You are here > Home > Hall of Merit > Discussion
 
Hall of Merit — A Look at Baseball's AllTime Best Monday, May 15, 20061977 Ballot Discussion1977 (May 15)—elect 2 Players Passing Away in 1976 Candidates Upcoming Candidates Thanks, Dan! John (You Can Call Me Grandma) Murphy
Posted: May 15, 2006 at 11:05 PM  178 comment(s)
Login to Bookmark
Related News: 
BookmarksYou must be logged in to view your Bookmarks. Hot Topics2018 Hall of Merit Ballot Discussion
(221  11:53pm, Sep 19) Last: Bleed the Freak Most Meritorious Player: 2008 Discussion (7  4:02pm, Sep 19) Last: John (You Can Call Me Grandma) Murphy Bob Boyd (12  9:45pm, Sep 18) Last: Dr. Chaleeko Most Meritorious Player: 2007 Results (8  7:47pm, Sep 14) Last: Chris Fluit Latin American Players and Pitchers Home Page (55  1:25pm, Sep 08) Last: DL from MN Most Meritorious Player: 2007 Results (1  8:10pm, Sep 06) Last: ThickieDon Most Meritorious Player: 2007 Ballot (11  4:08pm, Sep 06) Last: DL from MN Most Meritorious Player: 2007 Discussion (23  10:52am, Sep 05) Last: DL from MN Most Meritorious Player: 2006 Results (14  9:36am, Aug 07) Last: DL from MN Heavy Johnson (18  10:45am, Aug 04) Last: KJOK Most Meritorious Player: 2006 Ballot (18  5:00pm, Aug 02) Last: DL from MN Most Meritorious Player: 2006 Discussion (17  10:04am, Jul 29) Last: Tubbs is Bobby Grich when he flys off the handle Most Meritorious Player: 2005 Results (9  3:12pm, Jul 17) Last: Tom Riddle Most Meritorious Player: 2005 Ballot (13  4:39pm, Jul 05) Last: DL from MN Most Meritorious Player: 2005 Discussion (17  10:11pm, Jul 02) Last: Tubbs is Bobby Grich when he flys off the handle 

Page rendered in 0.7270 seconds 
Reader Comments and Retorts
Go to end of page
Statements posted here are those of our readers and do not represent the BaseballThinkFactory. Names are provided by the poster and are not verified. We ask that posters follow our submission policy. Please report any inappropriate comments.
1903 6th
1904 10th
1905 24th
1906 5th
Unlike today's game, many who finsihed ahead of him were pitchers. Considering his team's success, KJOK's posit is at least plausible.
overall for 4 years
Wagner 170
McGinnity 128
Mathewson 122
Chance 120
Chesbro 120
Yeah, that Honus guy was good. Especially considering those weren't necessarily his best years.
Small note on Honus: MVP voting did not occur until 1911. In 1912, the 38 yr old shortstop (finally) didn't win a batting or slugging title. And he finished 2nd in MVP voting.
1904 10th
1905 24th
1906 5th
Unlike today's game, many who finsihed ahead of him were pitchers.
That begs the question as to what his rankings are among position players in those years. How about among NL position players?
(sorry, I'd check myself but my digital win shares is at home).
name.... yrs used .......games OWP
Browning 188293 (x89) 1097 .759
Minoso..... 195161 .......1043 .655
Kiner....... 194655 .......1472 .693
Keller..... 193951 (x44) 1168 .748
C Jones... 187687 ........ 875 .700
Johnson... 193345 .......1863 .651
Sisler...... 191722 ........ 966 .737
Chance....190011 ....... 1155 .735
Since some have been comparing Keller to Sisler, I'll use Keller's 7yr prime for a more direct comp:
Keller..... 1939, 4146 ... 738 .777
Sisler...... 191722 ........ 966 .737
Okay, I cheated, it subs Keller's 39 for his 40, and of course he missed time in 194445 for the war. This shows that at their best:
1. Keller was a better hitter. Point not even really up for arguing, is it? The diff in OWP of 40 pts is about the same as .025 in batting avg.
2. Sisler was a little more durable, but not much, IF you assume Keller would have been as healthy in 4445 as other years.
I haven't had Charlie on my ballot yet, but he may vault Sisler this week.
name yrs used PA .OBP SLG
Keller 3951 .4618 .402 .607
Sisler 1622 .4823 .384 .567
one thousand SIX HUDNRED forty three games, Tom, get it right!!!!!!!
Since I reject Win Shares as the proper method to value players, I'll post a different measure:
NATIONAL LEAGUE RCAP LEADERS
1903 NL
1 Honus Wagner 77
2 Frank Chance 66
3 Roger Bresnahan 46
4 Jimmy Sheckard 44
5 Mike Donlin 34
6 Johnny Kling 32
7 Fred Clarke 31
8 Harry Steinfeldt 25
9 Roy Thomas 23
T10 Ginger Beaumont 21
T10 Cy Seymour 21
1904 NL
1 Honus Wagner 89
2 Mike Grady 39
3 Frank Chance 35
4 Harry Lumley 33
5 Roy Thomas 28
6 Art Devlin 25
7 Mike Donlin 24
T8 Jake Beckley 20
T8 Roger Bresnahan 20
T8 Jim Delahanty 20
1905 NL
1 Honus Wagner 89
2 Cy Seymour 65
3 Mike Donlin 50
4 Frank Chance 47
5 John Titus 35
T6 Roger Bresnahan 33
T6 Mike Grady 33
T8 Miller Huggins 29
T8 Dan McGann 29
T10 Art Devlin 26
T10 Sam Mertes 26
1906 NL
1 Honus Wagner 81
2 Harry Lumley 59
3 Frank Chance 53
4 Roger Bresnahan 47
5 Art Devlin 44
6 Sherry Magee 42
7 Sammy Strang 41
8 Harry Steinfeldt 37
9 Johnny Kling 28
10 Tim Jordan 26
1907 NL
1 Honus Wagner 88
2 Sherry Magee 53
3 Ginger Beaumont 32
4 Roger Bresnahan 28
T5 Frank Chance 26
T5 Fred Clarke 26
7 Dave Brain 24
T8 Tim Jordan 21
T8 Johnny Kling 21
T8 Sammy Strang 21
Really a man among boys.
