total baseball replacement value
j albright
Hello!
I wonder if anyone has worked out a reasonable approximation of
replacement value for Total Baseball's TPI and TPR? IMHO, it would have
to be expressed in terms of per 150g for positional players, and for
starting pitchers per 200 or 250 ip, and for relievers per 50 gp. My
seat of the pants estimate would be about -2.5g for each of those
standards (choosing 250 ip for pitchers), but I would welcome any more
researched views (or other informed guesses).
--
* Keith Woolner, Moderator for rec.sport.baseball.analysis *
* Post/Info/Contact: send email to base...@cedar.cic.net *
Dale Stephenson
In <5f366t$b2u$1...@news.cic.net> j albright <jalb...@earthlink.net> writes:
>Hello!
> I wonder if anyone has worked out a reasonable approximation of
>replacement value for Total Baseball's TPI and TPR? IMHO, it would have
>to be expressed in terms of per 150g for positional players, and for
>starting pitchers per 200 or 250 ip, and for relievers per 50 gp. My
>seat of the pants estimate would be about -2.5g for each of those
>standards (choosing 250 ip for pitchers), but I would welcome any more
>researched views (or other informed guesses).
I believe in crediting a starting pitcher with an extra half run per
start. For modern times, I can't see giving more than a run at the
most -- that would have a replacement starter allowed to pitch 6
innings with an ERA 1.50 below average, which seems unrealistic to me.
I see the number of innings per start to be less important, since the
innings taken away from the bullpen are likely to be of approximately
league average quality, and the beneficial effect of the extra rest
for the bullpen doesn't amount to much IMO. At a half run per start,
a 35 GS season (probably over 200 IP) would be worth less than two
wins, but I think 2.5 wins is within the realm of possibility, if unlikely.
For relievers, it's less clear which reliever is being replaced. A league
average pitcher pitching 50 innings in a closer role is probably hurting
his team, since his replacement should be the best reliever on staff.
A league average pitcher pitching 50 innings in a mopup role is probably
*well* above the level of his replacement, but the effect on wins and
losses is probably close to nil.
For position players, 2.5 wins seems feasible, but still strikes me as too
high. For DHs, I don't believe they should be given any value above
replacement at all.
--
Dale J. Stephenson |*| (st...@cs.uiuc.edu) |*| Baseball fanatic
"Even the wise cannot see all ends." -- Gandalf
Keith Woolner
In message <5f366t$b2u$1...@news.cic.net> - j albright
<jalb...@earthlink.net>Wed, 26 Feb 1997 10:14:34 -0500 writes:
:> I wonder if anyone has worked out a reasonable approximation of
:>replacement value for Total Baseball's TPI and TPR? [...]
I've done quite a bit of work in this area with a measure I call VORP (Value
Over Replacement Player), which is expressed in runs contributed above what a
replacement level player would contribute. The toughest part of this process
is determining what a sensible value for replacement level is. For my
purposes, I define a replacement level player as one who: (a) hits as far
below the league average for the position as an average backup does below the
overall league average, (b) is a breakeven base-stealer, and (c) plays average
defense at the position (the notion that reasonably competent glovemen are
easier to find in the minors, but their bats are what keep them from
advancing).
Basically, what I've done is to look at the typical difference
in OPS between the average player at a position, and the typical
backup at a position. In particular, I compute backup averages by removing
players from the positional totals who are not listed in Total Baseball with a
a '*' at the start of their positions played entry (indicating at least 100
games at their primary position). That doesn't guarantee that the player
was truly a backup, as it could have been an injured regular, or platoon
player, but it seemed like the best empirical way to look at it.
Taking the mean difference between positional average and backup averages
across all years in the live-ball era (since 1920) yielded a gap of about
50-55 points of OPS. Interestingly and surprisingly enough, there was not a
significant correlation between the magnitude of the gap and the year of
play, or the overall offensive level.
Separately, I did some hand calculations in 1994 and 1995, removing the top 14
players at each postion with the most plate appearances, and figuring the
difference in average of the remaining players. Though there was wide
variations in the one-year figures, the average for all position across for
each year was about the same -- 65-70 points of OPS.
As a result of the latter study, I've set replacement level at 35
points below positional average in both OBP and SLG, slightly greater
than what the historical study showed, but still in the right range
(at most a 10 point difference). I haven't yet investigated the
replacement-level delta of individual positions enough to know whether certain
positions have more of a dropoff (though there are indications that
high-offense positions like 1B and OF may show exactly that trait), so for now
I use the same size gap for all positions.
