Relative Ratings Mode Confirmed as Default in OOTP 19

hjrrockies · 01-29-2018, 05:54 PM

See this interview transcript with Markus and Rich: http://www.gameinformer.com/b/featur...seball-19.aspx

The relevant quote:

Quote:

MH: "The rating system is now by default one that calculates the player ratings relative to league average. In the past the ratings were absolute values, so if you played a 1910 league in the dead-ball era, no player had a high power rating and most looked alike. Now, you can see who the best in the league are because the individual HR ratings are based on the league average HR rating, and a 80/80 is three standard deviations better than average, if I recall correctly."

Very excited to see this fleshed out!

Matt Arnold · 01-29-2018, 06:25 PM

Yeah, we spent a lot of time to clean up a lot of the ratings. Hopefully everything will look much more consistent and clean

actionjackson · 01-29-2018, 06:52 PM

I hope this helps clean up the issues raised in this thread, which are not limited to seasons where you tie the stats to one particular historical season, but persist as well in straight historical replay, whether it's a regular historical game or a random debut historical game, as evidenced in this post, and this post from this thread.

Garlon · 01-31-2018, 11:14 AM

This has the potential to cause simulation problems in historicals. This is really not how the system was designed to work in conjunction with the league totals modifiers...I provided the design for the LTMs many years ago. They work because the absolute ratings are derived from the raw stats compiled from the league environment they were produced in by that player. This is then compared with the engine totals to produce results and then modified with some algebra to tune the league to perform very accurately. In addition this will also make things more confusing in the dead all era. HRs are not a significant contributor to offense in that era, so the rating will not be informative. I hope there is still a way to use the old rating system. Players like Cobb and Ruth are performing at a level so much above the league average that they may not perform properly with a ratings system like this.

JaBurns · 01-31-2018, 04:31 PM

This problem seems to have reared it ugly head already in 18 as is been discussed in one thread. See bugs report forum.

actionjackson · 01-31-2018, 05:06 PM

Quote:

Originally Posted by JaBurns

This problem seems to have reared it ugly head already in 18 as is been discussed in one thread. See bugs report forum.

I would guess that problem is something else, as this indicates that ratings were done the same way in OOTP18 as they were done in OOTP16, but I find the OOTP16 outcomes to be more realistic. OOTP18 seems a wee bit hot on an individual basis, not a league-wide basis.

JaBurns · 01-31-2018, 06:25 PM

I stand corrected, the “running hot thing” is bothering me in 18 because I try to introduce random results by moving the TCR to 140. This allows me to think that players will not precisely follow there historical path and allows for some deviation. I have just started a new random debut league. Now for example if Tony Gwynn hits .400 I will question whether it is my TCR setting or the league running hot. I really enjoy historical random debut and was trying to find my niche with this setup inside this amazing game. But now I have questions and doubts; hopefully I am over reacting and the design team will find and fix the problems.

r0nster · 01-31-2018, 06:34 PM

definitely interested in the scouting features and many others as well

Cobby · 01-31-2018, 07:13 PM

Quote:

The rating system is now by default one that calculates the player ratings relative to league average. In the past the ratings were absolute values, so if you played a 1910 league in the dead-ball era, no player had a high power rating and most looked alike. Now, you can see who the best in the league are because the individual HR ratings are based on the league average HR rating, and a 80/80 is three standard deviations better than average, if I recall correctly.

This quote bothers me a little bit - or at least needs some clarification I think. It seems to me that the talent of players in the league ought not to be distributed as a normal distribution. I think the normal distribution comes in when considering the entire talent pool of the country (or even the world). At top end you have Babe Ruth; at the bottom end you have a 96 year old man on crutches. In the middle - who knows? Possibly a 35 year old office worker somewhere. The people that become professional athletes are the elite of the population. Those that make it to MLB are the elite of the elite. The tiny sliver of the normal curve way at the right end.

So, I would expect the distribution of MLB players to be highly skewed, with a lot more players toward the left end (wherever the "replacement level" cutoff is) and just a few on the right end 80/80 level.

So I don't know what 80 = 3 standard deviations above average means in the context of a non-normal distribution. Some clarification perhaps?

hjrrockies · 01-31-2018, 08:45 PM

Standard deviation (the square root of the variance) is still well-defined for distributions other than normal.

Cobby · 02-01-2018, 02:20 AM

Quote:

Originally Posted by hjrrockies

Standard deviation (the square root of the variance) is still well-defined for distributions other than normal.

Defined, but not all that useful unless we know more about the distribution. For example, if the distribution is highly skewed to the left and I would assume the mean is assigned to a rating of 50, then that would mean that the 30 point rating difference between a 20-rated player and 50-rated player wouldn't mean too much in absolute terms, but the 30 point rating difference between a 50-rated player and an 80-rated player would mean a great deal.

And if the ratings were distributed somewhat normally, I think that would a problem too. They really should be skewed.