Really a man among boys.
RCAP actually *underrates* Honus. With only 8 SS's in the league, he does a pretty good job of raising the average RC for the position.
Note that KJOK stopped with the 1907 season. Wagner had his best single season in 1908.
In my ballot comment I said something about not having figured out where to rank Banks agains the likes of Vaughan, Cronin, and Boudreau. There's a reason that Wagner's name didn't appear in that comment.
I want to start with DERA and translated IP. For those unfamiliar DERA is Baseball Prospectus' Defense Independent ERA. "Normalized runs" have the same win value, against a league average of 4.5 and a pythagorean exponent of 2, as the player's actual runs allowed did when measured against his league average. DERA adjusts this for defense. It's not DiPS related.
I think there is a problem with that. At a 4.50 R/G environment, the exponent should be 1.87, not 2.00. But that's if the hitters and pitchers are at 4.50. The pitchers we are evaluating a generally well below 4.50, especially in their big seasons. So the exponent should generally be even lower than 1.87.
Take Walter Johnson 1913  He allowed 1.46 R/9. League was 3.93. Using PythaganPat, I get an .833 WPct.. Plugging in his NRA (2.00) with 4.50 as league and an exponent of 2, I get an .835 WPct.
Or his 1907  Johnson allowed 2.86 R/9. League was 3.66. Using PythaganPat, I get a .599 WPct. Plugging in NRA with 4.50 as league and an exponent of 2, I get a .594 WPct.
That seems OK especially since they are using PythaganPort, and not PythaganPat, which accounts for being a few % points off. But the problem is that in a 4.50 R/G environment 2.00 RA is only an .800 WPct, not .833, because the exponent should be 1.708, not 2. 1.74 RA in a 4.50 environment has the same win impact as 1.46 RA in a 3.93 environment. So they've overstated his NRA by 0.26, which is a huge difference.
Or take 1907. In a 4.5 R/G environment 3.55 RA/G has the same win impact as 2.86 RA/G does in a 3.66 R/G environment. So his NRA (3.72) is overstated, it should be 3.55.
What they should be doing is using a PythaganPat exponent for figuring the pitchers actual win impact based on his runs allowed  and then translating to a 4.50 environment with the PythaganPat exponent there as well.
To get this spreadsheet to calculate right, I need to figure out how to take
WPct of (lRA/pRA) = WPct of (4.5/nRA)
where WPct =
(lRA^((lRA+pRA)^.286))
________________________________________________
((lRA^((lRA+pRA)^.286)) + (pRA^((lRA+pRA)^.286)))
And solve for nRA
It's easy enough to do for a single season by just changing ERA manually until the WPct match up  but that's not feasible for something this big (every season of every pitcher worth evaluating).
Any math guys know how I can get excel to do this?
Adjusting . . .
For 1913, league becomes 3.969, WPct is .836 given Johnson's RA. At 4.50, with 2.00 RA/G, we get a WPct of .835 with a 2 exponent. But the exponent should be 1.708, so at 2.00 RA/G you are really only winning at an .800 clip. RA/G should be 1.71 in a 4.50 R/G environment to get a WPct of .836.
For 1907, league becomes 3.477, WPct is .583 given Johnson's RA. At 4.50, with 3.72 RA/G, we get a WPct of .594 with a 2 exponent. But the exponent should be 1.827, so at 3.72 RA/G you are really only winning at an .586 clip. RA/G should be 3.75 in a 4.50 R/G environment to get a WPct of .583.
So you can see it makes less difference the closer to 4.50 the pitcher's ERA gets. At the extreme levels, big years (ERA wise, IP don't matter here) are being underrated.
So in a nutshell I still have the issue of solving for one independent variable (RA/G) and one dependent variable (PythaganPat) in the same equation.
Assuming I can get this figured out, I'll just take the number I get, and then do DERA/NRA*myRA to adjust for defense  since I trust Prospectus' adjustment for defensive support. That's the proper way to do it right?
His overall stats are good (SLG above even his regular season avg), he scored and drove in runs, and he had one of the 20 (maybe 10) most important World Series hits ever  after the famous Mickey Owen dropped strike / passed ball that would have ended game 4 of the 41 Series, Keller later doubled home two runs, turning a 43 deficit into a 54 Yankee win. Huge hit with 2 outs in the 9th inning.
Maybe that's why Davenport is using 2? Try dropping him an email about this?
Even though my initial reply above was premature, I think it gives the roadmap to what you'll need to do. Goal Seek is Excel's way to do each pitcherseason separately. Since this is very time consuming, I suggest doing each season of a specific pitcher all in one fell swoop. Goal Seek's big brother is Solver. It can reach a goal by moving multiple cells.
So for each season create a cell for the required difference between the two win pcts (like you displayed above). Create another cell for the square of this difference. Then create a cell for the sum of all these squared differences across all seasons for one pitcher. Have Solver try to minimize this "metasum" by adjusting the collection of the pitcher's seasonal NRA's (these are the figures you seek). Of course, the minimum value will be 0 where all the squared differences are zero, so it will find the NRA for each season that equates the two win pcts you are analyzing.
Now that I think about it, I suppose you can try to set up a "metametasum" over ALL pitcherseasons in your spreadsheet since the overall minimum would be found where all the pitcherseason squared differences are zero. To make sure things are working properly I would start first with one pitcher and one season, then hook up all the seasons for one pitcher, and, if that works, hook up multiple pitchers.
Let us know if any of this is helpful!
Alternatively  I think I have an idea  what if I figured out what the myRA (my tweaked NRA, we'll get a better name later) for every WPct from 1 to 0. That would give me a table. Then I could have excel refer to the table based on the pitcher's actual RA WPct, right?
