Ratings for Negro League players

[Hoiles]

I really appreciate Negro Leaguers being rated as major leaguers in this year's version, so that Hall of Famers like Satchel Paige are no longer AAAA players as in previous versions.

However, looking more closely, anyone notice that Negro League players might have too much of a boost? I imported a 1927 season (single-season replay at defaults) to test this out. For example, Lou Gehrig and Rogers Hornsby are rated as 5 stars when viewed with "Ratings relative to MLB", but respectively, 4.5 and 4 stars with "Ratings relative to NeL". I also ran a 1620 game (or 10 season) simulation of that year's Yankees (often regarded as the greatest of all time) vs the year's St. Louis Stars. The results were pretty close (I don't necessarily have a problem with the Yankees losing the series), but some of the stats look wonky. For example, the Stars' sluggers went haywire on Yankee pitching and outslugged Babe, Lou and Co in HR by a 2:1 ratio. Mule Suttles averaged 66 home runs in the sim, while Babe Ruth averaged 38 home runs (the same year he hit 60 IRL). Obviously we can't know for sure what would have happened in real-life, but assuming that the AL and the Negro Leagues were roughly the same level, that wouldn't have happened.

I also tried this out in other seasons (1920 and 1946) and get similar results. Any ideas on how to tweak this? I noticed that adjust/weaken hitters and pitchers limits don't seem to have an effect when importing from the minor league database. Perhaps the ratings need an adjustment to account for the variation from the small sample size. In any case, it's still a major improvement from previous years, where the same St. Louis Stars would have averaged 30 wins over a full season vs an NL/AL team.

[LansdowneSt]

What's the frequency of the recalc and over what time period is the recalc done? One year as a basis of the skills is going to amp up the NeL stars but a wider time frame, say three or five year will pull them away from one-year out-of-whack seasons. I've run some career sims over the time period and when retired according to history, the actual numbers come out pretty well.

There's a couple of next steps, the first of which is scrubbing the NeL stats themselves that form the skill basis of the NeL players. Seamheads' database has grown by leaps and bounds even in just the past few years. Some of the new data when it comes in will help smooth it out and moderate the results. It's a first step though one, like you say, overdue for attention and more eyes, such as yours, running both long-term sims and play-by-play under a variety of different conditions. As opposed to a bug, I think it is more of a monitor and adjust period for the NeL part of the game. At some point, once the release if fully in the rearview mirror, I'd like to set up a thread in the Historical Forum where - under clearly articulated settings for control purposes in each post, a bunch of us interested in this part of the game - can run tests under multiple condition sets to provide a data-based approach to what those adjustments ought to be. But it is exciting to be on the cusp of resetting the NeL and glad you are diving in.

It's a shame the pandemic impacted MLB 2020 and the celebration of the NeL that centennial would have brought. And whether it is MLB or baseball-reference or OOTP, so glad it's getting more focus.

[sprague]

Originally Posted by Hoiles

I really appreciate Negro Leaguers being rated as major leaguers in this year's version, so that Hall of Famers like Satchel Paige are no longer AAAA players as in previous versions.

However, looking more closely, anyone notice that Negro League players might have too much of a boost? I imported a 1927 season (single-season replay at defaults) to test this out. For example, Lou Gehrig and Rogers Hornsby are rated as 5 stars when viewed with "Ratings relative to MLB", but respectively, 4.5 and 4 stars with "Ratings relative to NeL". I also ran a 1620 game (or 10 season) simulation of that year's Yankees (often regarded as the greatest of all time) vs the year's St. Louis Stars. The results were pretty close (I don't necessarily have a problem with the Yankees losing the series), but some of the stats look wonky. For example, the Stars' sluggers went haywire on Yankee pitching and outslugged Babe, Lou and Co in HR by a 2:1 ratio. Mule Suttles averaged 66 home runs in the sim, while Babe Ruth averaged 38 home runs (the same year he hit 60 IRL). Obviously we can't know for sure what would have happened in real-life, but assuming that the AL and the Negro Leagues were roughly the same level, that wouldn't have happened.

I also tried this out in other seasons (1920 and 1946) and get similar results. Any ideas on how to tweak this? I noticed that adjust/weaken hitters and pitchers limits don't seem to have an effect when importing from the minor league database. Perhaps the ratings need an adjustment to account for the variation from the small sample size. In any case, it's still a major improvement from previous years, where the same St. Louis Stars would have averaged 30 wins over a full season vs an NL/AL team.

I found the same things when I did a look around of 1930 and 1934.
I do all my work in commissioner mode to look under the hood at actual values.



8 of the top 10 pitchers were negro players, 6 of the top 6 pitchers. And the negro pitchers had far greater stuff ratings than Lefty Grove or later Carl Hubbell.
Willie Wells had a higher HR rating than Babe Ruth in 1930 as did a player with the last name Palm if I recall.
Satchel Paige though is one player who seems to get shafted by the rrating system however. He is one who needs boosting.


