Strategy Settings (Garlon & Others)

[Brad K]

Originally Posted by Garlon

The neutralized stats are not converted to modern era or anything like that. They are simply adjusting for park effects and as I have mentioned for the bulk of the players their neutralized statistics are very similar because park effects are not nearly as significant as what people think they are.

Well, OK, so then what is the answer to Mantle being rated to hit 9 more HRs than Maris in a year where he hit 7 fewer. (The previous answer was park factors.)

[Syd Thrift]

One possible factor is that Maris was a lefty and Yankee Stadium was a 76 for righties and 100 for lefties that year. I wonder if there’s some issue where switch hitters are neutralized as right handed batters instead of like 67% lefty.

[Brad K]

Originally Posted by Syd Thrift

One possible factor is that Maris was a lefty and Yankee Stadium was a 76 for righties and 100 for lefties that year. I wonder if there’s some issue where switch hitters are neutralized as right handed batters instead of like 67% lefty.

Mantle had a total of 167 ABs right handed, so that would be an estimated 84 in Yankee Stadium in which he'd be expected to hit 8 or so HRs. So give him a right handed park factor of zero and take away all the HRs doesn't account for a 16 HR swing.

Besides, the park factors Garlon has (I'd have to look it up again to be exact) are about 92 for RH HRs and 109 for LH HRs. It only takes a cursory glance at the numbers to disprove the earlier claim park factors caused the difference and this sort of thing - comments on a cause that is easily refutable - has happened several times. I've concluded this is a wear them out until they give up operation. If the users are to find out how the game works and what problems are present - not just the symptoms of problems which we can see but the actual problems - we will have to find it out on our own by testing. We're not going to find the answers asking questions.

[Garlon]

1. From 1871 to around 1905 we still have the quadratic park factors. This is because there is no discrete data available to create discrete factors. For pitchers, the only things that are getting adjusted are HR and pitchers simply did not give up many HR in 1901. So you will not see much of a difference and the park factors themselves will also be quite subtle.

2. We do have L/R data since 1901 but I need to figure out how to get them into the format we need to use them in the game. The current file in the game maybe starts around 1920 or so, I do not remember, but I know we looked into this and realized we had the opportunity to improve this. This is on my list of things for the next game.

3. My test leagues use neutral stats and neutral park factors and 5yr recalc with 200/67 batter imports and 25/8 pitcher import, 3yr fielding, pitcher stamina based on career, coaching off, injuries off, position player fatigue set to high, player personality ratings turned off, and some adjustments to strategy settings for different eras.

4. The 750 run neutralized stats is something Bill James published which then baseball reference decided to use. It uses the quadratic formula to adjust to a run environment. It is not designed to adjust across different seasons or different eras. This is not used in the current game.

As for Mantle, I posted that he hit 58 HR per 550 AB and Maris hit 57 HR per 550 AB in 1961. Mantle gets a favorable adjustment for batting switch but it is not to the resolution that I am going in to look at exactly how many times he batted R at home vs on the road. I said he receives the overall combined Yankee Stadium HR adjustment for 1961.

There is also the fact that OOTP then creates its own estimate for the editor based on a standard number of AB vs RHP and LHP for batters. There are no issues there.

The issue earlier was that Maris does not approach 61 HR when using Real stats and I said that Markus has two adjustments built into the game when using real stats so that HR totals do not get too extreme. One adjustment is HR power in general and another for his own adjustment for the home ballpark. I do not know what process is used there but I brought it up in testing that I felt HR were a bit too reduced for the best HR hitters when using real stats. For the neutralized stats the game does not make those adjustments though. I tried a 1961 with neutralized stats and Maris and Mantle both had over 60 HR when using the default import settings, settings which I think is too high and weaken too many players in the league which causes even more HR to be distributed to the top players. The adjustments made to HR power when using Real stats counteract the default import settings. I personally think the default import settings are too high even for replays but that is just my opinion. I think they are set so high to avoid issues with sample size but I think it ends up reducing the ratings of far too many players. I do not think sample size is even an issue at 150 AB or even lower. The last player to hit .400 with more than 70 AB was Todd Haney in 1995 when he went 30 for 73. If you are using real transactions and lineups I do not see how lower the import settings can hurt because any player who was really good for a small sample of AB still is not going to play more than they should. If you lowered the import settings to 70/50 in 1995, Todd Haney would have great ratings, but would he really get more than about 75 AB anyway?