Assuming I get something working, idea for using DERA/NRA*myRA to get my adjusted DERA works fine, right?
Jim, I thought about dropping him an email  but I imagine he has access to much better math tools than Excel  I'm guessing he just uses 2 because it made sense back then. But you are right, I should drop him an email and ask him about it.
As far as what I wrote above, does it all makes sense? I'm on solid ground right, I'm not finding something that isn't there, am I?
Now I need to figure a way to have it do that automatically. Very, very good. Thanks again.
Any chance you could explain that a little bit differently (like very basically?) I'm not really sure I understand  is it basically forcasting for each individual pitcher?
<<
Commonly, people solve these problems by writing programs rather than using canned applications (math tools). Davenport is a meteorologist, probably from a generation where everyone in the meganumberbashing observational sciences writes programs, probably trained on mainframe computing.
You may be able to write a spreadsheet that will assist immensely in a solution by iteration.
The assistance will be immense it enables one iteration for every pitcher(season?) at once.
Here's a sketch supposing a single column of data that needs correction, one magnitude for each pitcher.
 Data in column one. For now this is uncorrected data, iteration zero.
 Supporting data (doesn't need correction) and intermediate calculations in columns 3,4,5...
 Result of *one iteration of* the correction in column two. For now that is corrected data, iteration one.
That is the hard part.
 Copy magnitudes (not formulae) from column two to column one.
Now column one is corrected data, iteration one, and column two is the iteration two values.
 Iterate.
I think one could do this in the earliest GUI spreadsheet programs.
Can one generally do better in Excel?
Here's a trivial example of how to use Solver just in case it is not obvious. In a blank spreadsheet enter 3 in cell A1, 5 in A2, 2 in B1, and 8 in B2. Then create the following formula: C1=A1B1, C2=A2B2, D1=C1*C1, D2=C2*C2, D3=D1+D2. In this silly example, we are seeking values in B1 and B2 that minimize the value of D3. Clearly the answer is B1 should be 3 and B2 should 5.
Go to Solver under the Tools menu. Set target cell to be D3; select the Min radio button; and select B1:B2 in the by changing cells box.
The idea is that you can use Solver to do multiple goal seeks at once. The analog in the silly spreadsheet is that you can run Goal Seek on cell B1 (to make D1 0) and then run Goal Seek again on cell B2 (to make D2 0). But using Solver allows you to do all the separate Goals Seeks all at once.
Anyway, feel free to email me one of your spreadsheets and I can hook it up for you as long as I can follow what you are doing.
I just loaded the add in for solver  but I can't see how I can get it to calculate more than one cell at a time. It won't let me use a range for the 'set target cell', it has to be a single cell.
What would be nice would be if you could set a range on the target cells, and use a cell for the 'value of' portion, instead of having to manually input a number.
I'll send my sample spreadsheet along . . . I've been using Walter Johnson to work out the kinks.
I set up a cell that takes the sum of all of a pitchers career RA/WPcts at his normal environment  his RA/WPcts when the league is converted to 4.50 R/G. Basically (1907 RA/G WPct  1907 4.50 league RA/G WPct) + (1908 RA/G WPct  1908 4.50 league RA/G WPct) etc..
Then I tried to solve for making this cell equal to zero. What it did was make some over zero and sum under zero to get a total of zero, but the individual years are off.
So I then tried it with making each year's portion of the equation an absolute value. BINGO!!! Total error is zero. I think we're onto something here  thanks Rob!
Being top 5 in an 8 team league isn't nearly as impressive as being top 5 in a 16 team league. This biases the IP in favor of pitchers from smaller leagues. If 5 worked for an 8 team league, then we should use 6.25 for a 10 team league, 7.5 for a 12 team league, 8.75 for a 14 team league and 10 for a 16 team league. I can fix this in this spreadsheet.
Also if they are trying to set pitchers equal to a historic average runs level, why use 4.50? Since 1901 the AL has averaged 4.47 R/G, the NL 4.29. Something like 4.35 seems more appropriate to me, but that's a minor nitpick, but I can fix that too.
Here's another issue I have here regarding hitting.
If I take the IP and normalize to league leaders = 275 or whatever (I'll probably use the historic average through whatever the election year is), how do I equalize hitting?
Over the last 105 years, the average team has played a 155.6 game schedule. But pitchers have only had 130.2 games to bat  due to the DH effectively removing half the pitcher/hitter games over the last 34 seasons. If you go back to 1876, it's more like the average season was 148.0 games and only 127.6 pitcher batting games.
So what I think I should do is take RC above position (RCAP), since replacement level pitcher hitting is the average pitcher; divide by team games and then multiply by the average season's pitcher hitting games through the election year. We're normalizing innings based on the same thing, so that makes sense.
Of course, this will reduce the value of pitcher's who hit well over time  which is an issue. In one respect, I can see the side that says what this guy did had real value at the time. But we are also putting pitchers in a historical context here  and if that skill isn't transferable across time, it isn't as valuable historically. I'd be interested to hear different sides on this one.
OK a few other issues  I'm not sure why they set IP of the average of top 5 in the league = to 275 throughout history.
Being top 5 in an 8 team league isn't nearly as impressive as being top 5 in a 16 team league. This biases the IP in favor of pitchers from smaller leagues. If 5 worked for an 8 team league, then we should use 6.25 for a 10 team league, 7.5 for a 12 team league, 8.75 for a 14 team league and 10 for a 16 team league. I can fix this in this spreadsheet.
I agree that this is a problem: I suspect that they did it for the sake of simplicity. Your change to the system sounds like an improvement, but I think that using the top values in the league, while attractive because it is fairly easy, is problematic because the top performers are the outliers, and we shouldn't expect their performance to be consistent. In effect, this approach to normalizing workloads can penalize pitchers of great durability, because they bring up the averages in their era. I think it's better to normalize at the level of the average pitcher, but that (as discussed on the Bunning thread) is difficult to arrive at by purely empirical methods.