I think the negro players were true stars ON PAR with their MLB counterparts, but right now the measurement system seems to make them out to be quite a bit ahead.
Early versions of ootp 20, 21 had the negro players under rated, now it seems the rating pendulum has swung a bit the other way. A slight tweek downward (maybe a bit more for pitchers) and it might be good to go

[LansdowneSt]

Here's three of the players in the Sim I ran two days ago. it was 1920, recalc every year, use three-year recalc as basis, retire according to history. When it came to actual resulting game play, I didn't find any of these results out of sorts. The one-year ratings mattered less to me that the results.

Just need more folks trying it out under various conditions to see how they perform in an integrated what-if type environment. As I said, glad y'all are doing so and providing said feedback.

[David Watts]

Originally Posted by LansdowneSt

Here's three of the players in the Sim I ran two days ago. it was 1920, recalc every year, use three-year recalc as basis, retire according to history. When it came to actual resulting game play, I didn't find any of these results out of sorts. The one-year ratings mattered less to me that the results.

Just need more folks trying it out under various conditions to see how they perform in an integrated what-if type environment. As I said, glad y'all are doing so and providing said feedback.

When you use 3 year recalc, do you change the make bad setting at the bottom of the first page of the setup wizard, or just use the default?

[LansdowneSt]

So, all the wizard inputs were pretty straight forward. The only wizard toggles I used were to turn on NeL and Minors on the first page and I switched to 1-100 ratings. Once the league was created, I enabled DH's as I wanted to see hitters have a chance to hit (so even more offense to distribute to hitters suppressing some HR extremes in a given season), turned off color barrier, turned on retire according to history and everyone came in each year via a draft so they didn't align with historic teams as rookies. I didn't go into advanced and mess with the 300 AB / 40 IN player nerf line for partial years, if that's what you mean by "make bad setting" - though I like the name of such a setting Then I hit do-not-disturb and let it run and went to work...

[David Watts]

Originally Posted by LansdowneSt

So, all the wizard inputs were pretty straight forward. The only wizard toggles I used were to turn on NeL and Minors on the first page and I switched to 1-100 ratings. Once the league was created, I enabled DH's as I wanted to see hitters have a chance to hit (so even more offense to distribute to hitters suppressing some HR extremes in a given season), turned off color barrier, turned on retire according to history and everyone came in each year via a draft so they didn't align with historic teams as rookies. I didn't go into advanced and mess with the 300 AB / 40 IN player nerf line for partial years, if that's what you mean by "make bad setting" - though I like the name of such a setting Then I hit do-not-disturb and let it run and went to work...

Yup I was talking about that nerf line for partial years. Extremely cool. I didn't know it was even possible to use the retire according to history if you also used minor/NL. Learn something new every day.

[Hoiles]

Thanks for the input, Lansdowne. I'll probably be messing around with these leagues over the year, so I'll definitely be interested in a thread in the historical sims forum. I think many of the issues are due to the lack of quality in available data for these leagues, improvements to the dataset such as the ones from Seamheads should help out.

In a similar way, I earlier did a 1920-1950 sim with the default annual 3-year recalcs and color barrier off. I found that the integrated NeL hitters generally had reasonable stats, but instead (as Sprague) pointed out, the pitchers overly dominated. They filled out almost all the "Top 20 pitcher" rankings every year, as well as the Cy Youngs.

As evidence, the 1920-1950 awards were won by integrated NeL players:
AL MVP's: 3 out of 31 times
AL Cy Youngs: 27 out of 31
NL MVP's: 12 out of 31 (not counting 3 from Larry Doby)
NL Cy Youngs: 22 out of 31

Without knowing about the under-the-hood stuff in the import and sim engine, I think it could be related to some sort of disconnect on how AL/NL players are imported from the Major League database, and how NeL players from the Minor League database (I see the NeLs no longer get a "level nerf" in the CSV files).

I have some guesses off the top of my head for these differences:
- The adjust/weaken settings don't work the same way (which would be an issue as players often had stats from a limited amount of games each year)
- The NeL player stats aren't being compared and adjusted to the appropriate NeL totals, in the same way as they are for the AL/NL
- The batter/pitcher imbalances may have something to do with how the integrated player stats affect autocalcs/league modifiers
- For pitchers, perhaps the "three true outcomes" DIPS model doesn't apply to NeL stats, either due to lack of data quality for detailed stats such as BB/K/HR allowed, or perhaps the DIPS model didn't actually apply in these leagues?

[luckymann]

Check out my thread on this HERE. From Post #113 covers v23.

LSt and I ended up with very similar stat outputs in our sims even though he used 3-year recalc and I used 1-year.

G

[David Watts]

Man I wish the Negro Leagues could be made an option by itself not including historical minors. I also wish the players could be made available for random debut drafts. Historical minors was such a cool idea, but it just gets way too massive in size through the years.

[LansdowneSt]

Originally Posted by David Watts

Man I wish the Negro Leagues could be made an option by itself not including historical minors. I also wish the players could be made available for random debut drafts. Historical minors was such a cool idea, but it just gets way too massive in size through the years.

While I don't know the practically it of it, I've been thinking a lot about this very thing. It would allow processes/modifiers/options for just that limited set of players. That way a change that isn't to a player's row in the database but must apply to the dataset as a whole could be done to just the NeL without consequences to the traditional minor leaguers. If this were possible, imagine you were at the screen in the historical game set-up wizard, it would ask NeL 1915- (Y/N) and separately Real Minor Leagues 1915- (Y/N).