[Brad K]

Originally Posted by thehef

- If you are using real transactions (thus, players playing in the same ballparks as in real life) and real stats, then the affect will be..

Since you're still lacking an answer despite multiple requests, let me attempt an answer based on my understanding of what has been posted, observations of ratings, and testing.

Assuming one real recalc and no TCR of development, a players rating is adjusted by the information on park factors in the game's era_ballparks file. However this adjustment is largely irrelevant when players are playing in their historical ballparks.

For example (and ignoring the possible effects of the "in a modern neutral environment" statement in the editor concerning raw ratings), if a player hit 40 HRs but his home ball park was hitter friendly to the point he was helped by 4 HRs then his rating is 36. However since in the game his home park is his historic one, he performance will be boosted so his effective rating is 40 HRs.

However if his 40 HR year was an outlier for him - say he never hit more than 22 in another year - his rating is adjusted further to make it extremely likely he won't exceed 40 and extremely likely he won't get close to 40. Another player who hit 35 and has multiple mid 30ish HR years in his career is likely to hit more HRs than the 40 HR player since the 35 HR player will not have his rating reduced as much or probably at all.

Middle range performers even perhaps up to the level of the consistent 30ish home run player get additional performance as there is a transfer of output to them from low AB players. High performance low AB players have their performance reduced as the low sample size makes their raw rating unreliable. Although it could be considered low performance low AB players is also unreliable due to low sample size and thus the output should be transferred to them, it is not. Instead is it sent to players who are not low AB players even though their sample size is large enough that their performance is reliable.

If three or five year recalc is used the 40 HR player's rating for the year will be even lower than the downward adjusted single year rating while the consistent mid 30ish HR player will have ratings similar to single year ratings.

With players playing in their historical parks and with historical lineups, the main effect of the adjustment of the ratings with the park factors file is to change the distribution of player's performance among the various parks. Overall performance is not changed by the park factors. Any change in the per park performance is so low as to not be discernable amid the normal randomness.

In summary, peak HR performance is reduced in players best years. Performance of low AB players is reduced. In both cases the reductions are transferred to players with AB totals that are greater than the low AB threshold. As a group, players above the low AB threshold will perform better (except those having an extraordinarily good HR year compared with their other years),

Except for HR performances like the peak years of Roger Maris and Davey Johnson, casual examination is unlikely to reveal the deviations from historical performance. And people who play "historical what it", perhaps the most popular form of historical play, will find the performance differences lost in the noise caused by players playing for different teams, missed time due to non historic injuries or time not missed due to historic injuries, development settings, TCR, and the other things used to make the game more interesting to a human GM. But it's still there.

OK, I know there are four questions, but as can be seen from the amount of material needed to explain just this one, the others are going to have to wait until my typing fingers get a rest!

In the meantime, amplifications and corrections to the above are welcome.

[Brad K]

Originally Posted by Garlon

As for Mantle, I posted that he hit 58 HR per 550 AB and Maris hit 57 HR per 550 AB in 1961. Mantle gets a favorable adjustment for batting switch but it is not to the resolution that I am going in to look at exactly how many times he batted R at home vs on the road. I said he receives the overall combined Yankee Stadium HR adjustment for 1961.

Why does he get a favorable adjustment for batting RH part of the time? On you numbers Yankee stadium is easier for LHB than RHB for HRs, from memory 1.09 to 0.92.

[Brad K]

Originally Posted by Garlon

There is also the fact that OOTP then creates its own estimate for the editor based on a standard number of AB vs RHP and LHP for batters. There are no issues there.