See that discussion for more on the issue of normalizing pitcher workloads.
Also if they are trying to set pitchers equal to a historic average runs level, why use 4.50? Since 1901 the AL has averaged 4.47 R/G, the NL 4.29. Something like 4.35 seems more appropriate to me, but that's a minor nitpick, but I can fix that too.
Here, I suspect that they use 4.50 because 4.5/9 = 1/2. It's convenient to know that average runs allowed is always 1/2 of IP. I don't believe there's any "baseball value" rationale for this figure.
On pitcher hitting: I don't understand all the programming of Excel that you will be doing, so I don't know if the following proposal would be workable at all, but I would suggest that you use RCAP and normalize it by the average season's pitcher hitting games, but don't take out for the DH. Just consider all pitchers under the DH to be "average." That way, goodhitting pitchers and badhitting pitchers will get full credit/debit for their the contextual value of their hitting, prorated to the same workload standard as their pitching.
We're actually reducing the number of innings they would have played to put them on equal ground with their modern counterparts  should their hitting contributions (or failures) be reduced proportionally as well? Obviously this would only apply if they didn't hit at other positions also.
I guess my options are to
1) take RCAP/tG*X with X being a constant # of games (130.2, 154, 162 are all reasonable)
2) take RCAP/IP*tIP
Still going to be a bear to properly evaluate the hitting of a guy like Caruthers.
And a guy like Ferrell who played 161 nonpitcher games (not counting his 13 OF games in 1933) should really have those extra PA compared to a PH, not a pitcher. So RCAP is going to overstate his hitting.
Lots of little things to deal with.
****************
Anyway the goal of all of this is to come up with a normalized pennants added superstat.
Basically once I get the RA/9 (adjusted NRA I guess) set to a neutral environment at the same WPct, I'll convert that to an adjusted DERA using aNRA/NRA*DERA  this will adjust for defense, using the Prospectus' defensive adjustment.
Then I'll take the aDERA (adjusted DERA) and come up with RSAR using 5.75 as the replacement level (at a 4.50 environment this translates to a .383 WPct, or basically gives a pitcher credit if they move a team past 100 losses in a 162 game season). The new translated IP will be used as the 'playing time' factor.
I'll take RSAR add RCAP and that will get the pitchers total runs above replacement.
In a 4.50 environment, it takes 9.558 runs to flip one game in the win column, to TRAR/9.558 will give an adjusted WARP. This can then easily be converted to Pennants Added, which gives more credit for bigger years. For example Johnson ends up with 4.9 in 1908, 5.1 in 1921. That gets a total of .100 PA. However his 10.0 season in 1915 gets .116 PA. That's using 18761943 NL as the Pennant context  it's been awhile since I've updated that part of it.
Anyone see any flaws in this methodology?
As for time to calc, once I finalize the template, I'll need to:
1) input seasonal RCAP (from the Sinins Encyclopedia)
2) input seasonal IP (either pull from Lahman or copy from BR).
3) input seasonal RA (see above)
4) input season NRA (either pull from Prospectus website automatically somehow or manually enter)
5) input season DERA (either pull from Prospectus website automatically somehow or manually enter)
6) point to the cells for the pitcher that contain team games played for each year (I have another worksheet in the spreadsheet that stores this for every team)
7) point to the cell for the pitcher that contains his teams park factor for that season
8) point to the cell for the pitcher that contains his teams league runs allowed per game.
9) run the 'solver' cell Rob Wood turned me onto.
I already have tables in the worksheet that contain the data for converting pennants added, the top X number of pitchers in the league in IP, etc.. I've got 73 pitchers in the consideration set right now, once I get caught up it should be easy to maintain. I'm thinking at most 10 minutes per pitcher, which would mean 12 hours of work, once I finalize how I want to do it.
What I do right now, is edit the cell, then copy down until the player changes teams and do it again.
For Walter Johnson it's easy.
Under the "tG" column (team games), I enter "=Games!Y33"  Y33 on the games worksheet is the cell that says the SenaTwins played 154 games in 1907. Then I copy that down through all of the other cells for Johnson.
I do the same thing for park factors and league runs allowed.
Now in the template, I already have "=Games!Y33" in there, and I just adjust the Y33 to whatever is appropriate for the new pitcher. If it's Ron Guidry, I adjust the Y33 to R101 (Yankees/1975) and go from there.
What would be nice would be if I could just add a column for team and another for league for each pitcher season, and the cell would just know where to look, based on instructions, it could look at the Year, Team, League and know which cells, without my having to change it.
For Johnson and Guidry, this takes seconds, for Bobo Newsom, it will take a lifetime.
If anyone can help with that part, please let me know. Thanks!
I would say that my initial position, before reading, is that I don't mind setting the league leaders for innings pitched as the standard for that year. One pitcher in a year can be an outlier (Wilbur Wood in 1971 for example), but the 5 or 10? That's the era norm for top pitchers, which is what we are evaluating.
If we had stats on IP per GS for every league ever, I agree that could work (actually batters faced would be even better).
I would say ideally you could use say drop the #1 guy in the league from the equation  so you don't have a Wilbur Wood throwing the numbers off and changing the average. I think once you get past the top guy or two, it starts to level off.
Looking at 19011977 NL and AL, here are the averages of the top 5 in each league for each season, the number in parenthesis is the distance to the next guy up the chain:
NL1: 322.4
NL2: 303.4 (19.0)
NL3: 291.8 (11.6)
NL4: 282.8 (9.0)
NL5: 277.1 (5.7)
AL1: 322.9
AL2: 301.4 (21.5)
AL3: 290.2 (11.2)
AL4: 281.2 (9.0)
AL5: 274.4 (6.8)
After the first guy throws the number off, it follows a pretty steady progression with the difference getting smaller and smaller. I'd say drop the #1 and #2 and set your league norm based on the others.
