Bump Wills.

I should point out that the 6.0 WAR in 1938 is Bill Byrd's peak season. Here's his rank among black pitchers by season

1932 - 27th
1933 - 11th
1934 - 20th
1935 - 27th
1936 - 6th
1937 - 14th
1938 - 8th
1939 - 3rd (5.6 WAR behind Roy Partlow and Jonas Gaines)
1940 - 6th
1941 - 2nd (highest finish, 5.7 WAR, Dave Barnhill first)
1942 - 9th
1943 - 2nd (5.7 WAR, beat out by Dave Barnhill again)
1944 - 7th
1945 - 7th
1946 - 23rd

Compare to another famous spitball pitcher who gets votes here - Burleigh Grimes. Grimes' best finishes among NL pitchers in bWAR were 1st, 2nd, 2nd, 3rd and 3rd. That is a more impressive peak.

Hilton Smith bounces up and down in every MLE revision

I'm not sure he bounces all that much more than lots of players bounce around. The Fosters, Conrado Marrero, Webster McDonald, Davis Rooselvelt, and Roy Welmaker all have bigger shifts; Barnhill and Holland seem to swap around roughly as much with revisions. Bragana is up 7 WAR and 7 WAA. Byrd has generally been fringey to ignore-y, until the last few revisions have him decent to solid.

On the hitter side Marvin Williams, Julian Castillo, Bunny Serrell(!!), Carlos Moran, Ray Dandridge, and George Scales have way bigger moves...in fact there are dozens more with much bigger moves among hitters. There are a not-inconsiderable number of hitters with 20+ WAR swings.

The MLE's get translated into one calculator but white players' numbers get sliced into bWAR, fWAR, gWAR, Win Shares, pWORL, etc. I wonder what would happen if we translated NGL numbers into all of those different calculators.

I wonder what would happen if we translated NGL numbers into all of those different calculators.

Well, at the level of WAR for Negro-League play, without the MLE reconstruction, we can get a look at those kinds of differences right now, with three different versions of NeL WAR being available at Seamheads, Baseball Reference, and Fangraphs.

With respect to using Dr. Chaleeko's MLE pitching data for making single-season comparison, as DL does in post #102 above, I would strongly recommend being cautious about placing much weight on the result of such comparisons at the level of individual pitchers, primarily because we don't yet have a really good technique for modeling and translating pitcher workload from an NeL context to an NL/AL context, partly because we still don't have good data of what that workload actually was or what it meant for pitcher performance.

Since we're talking about Byrd, let me present him as an example.

Byrd is certainly a pitcher who looks like he was probably a workhorse; he pitched year in, year out, with no history of arm injuries. Dr. Chaleeko's MLEs project him, though, with a maximum number of IP in a single season as 250. He is given this total in 1939 and 1940. 1939 is one of his best WAR seasons. In that year in the NL and AL, there were eight pitchers who threw more innings than that, with Bucky Walters topping the majors with 319. I think that it is extremely likely that if Bill Byrd could have been a top pitcher in the NL or AL during this period, his maximum single-season IP would have significantly exceeded 250 in several seasons, but Dr. Chaleeko's system caps MLE IP in most cases at 250 for this period, with a few 260s scattered around.

The amount of MLE WAR that NeL pitchers earn on a seasonal basis is strongly influenced by their seasonal IP totals, and these totals do not closely map onto what an NL/AL pitcher's usage on a seasonal basis would have looked like. We can't readily compare Grimes to Byrd on a single-season basis because Grimes' IP have a range that Byrd's don't, and we can't readily compare Byrd to Dave Barnhill on a single-season basis, either, because their single-season IP totals are highly artificial.

That's not a knock on Dr. Chaleeko's MLEs in particular on this problem. It's the most difficult MLE problem we face.

As data becomes available, as it appears to be starting to through Retrosheet work on Negro-League games, it may become possible to figure out how to model a translation of Negro-League pitchers into NL/AL contexts that makes sense at the seasonal level for purposes of comparison, but we're not there yet.

DL's data on how many pitchers earned a given number of WAR in a single season, on the other hand, is still salient data for assessing the overall plausibility of the MLEs as an estimate of NeL player performance!

The single-pitcher comparisons are also interesting, but I don't think that we can say with any confidence that the pitcher with the most MLE WAR in a given season was the best pitcher that year. To make determninations of that kind, we would need to work with (not just look at) the actual seasonal numbers.