I'm wondering why this overall based on assumed splits still exists. Yes, pre "real" splits I understand the game took a players overall performance and created splits around the overall. Players with very low ABs against same side pitcher were seriously mis-rated on this method so I created my own formula.

Anyway, the game runs on LH ratings and RH ratings and combined is irrelevant. It's a relic of the past. Expected performance should show as vs LHP and vs RHP because that's what the game runs on.

[Garlon]

If you are using real transactions and real stats the effect will be

I addressed the fact that whenever you use real stats that the game has a built in adjustment the reduces HR power either for all players or for player who have more than X HR per 550 AB. I do not know what type of response anyone is looking for with those statements other than that. If you use real stats your single season HR results will be a bit muted in my opinion. Markus makes an additional adjustment for home parks in there somewhere too. But yes, the entire idea is that by adjusting the HR power accordingly that this will approximately offset the home ballpark influence on performance.

Regarding what you mentioned about outliers, there is no such adjustment. When you are using 1yr recalc the game is only looking at those stats from that season. When you are using 3yr or 5yr recalc it is taking the sum of those seasons and converting it to per 550 AB for the ratings. So in that sense an outlier year gets redistributed across those seasons. However, I just recently did a 1871-2020 game using 5yr recalc and neutralized stats and Andre Dawson hit 48 HR in 1987.

When you get to the editor to look at these things the game is still going to give you some expected aggregate assuming a certain number of those 550 AB are vs RHP and vs LHP to show you some type of approximate value.

[Brad K]

Originally Posted by thehef

- If you are using real transactions (thus, players playing in the same ballparks as in real life) and neutralized stats, then the affect will be...

This is quite interesting. Please recall the previous discussion about with these settings on 1 year recalc Mantle is rated at 9 HRs more than Maris in 1961. Following the idea that finding out how a system performs at the extremes is more informative that how it performs coasting, I set up the above conditions on one year recalc and ran the game 1951 - 2007.

I did this only recently and haven't done a thorough analysis but at a glance there are some astounding outcomes. In 57 seasons...

12 players hit over .400 for a season

15 players hit .390 to .399 for a season

24 players hit .320 or better for their careers

28 players hit 60+ home runs in a season (Maris hit 60 in 1961, Mays is the all time leader with 75 in 1955

23 players have 500+ career home runs. Aaron is first with 775.

[Garlon]

I do not recommend neutralized stats with 1yr recalc games with the default settings. My suggestion for replays with default settings are to use real stats. This is precisely why Markus built in the HR adjustment when using real stats, because those default import settings will redistribute quite a bit of HR to those players. I think 300 AB is too much for the default import settings.

I play plenty of career games with 5yr recalc and neutralized stats and my suggested settings and import values and do not get those types of results. I get very few players making it to 60 in a season and I get about the same number of 50+ players as there really were.

[Brad K]

Originally Posted by Brad K

This is quite interesting. Please recall the previous discussion about with these settings on 1 year recalc Mantle is rated at 9 HRs more than Maris in 1961. Following the idea that finding out how a system performs at the extremes is more informative that how it performs coasting, I set up the above conditions on one year recalc and ran the game 1951 - 2007.

I did this only recently and haven't done a thorough analysis but at a glance there are some astounding outcomes. In 57 seasons...

12 players hit over .400 for a season

15 players hit .390 to .399 for a season

24 players hit .320 or better for their careers

28 players hit 60+ home runs in a season (Maris hit 60 in 1961, Mays is the all time leader with 75 in 1955

23 players have 500+ career home runs. Aaron is first with 775.

It is quite expected that players will often exceed their single season bests when they're rated at or close to their performance in their best years. However career wise they should be close to career numbers.

Here are the top 20 career home run hitters 1951 - 2007 who are retired. Active players are listed but not included in the calculations. Plus and minus indicates their performance in the sim versus real life. Remember, this is with historical lineups and transactions.