So maybe drop the first 2 in an 8 team league, first 4 in a 16 team league. # of pitchers to base it on based on league size:
8 team: 3,4,5,6
10 team: (3*.5),4,5,6,7,(8*.5)
12 team: 4,5,6,7,8,9
14 team: (4*.5),5,6,7,8,9,10,(11*.5)
16 team: 5,6,7,8,9,10,11,12
Basically drop the top pitcher for every 4 teams, and take .75 pitcher for every team in the total.
Then normalize your pitchers to that standard, and pick the normalized number (Prospectus' 275) based on the all time average of the answers.
Now I'll read the Bunning thread and completely change my mind.
Also, by doing it for leagues as opposed to seasons, you inherently adjust for things like higher scoring leagues, DH (allows starters to throw more innings), etc.. Unfortuntately, it's still easier to throw more innings in a pitcher's park than a hitter's park, but I'm not sure how we'd adjust for that, or if it's worth the trouble.
By taking the top 5, if it's a year that's stacked tight at the top  say the 1990 NL), where 25 are within 5 2/3 IP of each other, they all get promoted to the 275 level. But that 275 is established based on a bunch of historical outliers, so they are getting unfair benefit for that. By using my system or yours, this doesn't occur.
I have a strong suspicion that WARP's assessment of AL pitchers vs. NL pitchers in the late 40s early 50s is distorted by this very issue, although I haven't had a chance to study the matter yet. In the NL, you have two workhorse outliers in Roberts and Spahn, where in the AL you have a lower, more tightly bunched group (I think). Thoughts?
When I'm done, I'll be able to provide two numbers  my system's number and WARP's. We'll be able to how much the differences between the leagues change . . . stay tuned!
I would probably want to swap runs allowed out with something like Component Runs Allowed though, since relievers RA totals are distorted by their coming in the middle of innings. Or maybe an average of the two (RA and CRA). Does that make sense?
Year NL AL DIF
1946 250.2 298.4 48.2
1947 271.4 273.3 1.9
1948 269.6 265.4 4.2
1949 268.0 283.4 15.4
1950 292.4 265.8 26.6
1951 298.4 259.8 39.6
1952 279.0 284.6 5.6
1953 270.2 269.9 0.3
1954 279.8 261.2 18.6
1955 257.0 246.2 10.8
1956 284.8 271.8 13.0
1957 261.0 254.0 7.0
1958 271.8 251.4 20.4
1959 281.2 244.0 37.2
1960 275.0 262.8 12.2
Now my system  which wouldn't be normalized to 275, all we care about right now are the differences between the leagues.
Year NL AL DIF
1946 235.5 271.3 35.8
1947 257.0 257.4 0.4
1948 249.5 265.4 15.9
1949 249.3 270.5 21.2
1950 278.3 253.3 25.0
1951 284.5 252.0 32.5
1952 253.8 266.3 12.5
1953 242.0 260.6 18.6
1954 258.0 254.5 3.5
1955 239.3 237.0 2.3
1956 267.8 260.3 7.5
1957 249.3 244.5 4.8
1958 261.0 244.5 16.5
1959 272.5 237.0 35.5
1960 271.3 250.8 20.5
For mine, the baseline drops to 265.1 alltime for NL, 264.7 for the AL.
I don't if this means anything . . . but using the top 5 in the league as the baseline, the sum of the absolute values of the annual differences between the leagues is 1544.7, or an average of 14.7 IP per season.
Using my system the total difference is 1391.3, or 13.3 IP per season.
I would think the closer the two leagues are in any given year the better, to a point  obviously run environments make it much tougher for starters to pitch as many innings as others and the two leagues shouldn't always be even.
Johnson's NRA and DERA were 3.32 and 3.31. Under my system they become 3.18 and 3.16 (some rounding issues are the reason for slight difference in the delta).
This makes sense as great pitchers from lower scoring eras (Johnson's career league R/9 was 4.14 park adjusted) have their win impact understated when you use a 2.0 exponent.
His translated innings changed from 5195.7 to 5202.7, he picked up 7 innings under my system, not much of a big deal there.
I get his career RSAR at 1579. That splits as 1495 PRAR, and 84 BRAR and This works out to a 165.5 Translated WARP and 1.887 Translated Pennants Added (using NL 18761943 as the Pennant context).
That's 84 RCAP or BRAR (same thing) vs. 96 in the Sabermetric Encyclopedia. If we are going to say that Johnson would have pitched 5202.7 IP instead of 5914.7, we need to reduce his offensive opportunity by the same amount (I think). If in a normal context he wouldn't have pitched as much as he did, he also wouldn't have hit as much.
I'd love any feedback if you have it.
I'll also try to update the pennant context through 1977.
As the teams get bunched together and the quality of the pennant winners moves closer to 500, the value of big seasons increases, as its easier to flip a pennant.
At some point, I'll have to add the AL to the mix as well. I'm wondering what impact the Yankees will have on all of this.
Does all of this assume that the value of pitching has remained more or less constant throughout history? Or maybe I should say, has the value of pitching remained more or less constant in the modern era, at leaste.g. since 1900 or so, or since 1893?
And all of this is concerned more or less with the distribution of that value among individual pitchers?
Is that accurate? Or has the value of pitching changed substantially since 1893 or 1900?
Of to put it another way: By normalizing IP so as to not disadvantage modern pitchers, are you increasing the cumulative value of pitching in the modern era?
Seems like I should know the answer to that after how many years of HoM duty, but if I've learned anything it's not to assume I know very much.
I know that sounds a little wishywashy  it's supposed to.
For something like pitching, I think it's important to normalize context. I don't think that being a pitcher in say 1913 when pitching was worth X amount  compared to fielding, should be an advantage or disadvantage to being a pitcher in 1976 when pitching is worth Y amount relative to fielding. You job, no matter what your birthday is to prevent runs. How well you did that, compared to your peers is what's important to me.