Updated ballot for updated MLEs from Dr C (as well as minor movements on other players):

1. Adrián Beltré
2. Chase Utley
3. Buddy Bell
4. Tommy John
5. Joe Mauer
6. George Scales - new MLEs say he was possibly an obvious miss
7. Sal Bando

* Small drop in quality, but still easily qualified IMO *

8. Kevin Appier
9. Roy Oswalt
10. Bob Johnson

* Drop off in quality here, all these guys are fringey I think *

11. Robin Ventura
12. Brian Giles
13. Chuck Finley
14. John Olerud - a clump of minor(?) stars from the 90s-00s all right here
15. Ron Cey
16. Chet Lemon
17. David Wright
18. Mark Buehrle
19. Sam McDowell - nobody talks about Sam McDowell, in fact he's only gotten 1 vote, ever. Well the expanded ballot let's me make it two. If you're looking at Gooden...why not McDowell?
20. Mickey Lolich

* Just off ballot *

21. Jerry Koosman
22. Cliff Lee
-- Mule Suttles - will go to pHOM in the next few years
23. César Cedeño
24. Dwight Gooden
25. Jason Giambi

-- Others moving from MLEs --

27. Heavy Johnson
38. Hurley McNair
44. Tetelo Vargas
50. Lazaro Salazar
57. José Muñoz - an oversight from DL's list of top pitchers in the revisions? Roosevelt Davis and Barney Brown may be others.
64. Bill Byrd
70. Bus Clarkson
75. Ramón Bragaña

I keep thinking a NGL MMP project could be useful. Negro League players were mostly overlooked or had token representation in our MMP voting. Rank just the players in the Negro Leagues from 1906 to 1946.

I keep thinking a NGL MMP project could be useful. Negro League players were mostly overlooked or had token representation in our MMP voting. Rank just the players in the Negro Leagues from 1906 to 1946.

I think this could be very useful! If this project were to be undertaken, in might make sense to do separate votes for position players and pitchers, just because the analysis of pitchers is going to be complicated. Having a carefully considered, season by season assessment of the top 5 to top 10 position players and pitchers in the NeL could be very useful and illuminating, and would be a good process for developing a solid approach to pitcher assessment at the seasonal level.

News on the merging of Negro League Stats with AL/NL.

https://theathletic.com/4503613/2023/05/11/negro-leagues-statistics-mlb-records/

Behind a paywall, some of the key takeaways, kudos to Kiko's continued work:

"MLB’s plan to integrate Negro League numbers and statistical legacies with its own remains years from completion. More than two years after its announcement, MLB is still in the initial phase of the project: data acquisition.

The league office was unable to reach an agreement with Seamheads Negro Leagues Database, the most complete set of Negro League statistics ever compiled, to use its data. The league ended its protracted negotiation with Seamheads this spring and now intends to use Retrosheet’s nascent database — a work in progress that Retrosheet president Tom Thress said likely won’t be finished for at least five years — as the basis for its records.

Representatives from the league office and Seamheads met on and off over the past two-plus years, but after a meeting around Opening Day this spring failed to result in a deal, MLB elected to pursue Retrosheet as an alternative. According to sources familiar with the negotiations, the sticking point for Seamheads was not compensation but rather concerns about control of the data, how it would be used and who would have a say in its implementation.
​
With Retrosheet’s Negro Leagues database, which Thress began in the summer of 2020, he hoped to provide what did not already exist publicly: game-level stats. For as comprehensive as the Seamheads database is, the data is displayed in season and career totals. “When they say Josh Gibson hit 18 home runs in 51 games in 1936, I don’t know what those 51 games are,” Thress said. Ashwill told The Athletic in 2021 that Seamheads planned to eventually display day-by-day numbers. That has yet to happen. That means others, like Thress, can’t simply look through Seamheads game logs to recreate season schedules. He had to start from square one.

It quickly became clear to Thress that he wouldn’t be able to achieve Retrosheet’s gold standard: the full play-by-play rundown. In many cases, it simply did not exist. Some box scores were available, but other times he’d find a line score with a sentence of summary. “In some cases, we don’t even have a line score,” he said. “There’s a two-sentence blurb that the Memphis Red Sox beat the Birmingham Black Barons last night, 3-2. That’s all you know. The challenge there is, how do we present this data? Because it doesn’t fit in Retrosheet’s baseline format.”

The answer was to modify Retrosheet’s standard. Now, the requirements for a game to be added to the site’s Negro Leagues database are that it be clear when a game was played, where it was played and its final score. “If that’s all you know, that’s all you know,” Thress said, “and we want to be able to present that.” For the 1948 season, Retrosheet found 542 games. It has box scores for 242; the others give as much information as possible.

