCD?)
--
Eric Roush "I'd have traded Sierra for Ferdinand
edr...@acpub.duke.edu the Bull."
also coa...@aol.com Sandy Alderson
ob?: Why'd he trade *for* Sierra?
David Grabiner
In article <473kv9$o...@news.cic.net>, Dale J Stephenson writes:
> Christopher Hill <ceh...@u.washington.edu> writes:
>> David M. Tate responded:
> [...]
>>> (This raises the $64000 question: why don't DHs hit better than all other
>>> positions, on average? It's a good question, with a complex answer.)
> In order to set a replacement DH's hitting below the level of a first
> baseman's replacement, we need to show that *the same player* would
> hit better as a first baseman than as a DH. You approach that argument
> below:
>> Why that should be I don't know, but here are three ideas. People often
>> say it is harder to hit well coming off the bench cold every two to three
>> innings. That could be.
> I've heard this argument about pinch hitters, but do they really say
> this about DHs? Conventional wisdom is that a less difficult defensive
> position allows better hitting -- and there is no less difficult
> defensive position.
> Every hitter comes off the bench cold [as far as hitting is concerned]
> every two to three innings. We're not talking about not getting
> enough at bats to stay sharp here.
> If this is true, we might be able to see it by comparing performances
> at DH to performances at other positions by the same players. I don't
> know of such a study. [You would have to correct for injury related
> effects somehow, though.]
Elias looked at this, but they did the study with their usual degree of
rigor. They looked at every player who played a defensive position and
DH for a fair number of games during the same season, and comarped
batting averages.
Most outfielders hit better when they played the field than as DH''s.
Infielders hit slightly better; these were mostly first basemen, but
there were a few players who played other positions. Catchers hit
better as DH's than while catching; this isn't surprising, given the
strain that catching puts on a batter's knees.
>> Some managers use the DH spot to get at bats for injured players or
>> good-hitting catchers on a day off from the rigors of defense. To be good
>> enough to play first you might only have to hit better than the next guy.
>> To play DH, you might also have to convince the manager that he should
>> give up the option of using the spot for injured players.
This is the principal problem with the Elias study. I would assume that
most of the infielders (other than first basemen) and many of the
outfielders were nursing minor injuries.
On the other hand, there is some evidence that players who don';t
like to DH don't hit as well there. Reggie Jackson, for example, spent
his last five years with the Angels as a DH/RF, and hit better as a
right fielder then as a DH all five years.
> This doesn't matter for assigning replacement level, since neither
> a replacement 1B or a replacement DH would be playing injured.
And this is why I don't trust the Elias results. For my purposes, I
would assume that a replacement-level DH hits as well as the
replacement level at the best offensive position.
> Of course, it's still *possible* that the same batters hit worse at
> DH. Anecdotally, I think the reverse may be true. Look around at
> the current crop of DHs: Chili Davis, Paul Molitor, Harold Baines --
> these guys had their *best* offensive years as DHs, and on the wrong
> side of thirty. Edgar Martinez was a good hitting 3B. He's moved
> to DH, and suddenly he's Frank Thomas. Dave Winfield had one of
> his best years as a DH when he was pushing forty.
This tends to work opposite to my anecdotal evidence. On the other
hand, there are other factors at work in some of these cases. Molitor
didn't hit well as a second baseman or third baseman because he kept
getting hurt; the DH spot was more important in keeping him healthy.
--
David Grabiner, grab...@math.harvard.edu
I don't speak for Harvard, or any other organization that speaks for itself.
Shop at the Mobius Strip Mall: Always on the same side of the street!
Klein Glassworks, Torus Coffee and Donuts, Projective Airlines, etc.
Jeffrey McGowan
: > I wonder if someone could help me understand why Edgar is considered to
: >be such a lousy fielder. I don't live in Seattle and so never got to see
: >him on a regular basis, but in my copy of _Total Baseball_ (the 1993
: >version, not the newest one), he is rated as an average fielder (in fact,
: >even a little better than Blowers).
: This is just slightly misleading phrasing on the part of the original
: poster. It's not that Edgar can't field because he's got no glove, it's
: that Edgar can't field because it's likely that he'll get hurt if he
: does. He's not like Sam Horn, who had as much use for a glove as a
: quadruple amputee. He's more like Paul Molitor, who can't field because
: he'd be on the disabled list faster than you can say "Today's players go
: on the DL for a hangnail."
: >But this book has many fielding
: >ratings that go against the conventional wisdom. Don Mattingly, whom
: >Yankeee fans are convinced is the greatest fielder on the planet, is rated
: >as below average (just for example).
: > Could someone give me some background on whether or not Thorn &
: >Palmer's fielding ratings are really useful, and if not, why not?
: They're not very useful. They're better than flipping coins, and they're
: better than fielding percentage, (which might be redundant), but that's
: about it. Essentially, they're range factor, with a bunch of (as far as
: I can tell) arbitrary modifications.
: I'd believe T&P before I'd believe "Peter Gammons said he was the best
: shortstop in the game."
: Has anybody tried to see how well DAs or ZRs and FRs correlate?
: --
Well, I haven't done that, but you can get some idea of how useful their
fielding ratings are by 1) checking the formulas in the back of the book
(which *are* basically range factors, but changed in weird ways, and if
memory serves , the 1B formula is almost worthless) and 2) noticing that
the numbers vary *wildly* - just check a bunch of players and you will
see people who seemed to be saving their team 40 runs one year, costing
them 10 the next, then saving them 25 the year after. That would seem to
indicate a serious lack of utility.
Clifford Blau
In <473kvd$o...@news.cic.net> niep...@pluto.njcc.com (David Nieporent)
writes:
>
>> Could someone give me some background on whether or not Thorn &
>>Palmer's fielding ratings are really useful, and if not, why not?
>
>They're not very useful. They're better than flipping coins, and
they're
>better than fielding percentage, (which might be redundant), but
that's
>about it.
>
>
with runs allowed. From 1946 to 1988, the leader in FR allowed an
average of about 23 runs less than the league average, while the team
with the worst FR allowed about 33 runs above average. Thus, it would
seem that they are somewhat useful.
Incidentally, fielding average correlates positively with preventing
EARNED runs, so it is also somewhat useful.