It's a little different with individual fielding positions  the difference between being a 3B in 1907 and being a 3B in 2006 is ginormous. So different that someone who is likely to be chosen to play 3B in 2006 wouldn't have been able to in 1907. That's the difference for me, and where I draw the line.
I hope that helps. So I'm not 'assuming that the value of pitching has remained more or less constant throughout history'. It most likely hasn't. But I think when trying to compare pitchers across eras, you have to act as if it has, or you are penalizing guys with the same job based on their birthday.
Suppose team data (such as SenaTwins) is in column A and the year (such as 1907) is in column B. Then you need to insert a new column C which is simply =A&B so in the example above you'd get SenaTwins1907 in column C for that row. I know it looks ugly but Excel will know that this is the row for SenaTwins in 1907. What I mean is that the formula for cell C1 would be =A1&B1;, the formula for cell C2 would be =A2&B2;, etc. (same thing will be done on the pitcher sheet below). Just enter the formula in C1 and drag/copy the rest of the column.
You'll need to create a named array of all the teamseason games played. Suppose the team games data is in column D and that you have 2000 rows of data in rows 2:2001. Then select (highlight) cells C2:D2001 and go up to the Insert menu, select Name, select Define, and then give the array a name (such as TeamGames) with no spaces. The cells that you highlighted should go in the "Refers To" box automatically. The idea, of course, is that you'll use this TeamGames array to look up each pitcher's team games based on the team and season.
Over in the pitcher's sheet you can create the similar concatenated column for team season too. Suppose you have the pitcher's team name in column E and the season in column F, then insert a new column G that is =E&F (see above for details). So you'll see SenaTwins1907 for Walter Johnson in 1907.
Then in the team games column, create a formula using the VLOOKUP function. Suppose the first row you need team games for is row 2 in your spreadsheet and SenaTwins1907 is in cell G2, and team games is column H. Then in cell H2 enter the formula =VLOOKUP(G2,TeamGames,2,false). G2 is the cell you are trying to get information on, TeamGames is the array of team games data you are trying to reference, 2 means that you want the 2nd column of data to be returned (in our case the first column of the array is the teamseason such as SenaTwins1907, and the second column is the number of team games for the Senators in 1907 or 154). Then just drag the formula down to copy it to all the cells you want team games for in the spreadsheet.
This may sound a little complicated, but it is pretty easy once you get the hang of it. Plus, VLOOKUP (and its brother HLOOKUP) is very powerful and has many applications.
Let me know if this works or if my description is not clear.
I'll try that, but probably not before next week . . .
************
Paging jimd (and anyone else with an opinion on this) . . . I believe you've mentioned in the past that the adjustment for league quality shouldn't just be (Value * Factor). It should be more like (Value  constant) * Factor. With the contant being the difference in replacement level between the two leagues.
I'm talking about different leagues within the same season, like the 1890 PL, AA and NL; the Federal League things like that. Or adjusting for known weaker leagues, like the majors 194345. I'm not talking about comparing 1949 to 1953 or 1923.
So my questions are 1) what should the constants be for the AA, FL, and war years 2) what are reasonable factors.
Along those lines are the WARP3 adjustments reasonable for this type of quality issues? I'm thinking of seeing what a players surrounding year WARP1WARP2 differences are, then adjust for war, or weak league based on how much bigger the difference is in the questioned year. Does that make sense?
I agree with Rob, use concatenation and vlookup.
A couple of experienceworn suggestions about them. If you already use them a lot you might want to skip it. If you are concatenating and your worried about corresponding data having punctuation in it, this is not a problem. So if you had a first name and last name column where Walter and Johnson, resided, then instead of a2&b2; returning WalterJohnson, use b2&", "&a2; to return Johnson, Walter. I find this really helpful especially if I've made the killer mistake of not making all my data match the exact same formateliminates those pesky "VALUE!" or "NAME!" errors.
Also, when using Vlookup, you can nest a concatenation within it as needed. BUT the thing you need to know about vlookup is that it can be immensely draining on excel's computational resources. If you are pulling from a dataset that is especially large (like, let's say, a list of WS for every player for each year of the game), you may not be able to save the document which REFERS to the table (aka: the document in which the vlookup is originated, not the data.)
If you get the "can't save, not enough memory" error, try slicing your dataset smaller in the lookup function itself. So rather than selecting all 120 columns in my example table above, you might select only the ten you need for your lookup to work.
One last thing about vlookup (or any formula that returns a value if a condition isn't met). Let's say that your formula is
vlookup(a2, games a1:b1000, 2, false)
If you change the "false" to a zero, "0", then the formula will return a zero if it can't locate the data it's looking for. But because sometimes it's annoying to have zeros all over the place, you may want nothing to appear. In which instance
vlookup(a2, games a1:b1000,2, " ")
will return an empty space. Especially helpful if you are doing a count or countif function in conjunction with the result of the lookup and you don't want unconditional zeroes getting in the way.
Oh, one more thing. Make sure you lock down the target data range if you're using the same range for everyone but copying the formula from cell to cell (or worksheet to worksheet). You'll kick yourself if you don't (I've got bruises on my arse to prove it!)....
vlookup(a2, games $a$1:$b$1000,2," ")
Excel's very helpful feature of changing data ranges along the vectors at which you're cutting and pasting is then disabled for that formula, preventing that awful feeling in your tummy when you realize an hour into things that the data that's coming back seems kind of funny.
Use whatever replacement level standard is also being used for the hitters.
People will compare them to the hiters no matter what caveats you state.
So my questions are 1) what should the constants be for the AA, FL, and war years 2) what are reasonable factors.
I haven't worked much with the latest WARP numbers yet. And not for those years. (I don't revise everything; it's too much work. Though I will revisit a VanHaltren, etc.)