It is agonizingly slow work, but rewarding. Researchers liken unearthing a box score lost to time to striking gold. Once, while at a speaking engagement in Excelsior Springs, Mo., author Phil Dixon mentioned he couldn’t find a certain 1920s Kansas City Monarchs game in any newspaper in the state archives. An audience member told him bound editions of the local newspaper were kept in the vault at a bank in town. Dixon descended into the basement of the bank and emerged a short time later with photos of the lost Monarchs box score on his phone.

While working on the 1943 Negro American League season, Thress saw a reference in the Chicago Defender to a Chicago American Giants game against the Birmingham Black Barons in Kewanee, Ill. “I lived in Chicago for 29 years,” Thress said. “I had never heard of Kewanee.” None of the three newspaper archives he’s subscribed to have access to the Kewanee Star Courier. Then he discovered it was searchable through the town library’s site. “And damned if I didn’t find a box score for that game,” Thress said. “That was the most amazing thing.”
​
​Currently, the team working on Retrosheet’s Negro League database comprises Thress and four volunteers. Thress builds season schedules. Volunteers fill out a game file for each contest and return it to Thress. More volunteers might help, Thress said, but the bottleneck is at his desk. This isn’t the only thing he’s working on. On a recent Saturday, Thress was proofing the 1913 AL/NL season when he paused to discuss the Negro Leagues project with a reporter."
​

Bleed, thanks for linking that! I like that you included the Kewanee story. That's one of my favorite stories from doing this. Incidentally, I found not one, but THREE Negro League games played in Kewanee in 1943. It ended up being a little tournament. In June, Birmingham beat Chicago. In July, the Clowns beat Memphis. So, in August, Kewanee matched up not only the two winning teams from earlier in the season, but the two winning pitchers from those games.

Here are all three games.

110. Kiko Sakata Posted: May 13, 2023 at 10:21 PM (#6128215)
Bleed, thanks for linking that! I like that you included the Kewanee story. That's one of my favorite stories from doing this. Incidentally, I found not one, but THREE Negro League games played in Kewanee in 1943. It ended up being a little tournament. In June, Birmingham beat Chicago. In July, the Clowns beat Memphis. So, in August, Kewanee matched up not only the two winning teams from earlier in the season, but the two winning pitchers from those games.

Here are all three games.

I grew up about a half hour from Kewanee, my mom currently resides, and I try to visit on Wednesday nights for Rolle Bolle during the winter months, small world!

SPOILER ALERT - I have seen the light on Willie Wells, and he'll enjoy his new ranking from me.

whoops, that should be on the HOM SS Discussion Thread, obviously.

Some analysis of George Scales

As there is a surge of interest in George Scales following the latest MLE updates from Dr. Chaleeko, I thought it would be a good idea to take a close look at Scales. My own MLEs for Scales using the Seamheads data, while they certainly present a view of Scales that should cause him to get a close look for a ballot spot, do not present him as a slam-dunk, top-of-ballot player. The conversion factors that I use are somewhat lower than those in Dr. Chaleeko’s MLEs, but the distance between my construction of his value and Dr. Chaleeko’s is quite a bit greater than is usual, so I want to raise some questions about Scales to see if the sources of the discrepancy can be identified and explained.

Here's the basic issue with Scales. Dr. Chaleeko’s calculations for Scales find that he would have earned a major-league equivalent 84.4 WAR in his career. My calculations come up with an MLE for Scales of 67.9 WAR.

For Mule Suttles, an exact contemporary of Scales, Dr. Chaleeko’s system finds Suttles’ career MLE to be 68.5 WAR. My calculation for Suttles finds his career MLE to be 65.9 WAR.

I think that the difference is occurring in Dr. Chaleeko’s calculations and not in mine. If I do a season-and-fielding-adjusted projection of Scales’ and Suttles’ Seamheads WAR by (a) dividing the fielding WAR by 1.75 to convert it to the Total Zone scale used by pre-contemporary bWAR and fWAR and (b) season-adjusting each season to 154 games, I get these results:

Scales: 92.2 adjusted NeL WAR
Suttles: 89.6 adjusted NeL WAR

My system finds Scales’ career MLE value to be .736 of his NeL season-adjusted value.
Dr. Chaleeko’s system finds Scales’ career MLE value to be .915 of his NeL season-adjusted value.

My system finds Suttles’ career MLE value to be .735 of his NeL season-adjusted value.
Dr. Chaleeko’s system finds Suttles’ career MLE value to be .764 of his NeL season-adjusted value.

Scales’ career was 1923-46; Suttles was 1923-44. They were not always in the same league, but I doubt that the superior competition level Dr. Chaleeko’s analysis has found for the ECL over the NNL in the 1920s is sufficient to account for such a large difference between the players in the ratio their NeL value to their MLE value.