Plus, in the example given - you might find yourself getting all excited about that player in the deadball era with an 80 power rating. Then you would find out that what that really means is that he hits 5 home runs a season instead of 1 like the 20-rated guy. So now in your mind, if you're in the deadball era, you have try to mentally disregard a player's power rating, which is hard to do with that long lovely blue line and 80 rating staring at you. I'd much rather those players have a 22 and 20 power ratings instead as they probably would have in absolute mode.

I don't mean to be a killjoy or anything - I'm amazed at the work the development team does to bring us such an amazing game.

But in this case - without further info about how this works and what the distributions are like and more of an explanation - the first thing I'll do is turn off the relative ratings mode and go back to absolute.

Huckleberry · 02-01-2018, 01:50 PM

Quote:

Originally Posted by Cobby

Defined, but not all that useful unless we know more about the distribution. For example, if the distribution is highly skewed to the left and I would assume the mean is assigned to a rating of 50, then that would mean that the 30 point rating difference between a 20-rated player and 50-rated player wouldn't mean too much in absolute terms, but the 30 point rating difference between a 50-rated player and an 80-rated player would mean a great deal.

I would think they need to explain it better, certainly. Because the net effect if you apply a normal distribution based rating system on this type of population is that the lowest end of the ratings set will have no members.

Quote:

Originally Posted by Cobby

And if the ratings were distributed somewhat normally, I think that would a problem too. They really should be skewed.

Is that really the case, though? Major league players in whatever OOTP universe you're playing have been selected from that skewed end of the population, and the minors or reserve rosters or free agent pool are filled with the guys on the lower end. However, because a player's full value is based on dozens of components, you will have some players in the majors who have very low ratings in certain skills. Here are a couple of charts from an OOTP major league. The first is the number of players in each range in a certain skill.

It's not a normal distribution but it's also not the tail-end that you're envisioning. The problem, though, comes when you convert that to a 20-80 scale using the 3 standard deviations idea (every 10 rating points is based on a single standard deviation from the mean. That results in this ratings set:

There are no players out of a 400+ set who have 20 or 25 ratings in this skill. Obviously a study would need to look at more than just one skill, but it's an interesting starting point.

Garlon · 02-01-2018, 09:24 PM

As for random debut leagues, the game was never designed to handle that. You'd really need to translate the player ratings to the new league environment to do that. This standard deviation stuff will not work for that either. Baseball statistics simply do not work that way. The Davenport Translations will allow for play with players from different eras but we do not have the Davenport Translations in the game database. Playing with Tony Gwynn in any other era than his actual career is meaningless in terms of results. He compiled his stats under a given set of playing conditions, so if you put him in the deadball era there is no real value to that. Batting .393 in 1994 is completely different than batting .393 in 1905.

If the way that the ratings are actually made from the stats is not different than in previous OOTP games then historicals are not going to work because the ratings will not be consistent with how the outcomes are generated by the game engine. The absolute values are the ratings the players need because that sets them relative to each other properly. The standard deviation system or setting these rating to 50 as league average or something will cause issues. If you look at major league ERA's there are many more pitcher below the median than above the median because pitchers who do not perform well will get sent down to the minors or will not accumulate as many innings pitched...it is not a normal distribution. Another issues with using some sort of standard deviation is that you need to consider the fact the pitchers have different amounts of Batters Faced and the positon players have different amounts of PA.

In 1921 the average major leaguer hit 6.05 HR per 550 AB. So are we now going to see this as an "average" Power rating of say 50 on the 250 scale? What is Ruth going to have as a rating as he hit 59 HR in 540 AB, nearly 10 times greater than the league average. The scale only goes to 250. So if there is not the same separation between Ruth and the average player in the league when using the absolute ratings then Ruth and other star players will always perform poorly. With absolute ratings this is not an issue because the average player in the league will have a power rating worth about 6 HR per 550 AB and Ruth will be 10 times higher and be much more likely to hit HRs. Whats the standard deviation for HR per 550 AB going to be in 1921, 3 HR (unlikely that the SD ifs going to be more than half the average probably)? Then Ruth, who averaged 60 HR per 550 AB is about18 standard deviations above the mean...how will the rating system be able to handle that? It can't, and therefore the great players will no be so great when you play. Everyone will get pulled toward the mean. The League Totals Modifiers will keep the league average rates correct butt getting realistic player outcomes requires that the player ratings be relative to each other properly and that is done with the absolute values. This pretty much compromises historical play.

Matt Arnold · 02-01-2018, 09:43 PM

Just a quick note - under the covers the ratings will work as they always have, It's more of a display issue that changed. And we'll make sure that things look good overall- nothing says that we can't tweak things to make sure they look accurate

Garlon · 02-01-2018, 10:21 PM

Ok, that's good. I hope we have the option to switch between these new ratings and the traditional OOTP absolute ratings when playing.