A .383 team will go 62100, and earn 186 WS. If all dept's of a team are equally bad, the distributions of the team shares will still be typical. There would 89.3 BWS, 31.4 FWS, and 65.3 PWS. This replacement level implies that a replacement level position player that plays all 162 games earns between 14 and 15 Win Shares (shares just for playing, not for being any good). The hitters Pennantsadjusted would need to be recalculated based on that level.
Wait a minute  I'm not doing that. Setting pitching at a .385 WPct assumes an average offense. An team with an average offense and replacement level pitching would lose 100 games is what I'm setting the replacement level at.
I've always felt a fulltime replacement player would get about 8 WS (7 with the DH) and a 220 IP pitcher at replacement level would get about 7 WS. I was trying to err to the side of not setting it too low  I should have thought it through under those terms. I know that shows pitching replacement being a little higher than position player replacement level but I don't think that's unreasonable.
Let me think that through again though. A true replacement level position player will hit at replacement level but field at average level.
An average team in a 162game season: 116.6 offensive WS, 41.1 D and 85.3 P Win Shares. That gives you 19.7 WS for an average position player in 162 games in a nonDH league, 18.1 in a DH league and 12.9 for an average pitcher. Drop that to 18.7 for a 154 game season position player and 12.25 for a 154 game season average pitcher (that's over 209 IP). BTW as a side note that should dispell any myth that Jake Beckley wasn an average player  he was in the 2027 WS prorated to 154 game seasons over his career, not in the teens and he wasn't playing every single game every year. Sorry for the digression, but it's an important point.
If the team remains average on offense and fielding, it would take 28.3 pitching WS to drop them to 100 losses. So I'm setting my replacement equivalent to 4.3 pWS over 220 IP being replacement level. That's probably too low.
If 7 WS per 220 IP is replacement level, then a team with average hitters and fielders would win 68 games with a replacement level staff. That would mean setting .420 as replacement level, assuming an average offense.
Now lets reverse it and see what a replacement level hitter would do to a team with average defense/pitching.
Setting replacement level hitters to where a team with average pitching and fielding loses 100 games, means 59.6 bWS. That sets hitters at 7.5 WS + 5.1 for fielding or 12.6 WS per 162 games  11.8 in a DH league.
Bumping it up to the 68 win mark like we did for pitchers would make their replacement level 14.8 WS, or 13.8 were it a DH league. That's too high.
What to conclude from this  pitching replacement level  in terms of the record of a team with all replacements as pitchers and average everywhere else  is probably higher than position player replacement level  at least under the WS system.
There's no way pitching replacement level is as low as 4 WS/220 IP. And there is now way that position player replacement level is 15 per 162 games. Just look at some 15 WS position player seasons if you don't believe me. And take a look at how bad you have to be to get 4 WS in a 220 IP season.
There's a logical explanation for this apparent paradox, IMO. It's that WS only gives 1/3 of the credit (35.1% to be exact) to the pitchers. Combine that with the fact that no one at the major league level with any signficant time is a replacement level fielder AND hitter, and that's what you get.
It doesn't surprise me at all that to get their replacement levels equivalent (for a full time player or a 220 IP pitcher) on a per player level, a team with replacement level hitters dragging down 48% the team would do worse than a team with replacement level pitchers only dragging down 35.1% of the team.
So here's where I'm going. If you set a .225 team at replacement level (a little worse than the 1962 Mets), with all things equal (batting and pitching replacement level, fielding average) you get:
Hitters 39.4 WS, Fielders 41.1 WS, Pitching 28.8 WS. That sets position players at 10.1 WS, (9.5 in a DH league) and pitchers at 4.4 WS.
Think about it though  James was wondering why pitchers came out too low. Hell most of us think that. That's why  when you adjust for replacement level the pitchers get the boost they are in need of.
BTW, that's probably too low, but it's where you are if you set the pitchers and hitters as equally bad.
If you want to bump the pitchers to 6 WS being replacement level you get a team at .278 WPct (45117). I think that's fair  the 1962 Mets certainly had some players that were way below replacement level  certainly many players that couldn't get time in other organizations were better than what the Mets put out on the field that first season.
So there you have it. I'm going to be setting my pitcher replacement level to 6 WS per 220 IP. My position player replacement level becomes 11.9 in a nonDH league, 11.2 in a DH league. Over a 154 game season, I'll go with 11.3/10.6/5.7 (the 5.7 is over 209 IP, not 220, in the shorter season).
What does this mean for my expected team WPct to use in this massive pitcher spreadsheet? Well a team with an average offensive and defense, and pitchers that pull in 6 WS per 220/IP would win 65.8 games, or play at a .406 clip. That means that I'll be using 5.48 as my replacement level aDERA  which is the equivalent of 6 WS in a 220 IP season.
Having looked at it this way, I'm pretty surprised that the replacement level for a position player comes out that high, but it makes sense. I mean an average season is generally about 22.5 WARP. A replacement player is about 2 to 2.5 in TPR. If an average position player gets about 19.7 WS in a full 162 game season, it would make sense that a replacement player would be about 8 WS below that.
I can't believe it took me 45 years of working with WS to approach it this way, thanks for triggering it jim!
Or on Tuesday?
My biggest fear with the 'open membership' thing is that more start joining in once they recognize the names being considered, basically destroying the candidacy of any borderline guys from way back when.
Ditto.
>But they should be able to look at least at the current backlog, and study those players and their effect on the game at the time.
Ditto ditto.
I only had a chance to lightly read your post as it is quite a dense one, but I do have a few comments.
1. James didn't use a strict split on all teams. if a team had an average offense and a replacement level pitching staff a higher percentage of the teams WS would be offensive. For intance 35% of that teams WS woudl not go to pitcher since their pitching staff was at replacement level. I didn't catch that you factored this in but I could be wrong.
2. While Beckley had a number of seasons in the 2025 (I have no season in which Beckley had 27 WS even when schedule adjusted) he had nearly as many in the 1520 range, which by your analysis is about average if only slightly above. And of course he still was never an MVP caliber player.