Career HRs

Hank Aaron +20
Barry Bonds -18
Willie Mays +76
Reggie Jackson +78
Harmon Killebrew +22
Frank Robinson +1
Sammy Sosa active player
Mickey Mantle +50
Eddie Murray +81
Ken Griffey Jr active player
Rafael Palmeiro +4
Alex Rodriguez active player
Mark McGwire -15
Eddie Matthews +50
Jim Thome active player
Ted Williams +12
Manny Ramirez active player
Willie McCovey 0
Stan Musial +38
Mike Schmidt -39
Frank Thomas active player
Dave Winfield +39
Fred McGriff +7
Willie Stargell +19
Cal Ripken Jr +58
Carl Yastrzemski +32


Sixteen players hit more than historical, three hit fewer, and one was even. In total the group hit 554 more HRs than real life, and average of 27.7 per player.

[Brad K]

Originally Posted by thehef

- If you are using real transactions (thus, players playing in the same ballparks as in real life) and neutralized stats, then the affect will be...

As a conclusion to information presented in several earlier posts, one year neutralized stats setting is likely not to be a solution for those unhappy with the suppression of peak HR years by top players using "real" stats. On a 57 year test with one year recalc playing in historic parks with historic lineups top HR hitters significantly exceed their historic production.

It thus seems likely that them exceeding their historical peak years performance is due to the characteristics of the neutralized stats file or other game characteristics, not randomness as would be hoped. So people doing a one year replay have the choice between intentionally reduced performance (the real stats file) or accidentally increased performance (the neutralized file).

Anecdotal comments from users indicates use of neutralized 3 or 5 year recalc results in diminished career performance by top players. However it is believed these users used setting of a somewhat "what if" historical game, so a test using historic transactions and lineups needs to be done.

At present, the best use of the one year neutralized file is for "what if" historical games with TCR, development, trading, free agency prior to its historic inception, game generated injuries, random retirement, etc activated. That way the randomness created by these settings will create enough noise that the accuracy errors will not be noticed.

[Garlon]

If you had coaching enabled and player personality ratings enabled that can influence results. I also think that the default strategy settings associated with different eras are not necessarily optimal. So when you do these tests you are actually testing many things.

I suggest trying your 1yr recalc test with neutralized stats again but turn off coaching and player personality ratings and make sure player development is disabled too. I suggest changing the batter imports to 75/50 and pitcher imports to 25/10 and see what results you get.

[Brad K]

EDIT NOTE: I'm rerunning this test. I discovered it was run with dev on.

Originally Posted by thehef

- If you are using real transactions (thus, players playing in the same ballparks as in real life) and neutralized stats, then the affect will be..

The following are career HR results using the same testing methods as before except with 3 year recalc instead of 1 year. 1951 - 2007. In the one year recalc test the top 20 retired HR hitters were +554 compared with their historical numbers

Hank Aaron -42
Barry Bonds -135
Willie Mays -36
Sammy Soasa active player
Harmon Killebrew +32
Ken Griffey Jr active player
Rafael Palmeiro +30
Mark McGwire -2
Frank Robinson -28
Willie McCovey +32
Mickey Mantle +11
Eddie Matthews +33
Alex Rodriguez active player
Eddie Murray +21
Mike Schmidt -25
Manny Ramirez active player
Reggie Jackson -54
Ted Williams -12
Gary Sheffield active player
Frank Thomas active player
Dave Winfield +25
Jose Canseco +21
Carlos Delgado active player
Carl Yastrzemski +11
Jim Thome active player
Juan Gonzales +23
Dave Kingman +14

Overall -81
Overall excluding top 3 +132

I don't know what to make of this. It seems rating on a three year average over a career should bring the same career results as rating on one year at a time. Particularly troublesome is the poor performance of the top three, particularly since it became known the "real" stats file reduces the HR performance of top players. Is that happening here too?

Due to anecdotal reports of top players not achieving career HR total in historical games with "what if" features included, I think it can be concluded at this time the reduced performance is in the basics, not the what if factors.

[Reed]

Originally Posted by Brad K

As a conclusion to information presented in several earlier posts, one year neutralized stats setting is likely not to be a solution for those unhappy with the suppression of peak HR years by top players using "real" stats. On a 57 year test with one year recalc playing in historic parks with historic lineups top HR hitters significantly exceed their historic production.