As the game looks to update the stats going-forward, it's a good time to consider this. This could also include updating some of the more obtuse naming conventions for importing the NeL'ers. I understand why Spritze created those but if the NeL is undergoing a reset in the years ahead, aligning with bRef is a good idea. I don't mind renaming a few facegens to conform.

The RD thought you have would be nice as well. Could RD pull from two databases? Some folks have asked for that before, with the opening of the MiLB to be included too. The fact it hasn't been made an option implied it might be tricky to pull from anything other than the csv where just the "MLB" players reside. But if the NeL'ers are done/modified/imported in an accurate and balanced way, someone will have a master excel sheet with a built-in random number generator, and one can easily import X number manually to supplement the traditional RD drafts depending whose number gets called.

[sprague]

So I have looked under the hood. Pitchers are the biggest factor out as mentioned. So I loaded 1930, no scouting, and here is the top 3 negro league pitchers.

[sprague]

Now here is the top mlb pitcher for 1930, lefty grove, and he does not come close to the negro players.
Then I provide the overall list of starters dominated by negro league pitchers.

[LansdowneSt]

Thanks for posting those, sprague.

This is also one of the reasons I'm anxious to test the settings with the Seamheads data being inputted to replace the older OOTP real stats in their NeL database.

Take Bullet Rogan, for example, here's his Seamheads page and here are the stats feeding Bullet in 1930. The OOTP 1930 year isn't his or isn't sufficiently sourced to qualify for inclusion in Seamheads. I'll spot check some of the others - and there were meteoritic, excellent years for NeL pitchers - but 1930 for Bullet Rogan won't be one of them once the Seamheads data is utilized.

[LansdowneSt]

I looked at Helio and Willie too and they only imported as three-star players (scouting off and using a 3-year recalc) for me in 1930. There are similar issues in the real stats though to a lesser extent than what Bullet had. Their OOTP real stats had lower ERAs, higher SOs and lower BBs than what subsequent research into those years for those players have shown per Seamheads.

Are the older, incomplete, and sometimes just incorrect real stats (well, in fairness they were what was known when the OOTP database was populated) the whole issue? Probably not... but it's certainly the next logical step in getting the NeL where it needs to be. Thereafter, it will be much easier to isolate game elements that may or may not be translating those stats into the game appropriately.

[luckymann]

Here they are in my full sim.

Heliodoro Diaz

39.2 career WAR vs 13.7 from SH and 21.1 given him by Eric Chalek in his MLEs




Willie Powell

44.7 career WAR (more than a quarter of this in 1928) vs 8.5 SH / 9.2 MLE




Bullet Rogan

46.6 vs 61.7 SH / 85.4 MLE

[luckymann]

Also, have you noticed how Bullet's IRL stats import with no age correlation?

[LansdowneSt]

Originally Posted by luckymann

Also, have you noticed how Bullet's IRL stats import with no age correlation?]

Yup, a missing data field somewhere. More at issue is the extra 1,000 innings of stats in Bullet's OOTP Real Stats vs. Seamheads. That's the reason revision of the stats is the next task ahead. Takes time. At least this aspect of the game is finally on tap to get some attention. That will come once there is bandwidth, I'm sure.

[luckymann]

As I'm sure LSt will readily attest, I've done as much "under the hood" work as anyone with NeL players.

The pitchers benefit from the same criteria that 19th century guys like Toad Ramsay - who just come into OOTP saves and almost always dominate - do.

Getting the mix right on pitchers is the most difficult part of the engine's responsibilities. You'd be amazed at how much each of the data inputs makes to their overall profile - pitch repertoires, first and foremost, but also how much difference even incremental changes to GB%, control and movement make. The era is another key factor.

I find that the raw stats struggle to properly convert into ratings for these guys, almost universally making them stronger than they should be. I've got a bunch of "second-tier" NeL pitchers in my Bucs save all outperforming to various degrees the level you might expect. Have a look at some of these guys, this is thru the 1932 season, with my having curated them using my own methodology.

Active




Retired




I have, as time has gone by, gotten better at figuring out their proper ratings, working with the engine rather than against it and just fine tuning to get things where I think they should be. But even now, they tend to come in slightly above where you might expect. Many come in with low stuff but high CONT / MOV. Here are a couple profiles from my Eclipse League save set in the late 1960s, now in 1971, again curated by me. None of the "big" pitching names have entered as yet.












Compare them to CC Sabathia and Juan Marichal, the two leaders in career pWAR in that save.






You'll notice that, despite the low STU they are still OK on the K9 rates, roughly between 3 and 5 - the high MOV comes into play here to make up somewhat for the low STU - and do pretty well without being overly dominant.

But, as the game relies purely on stats to generate the ratings and, with those stats not having been sourced correctly, the default will see them outperform more often than not. Then again, as someone pointed out, how are we to really now what they would have done as MLBers?

[kq76]

Please keep it above board, people. I moved and hid the nationalistic talk.