And I dont' want it to look like we are shunning possible contributors here either. Of course there is a fear that some people will join to vote in their favroite HOF cases form thier childhood, but it shouldnt' be too hard for any new voter to take a look at the 19th and early 20th century players as well. This is why we force new guys to turn in a ballot, gives us a chance to ask them why they may look over Duffy, Beckley, GVH, Waddell, etc. They may have really good reasons (all of these players have warts, especially Beckley ;)) or they may have simply forgot/overlooked/are underrating them.
I'd then suggest voters try to read the threads of each candidate, especially Negro League holdovers getting votes, and even better if they can peruse some old ballotdiscussion threads as well. Particular emphasis should be paid to the top 25 or so; a new voter needs to recognize the many pros and cons of each (that's why their holdovers still getting decent support), and come to a decision on which side they fall.
What they would not necessarily have to be are the ones to resuscitate the prospects of Jack Chesbro or Billy Nash or Joe Tinker or Oliver Marcel (not that there's anything wrong with that).
That's meant as a welcome to prospective new voters, with a rational caveat. If it's asking too much of them, they're free to keep just 'auditing' the course. If it's manageable, then by all means join the party!
lol
Childs and Bresnahan, too. :D
When I joined, I first examined every player who had received votes in the last election, plus the newly eligibles. Then, when I had more time a few weeks later, I started poring through the Bill James Abstract looking at other players who had made his lists and sifting through them using my own standards of excellence which has allowed my list to expand and become gradually richer and more refined.
Going over your ballots, you definitely are not ignoring the earlier candidates, which is the fair thing to do.
Thanks for taking the time to read it. I understand it my have been too 'dense' very hard to keep it 'undense' with the subject matter :)
Anyway, I understand what you are saying, but it's not an issue in this case. James does ties the splits to certain things like how much over the margin teams were on offense/defense, how many K/BB/HR the pitching staff allowed, etc.. But since I'm starting with an average team  and moving the team off average by adjusting the offense or pitching win shares, it shouldn't make a difference. In effect, when I drop the team from 116 bWS to 60, I'm changing the overall offense/defense split, etc.. So I think I'm covered there.
*************
Regarding Beckley, using WS from the WS Digital Update, and adjusting based on team decisions (not games) to 154 decision seasons, since WS is based on decisions not games, I get:
YEAR WS
1888 16.0
1889 21.6
1890 25.5
1891 18.3
1892 19.4
1893 20.1
1894 20.4
1895 20.7
1896 11.8
1897 17.6
1898 14.5
1899 20.7
1900 23.6
1901 19.4
1902 19.7
1903 18.3
1904 22.7
1905 15.7
1906 4.7
1907 0.3
18.7 would be an average player if he played every single game. Beckley didn't play every game every season, obviously.
Adjusting for his games played (which treats every game as a full game, but it's close enough) as a percentage of his team, he was above average every year from 1888 through 1904, except for 1896 (he was 3 below average that year). In 1891 he was only .1 WSAA and in 1898 he was .4 WSAA.
He was at least 3 WSAA from 18881890, 18991900, 1904.
This is giving no credit for the fact that 1B was a more valuable defensive position in the 1890s and 1900s than it is in modern times.
According to the Sabermetric Encyclopdia, Beckley was 245 RCAP (RC above position) for his career, 330 RCAA.
But it gets interesting when you look at 189396. During those years Beckley was 77 RCAP, 64 RCAA  that's right, an average 1B in those years would have been 13 runs below an average hitter in Beckley's number of outs. WS doesn't account for this shift on the defensive spectrum at all. In 1898 Beckley had an off year, but 1B overall were still slightly below average hitters. Same for 1899. And 1900. And 1901.
Something happened in 1902 and 1B started hitting better than average. But for much of Beckley's prime, 1B were average to below average hitters. Maybe it was the wear and tear of playing in the infield in the rowdy 1890s? Or the bad gloves? But WS does not account for this at all  and it still shows Beckley as an above average to sometimes very good player throughout his career. For his career, 1B AVG OWP was only .531.
Moving to modern times, I just pulled Jeff Bagwell, since his career covers most of the last 15 years (19912004, a little 2005). For Bagwell's career the average 1B had an OWP of .559.
Over George Sisler's career the average 1B had an OWP of .544.
So for whatever reason, 1B didn't hit as well during Beckley's career as they have in the future. WS doesn't account for this, and as such underrates him. I still get him 351 WS when adjusting for schedule. I would think this underrating could have cost Beckley as much as 23 WS per season in the mid 1890s. An average defensive player gets 4.9 fWS in a 154 game season. Beckley who was at least an average 1B was getting 1.53.0 fWS per season. There were several years in the mid1890s where 1B could have been considered at least middle of the pack in defensive responsibility.
Basically, Beckley is beginning in an era when play was VERY different. I think Beckley was quite good at fielding the bunts and slaps and such, and therefore was quite valuable. His arm wasn't always as accurate as you'd want, but handling the 'small ball' seems to have been quite taxing, if you go by the neartotal lack of 1B longevity in his era.
Joe, you are correct. A team made entirely of replacement players would both score runs (hitting+baserunning) and prevent runs (pitching+fielding) at a .383 clip. So the proper determinant of this "team's" wonlost record is the <u>square</u> of .383, which is .14669. The square of a .500 team is, of course, 1/2 * 1/2 = 1/4. So a replacementlevel team would win (.14669/.25) or 58.6% as many games as a .500 team. With a 162game schedule, a .500 team posts an 8181 record. The replacementlevel team wins (81*.586) games, which works out to 47115.
That 47115 record, while not as bad as the Spiders, Mets, or '03 Tigers, seems intuitively correct.
I don't understand why he'd have his replacement player pool that large (as many as the regulars).
Personally, I think replacement level is the bottom 1520% of the regulars. Replacement level players play all the time. So I think replacement level might be a little higher than he indicates but overall that makes sense.
Thanks for the link!
You must be Registered and Logged In to post comments.
<< Back to main