It thus seems likely that them exceeding their historical peak years performance is due to the characteristics of the neutralized stats file or other game characteristics, not randomness as would be hoped. So people doing a one year replay have the choice between intentionally reduced performance (the real stats file) or accidentally increased performance (the neutralized file).

Anecdotal comments from users indicates use of neutralized 3 or 5 year recalc results in diminished career performance by top players. However it is believed these users used setting of a somewhat "what if" historical game, so a test using historic transactions and lineups needs to be done.

At present, the best use of the one year neutralized file is for "what if" historical games with TCR, development, trading, free agency prior to its historic inception, game generated injuries, random retirement, etc activated. That way the randomness created by these settings will create enough noise that the accuracy errors will not be noticed.

What was your adjust/weaken hitters setting while using 1 year recalc with historic lineups? I think they should be set very close to 0 otherwise you will be taking home runs away from some players with less than xxx real life at bats and giving them to players with more than xxx real life at bats.

[Brad K]

I used default adjust/weaken settings.

Do you think there are enough home runs to be taken from the low AB players to ad 554 HRs to 20 players plus HRs to the thousands of others who played during that 57 year span? I don't. If you have some support for the idea I'll look at it.

The developers appear to consider the effect of extra home runs for the most active players to be minimal. If it was significant they'd probably have chosen another way to distribute those home runs.

People generally aren't going to play the game just watching AI toil along with historical lineups. They're gong to play "what if" historical. That means they're not going to use adjust/weaken numbers near zero because it means they and AI will misuse the bench players. What I'm trying to do is get the noise of the "what if" settings out and see what the basic game does.

[Brad K]

Originally Posted by Garlon

If you had coaching enabled and player personality ratings enabled that can influence results.

That's good. Baseless speculation an error was present and implying the unproven error had an effect far beyond its capability, 554 extra HRs for 20 players. You know what that's called! Conspiracy Theory!!!

[Rain King]

It is a pretty bold strategy to constantly insult the person who you are also asking to help you.

[Brad K]

Here is neutralized three year recalc, corrected to have dev off.

Neutralized 3 HRs

Hank Aaron +19
Barry Bonds -24
Mark McGwire +43
Willie Mays -55
Frank Robinson +9
Ken Griffey Jr active player
Mickey Mantle +44
Rafael Palmeiro +7
Reggie Jackson +4
Sammy Sosa active player
Eddie Matthews +48
Harmon Killebrew -29
Ted Williams +17
Eddie Murray +20
Willie McCovey -4
Frank Thomas active player
Jim Thome active player
Ernie Banks -20
Stan Musial +17
Dave Winfield +25
Carl Yastrzemski +35
Alex Rodriguez active player
Mike Schmidt +32
Jose Conseco +15
Gary Sheffield active player
Willie Stargell -5

+198




Here is neutralized three year double weight current year

Neutral 3 weighted HRs

Hank Aaron -23
Barry Bonds -38
Willie Mays -17
Eddie Matthews +92
Rafael Palmeiro +35
Reggie Jackson +20
Mickey Mantle +47
Mark McGwire -1
Sammy Sosa active player
Ken Griffey Jr active player
Harmon Killebrew -1
Frank Robinson -15
Ted Williams +31
Willie McCovey +9
Eddie Murray +22
Manny Ramirez active player
Stan Musial +21
Alex Rodriguez active player
Willie Stargell +18
Frank Thomas active player
Carl Yastrzemski +31
Mike Schmidt +28
Gary Sheffield active player
Ernie Banks -45
Dave Winfield -3
Jose Canseco -1

+210

Not much to choose from between the two. Either is considerably better than single year.

[Brad K]

Originally Posted by Rain King

It is a pretty bold strategy to constantly insult the person who you are also asking to help you.

I didn't make the facts. I just reported them.

I don't consider that with a month of avoiding hef's questions there's much intention to help.