A next step in the analysis would be to look at career segments. Because Dr. Chaleeko’s MLEs include regression, single-season values don’t map directly onto one another. There are good reasons to include that regression, and so we shouldn’t expect to find a constant ratio between Seamheads’ NeL WAR and MLE WAR. With competition levels varying also from year to year, the expectation of a constant ratio is further diminished. Still, it could be illuminating.

I haven’t done a full analysis of this kind yet. I have, however, started it by looking at Scales’ peak from 1929 to 1932, because Scales’ 9.0 WAR in his 1929 MLE from Dr. Chaleeko stands out as a historically huge season.

Here’s what a fielding-and-season-length adjusted projection of Scales’ 1929-32 seasons looks like in Seamheads’ WAR

Year – Raw – G – Proj. G – Proj. War
1929 – 2.8 – 66 – 154 -- 6.1
1930 – 3.1 -- 55 – 154 – 8.1
1931 – 2.3 – 48 – 154 – 7.4
1932 – 0.7 – 28 – 154 – 4.6
26.2 projected WAR in 616 games
Rate = 6.55 WAR/154

Here’s what those seasons look like in Dr. Chaleeko’s MLEs
Year – WAR
1929 – 9.0
1930 – 6.9
1931 – 5.2
1932 – 3.7
24.8 projected WAR in 582 games
Rate = 6.56 WAR/154

For these four seasons, Dr. Chaleeko’s MLEs estimate that George Scales would have been just as much above replacement level in the NL as he was in the Negro Leagues. This seems unlikely.

DL in MN has already flagged this period in general and 1929 season in particular as ones that Dr. Chaleeko’s quality of play measures appear to overrate somewhat relative to others.

There may be good reasons why Scales’ MLEs come in significantly higher in relation to his Seamheads’ WAR than Suttles do, and it would be good to see what those reasons are, if they are there.

Until such reasons are identified, I would suggest being cautious in taking Scales’ MLEs from Dr. Chaleeko at face value.

I can buy Scales as a little better than Suttles. I'm not buying Scales as Chipper Jones / Charlie Gehringer.

I think it's helpful to use the numbers to rank the Negro League players among themselves first, then try to fit them into the larger context.

I agree with DL on this point. And thanks to Chris for his valuable post above (and of course to Dr. Chaleeko and everyone else who works on Negro League stats/evaluations).

I think it's helpful to use the numbers to rank the Negro League players among themselves first, then try to fit them into the larger context.

This is a great point. And, of course, arguing that George Scales is similar in value, maybe slightly ahead of, Mule Suttles is still a pretty compelling Hall-of-Merit argument.

Bringing this to 1929, which appears to be the big anomaly year flagged by Chris, Retrosheet hasn't gotten back that far, so I can't really add anything authoritative there. But I will note that, according to Seamheads, George Scales ranked 25th in the Negro Leagues in WAR. Two teammates of his on the NY Lincoln Giants - Connie Rector and Charlie Smith - ranked 4th and 5th. Seamheads has him splitting time between the Homestead Grays (12 g) and the Lincoln Giants (54 g) that season. Is it possible that's generating some double-counting somewhere?

Quick hit on the MLEs - I really like the idea of looking at the NeL players ordinally. For a quick-and-dirty gut check, for the 1920-1948 period, here is everyone who shows up at top 10 in WAR in at least 5 seasons:

Josh Gibson - 14
Buck Leonard - 12
Turkey Stearnes - 12
Oscar Charleston - 11
Martin Dihigo - 11
Willie Wells - 11
John Beckwith - 10
Jud Wilson - 9
George Scales - 8
Charlie Smith - 6
Lazaro Salazar - 6
Dick Lundy - 6
Mule Suttles - 6
Cristobal Torriente - 5
Bill Wright - 5
Willard Brown - 5
Heavy Johnson - 5
Alejandro Oms - 5
Dobie Moore - 5
Bus Clarkson - 5

My initial reaction is that this looks awfully good - the top results are all the top players are exactly who'd we expect, and the more borderline HoMers are mixed in with reasonable candidates or extreme peak guys like Charlie Smith. It's also another good showing for George Scales, who comes in closer to the no doubt guys than the borderline guys. Of course, if the MLEs are off on Scales in particular this doesn't change that narrative, but comparitively he looks strong. The arc in his career looks a little Beltre-esque - one BIG year when he's young and then a really nice run at the back of his career. This look should be a little less prone to single year weirdness (1929 and 1932 look very screwy, and 1924 and 1934 look a bit off too) and still points to Scales as a guy of note.