Unfortunately, I haven't given the same attention to setting replacement
level for pitchers. Currently, I'm using an ERA (or RA, depending on the
dataset) of +1.00 greater than the league average (E)RA. But I could be
persuaded to change that by a strong argument in favor of some other value.
Anyways, having determined replacement-level, I then compute VORP as follows:
For hitters:
* Start with the player's own park-adjusted AVG, OBP, and SLG.
* Determine the replacement-level AVG/OBP/SLG for the position most
often played.
* Determine the Marginal Lineup Value (MLV) for the player over a
replacement level player in the same percentage of team plate appearances
as the actual player had. MLV is a technique for estimating the impact
on the team's overall Runs Created projection by calculating his impact
on the team's AVG, OBP, and SLG. This quantity is called RPOSMLV.
* The linear weight of a player's net steals is calculated as 0.3*(SB-2*CS)
This value, added to RPOSMLV, yields VORP (Value Over Replacment Player).
* If you have a measure of intrapositional defense that you trust (which
could be Total Baseball's fielding runs, or something better like DA
or ZR based fielding runs), you can add that to VORP to get VORPD
(VORP plus Defense).
For pitchers
* Determine the pitcher's park-adjusted ERA (or RA, depending on your
preference)
* Calculated the number of runs prevented by the pitcher compared to a
replacement-level pitcher inthe same number of innings
(RepLvlERA = LgERA + 1.00)
* Add, as desired, a pitcher's batting runs and/or fielding runs.
* Note: a VORPD for pitchers would involve adjusting his contribution
for the quality of the defense behind him. I have not attempted this yet.
Giving credit where it is due: David Tate developed the original
concept for MLV, which I've extended to positional and replacement level
considerations.
(All the following lists are through 1993 only, as I haven't
yet gotten around to incorporating 1994-96 stats into my
database yet)
For example, the top 20 VORP seasons by hitters are as follows:
Year Lg Name Team Pos VORP
-------------------------------------------
1921 AL Ruth,Babe NY-A *O/P1 156.6
1920 AL Ruth,Babe NY-A *O/1P 144.1
1927 AL Gehrig,Lou NY-A *1 142.1
1923 AL Ruth,Babe NY-A *O/1 141.1
1922 NL Hornsby,Rogers StL-N *2 140.5
1927 AL Ruth,Babe NY-A *O 134.1
1930 AL Ruth,Babe NY-A *O/P 133.2
1924 AL Ruth,Babe NY-A *O 130.9
1925 NL Hornsby,Rogers StL-N *2M 130.7
1941 AL Williams,Ted Bos-A *O 130.5
1924 NL Hornsby,Rogers StL-N *2 126.9
1926 AL Ruth,Babe NY-A *O/1 126.8
1930 AL Gehrig,Lou NY-A *1/O 125.1
1931 AL Ruth,Babe NY-A *O/1 124.9
1948 NL Musial,Stan StL-N *O/1 124.7
1957 AL Mantle,Mickey NY-A *O 122.6
1993 NL Bonds,Barry SF-N *O 122.6
1933 AL Foxx,Jimmie Phi-A *1/S 121.4
1887 AA ONeill,Tip StL-a *O 120.5
1921 NL Hornsby,Rogers StL-N *2/OS 120.4
And a sampling of the top VORP seasons by pitchers:
Year Lg Name Team Pos VORP
-------------------------------------------
1912 AL Johnson,Walter Was-A P 130.3
1901 AL Young,Cy Bos-A P 126.5
1913 AL Johnson,Walter Was-A P 116.6
1923 NL Luque,Dolf Cin-N P 112.3
1905 NL Mathewson,Chris NY-N P 110.5
1939 NL Walters,Bucky Cin-N P 106.1
1914 FL Hendrix,Claude Chi-F P 106.1
1902 AL Young,Cy Bos-A P 105.9
1915 NL Alexander,Pete Phi-N P 105.1
1908 AL Walsh,Ed Chi-A P 104.8
1931 AL Grove,Lefty Phi-A P 104.3
1912 AL Wood,Joe Bos-A P 103.5
1968 NL Gibson,Bob StL-N P 101.7
1944 AL Trout,Dizzy Det-A P 101.7
1940 AL Feller,Bob Cle-A P 101.6
1937 AL Gomez,Lefty NY-A P 101.3
1930 AL Grove,Lefty Phi-A P 101.2
1912 AL Walsh,Ed Chi-A P 100.8
1985 NL Gooden,Dwight NY-N P 100.7
1953 NL Roberts,Robin Phi-N P 100.3
1965 NL Koufax,Sandy LA-N P 100.1
1972 NL Carlton,Steve Phi-N P 99.7
1953 NL Spahn,Warren Mil-N P 99.4
1966 NL Koufax,Sandy LA-N P 99.0
I've also figured out career VORP values, listed
here for position players by descending VORPD
(using TB's Fielding Runs, such as they are).
NAME YEARS PLAYED VORP VORPD
--------------- ----- --------- ------- -------
Ruth,Babe 22 1914-1935 1760.4 1766.4
Aaron,Hank 23 1954-1976 1326.2 1427.2
Mays,Willie 22 1951-1973 1223.1 1403.1
Williams,Ted 19 1939-1960 1427.3 1394.3
Cobb,Ty 23 1905-1928 1345.3 1389.3
Lajoie,Nap 21 1896-1916 972.4 1339.4
Speaker,Tris 22 1907-1928 1063.3 1312.3
Musial,Stan 22 1941-1963 1290.3 1251.3
Hornsby,Rogers 23 1915-1937 1330.0 1239.0
Wagner,Honus 21 1897-1917 1102.0 1204.0
Gehrig,Lou 17 1923-1939 1201.0 1140.0
Mantle,Mickey 18 1951-1968 1108.9 1097.9
Robinson,Frank 21 1956-1976 1042.3 1094.3
Schmidt,Mike 18 1972-1989 798.9 1062.9
Collins,Eddie 24 1906-1930 996.4 1036.4
Ott,Mel 21 1926-1946 1023.9 1013.9
Foxx,Jimmie 20 1925-1945 979.5 1007.5
Henderson,Ricke 15 1979-1993 793.9 955.9
Yastrzemski,Car 23 1961-1983 765.8 914.8
Delahanty,Ed 16 1888-1903 844.0 898.0
And now for pitchers:
NAME YEARS PLAYED VORP VORPD
--------------- ----- --------- ------- -------
Young,Cy 22 1890-1911 1522.6 1542.6
Johnson,Walter 21 1907-1927 1467.3 1443.3
Spahn,Warren 21 1942-1965 1160.5 1169.5
Alexander,Pete 20 1911-1930 1074.4 1107.4
Mathewson,Chris 17 1900-1916 1005.8 1075.8
Nichols,Kid 15 1890-1906 1053.3 1044.3
Seaver,Tom 20 1967-1986 979.6 991.6
Keefe,Tim 14 1880-1893 964.0 981.0
Grove,Lefty 17 1925-1941 988.3 973.3
Perry,Gaylord 22 1962-1983 889.0 907.0
Clarkson,John 12 1882-1894 846.6 891.6
Ryan,Nolan 27 1966-1993 911.4 882.4
Rusie,Amos 10 1889-1901 846.8 873.8
Ruffing,Red 22 1924-1947 910.4 873.4
Radbourn,Charle 11 1881-1891 839.2 847.2
Carlton,Steve 24 1965-1988 857.5 840.5
Sutton,Don 23 1966-1988 864.6 838.6
Palmer,Jim 19 1965-1984 830.1 831.1
Galvin,Jim 15 1875-1892 784.3 824.3
Lyons,Ted 21 1923-1946 809.9 822.9
Mullane,Tony 13 1881-1894 785.2 818.2
You'll note that VORP rates pitchers with long
careers more highly that you might otherwise think.
It's not clear to me (and I'm open to arguments
in either direction) whether this method doesn't
correctly value pitchers, whether replacement
level is set in such a way to distorts their
value, or whether the Don Sutton's of the world
really are that valuable over the course of a
career.
For those who are interested, there's more detail
on my Baseball Engineering Web page, which can be found at:
http://www.best.com/~kwoolner/bbeng
You'll find a complete explanation and derivation of MLV,
further details on VORP, complete 1996 VORP's, and
the top VORP position players in each league for each year.
---
Keith Woolner
kwoo...@best.com
http://www.best.com/~kwoolner