JaBurns · 02-02-2018, 04:11 PM

Quote:

Originally Posted by Garlon

As for random debut leagues, the game was never designed to handle that. You'd really need to translate the player ratings to the new league environment to do that. This standard deviation stuff will not work for that either. Baseball statistics simply do not work that way. The Davenport Translations will allow for play with players from different eras but we do not have the Davenport Translations in the game database. Playing with Tony Gwynn in any other era than his actual career is meaningless in terms of results. He compiled his stats under a given set of playing conditions, so if you put him in the deadball era there is no real value to that. Batting .393 in 1994 is completely different than batting .393 in 1905.

If the way that the ratings are actually made from the stats is not different than in previous OOTP games then historicals are not going to work because the ratings will not be consistent with how the outcomes are generated by the game engine. The absolute values are the ratings the players need because that sets them relative to each other properly. The standard deviation system or setting these rating to 50 as league average or something will cause issues. If you look at major league ERA's there are many more pitcher below the median than above the median because pitchers who do not perform well will get sent down to the minors or will not accumulate as many innings pitched...it is not a normal distribution. Another issues with using some sort of standard deviation is that you need to consider the fact the pitchers have different amounts of Batters Faced and the positon players have different amounts of PA.

In 1921 the average major leaguer hit 6.05 HR per 550 AB. So are we now going to see this as an "average" Power rating of say 50 on the 250 scale? What is Ruth going to have as a rating as he hit 59 HR in 540 AB, nearly 10 times greater than the league average. The scale only goes to 250. So if there is not the same separation between Ruth and the average player in the league when using the absolute ratings then Ruth and other star players will always perform poorly. With absolute ratings this is not an issue because the average player in the league will have a power rating worth about 6 HR per 550 AB and Ruth will be 10 times higher and be much more likely to hit HRs. Whats the standard deviation for HR per 550 AB going to be in 1921, 3 HR (unlikely that the SD ifs going to be more than half the average probably)? Then Ruth, who averaged 60 HR per 550 AB is about18 standard deviations above the mean...how will the rating system be able to handle that? It can't, and therefore the great players will no be so great when you play. Everyone will get pulled toward the mean. The League Totals Modifiers will keep the league average rates correct butt getting realistic player outcomes requires that the player ratings be relative to each other properly and that is done with the absolute values. This pretty much compromises historical play.

This question may sound dumb because I do not have near the experience of you others with historical random debut. But this my favourite part of the game, I am trying to learn and tailor a setup to my liking using this mode of play.
That said here is the question: Isn’t neutralize stats supposed to take care of the difference in eras?

Reed · 02-02-2018, 06:20 PM

It’s my understandiding that neutralized stats is NOT to make players stats neutral across season or eras. It’s to take account of ballpark factors so you can make a comparison between a player that played in Coors field and a player that played that year in Dodger stadium.

JaBurns · 02-02-2018, 06:41 PM

I did not know that now I have to rethink my whole league, it is not the first time nor probably the last that I have to blow up a league and head back to the proverbial drawing board.

Spritze · 02-02-2018, 08:05 PM

Quote:

Originally Posted by Reed

It’s my understandiding that neutralized stats is NOT to make players stats neutral across season or eras. It’s to take account of ballpark factors so you can make a comparison between a player that played in Coors field and a player that played that year in Dodger stadium.

Well it removes ballpark factors AND makes players stats neutral across all seasons in any era. 750 runs per team per year. That is why it is called Neutralized Stats.

Garlon · 02-02-2018, 11:23 PM

It does not neutralize stats across eras. Even though th DB is based on a 750-run environment, the way those 750 runs are accumulated are vastly different throughout history. The neutralized stats really adjust for ballpark factors. If you want realistic play for players from different eras then you need translated statistics and even then you have to decide what era you want to translate to. The Davenport Translations translate player stats from every year to a standard environment. This is done using EQA. I have tried to find the formulas used to do this but I don't see these published anywhere.

I was planning to do a couple roster projects using these as the basis. One project with the 25 best players from every franchise, and another project with the single best team for every franchise. This of course will take quite a while to put together and can only be used for a single season.

01-31-2018, 04:31 PM	#5
JaBurns All Star Starter Join Date: Sep 2015 Location: Kelowna, British Columbia Posts: 1,295	This problem seems to have reared it ugly head already in 18 as is been discussed in one thread. See bugs report forum. Last edited by JaBurns; 01-31-2018 at 04:32 PM.

02-02-2018, 06:41 PM	#18
JaBurns All Star Starter Join Date: Sep 2015 Location: Kelowna, British Columbia Posts: 1,295	I did not know that now I have to rethink my whole league, it is not the first time nor probably the last that I have to blow up a league and head back to the proverbial drawing board. Last edited by JaBurns; 02-03-2018 at 01:13 PM.

01-29-2018, 06:25 PM	#2
Matt Arnold OOTP Developer Join Date: Jun 2009 Location: Here and there Posts: 15,902	Yeah, we spent a lot of time to clean up a lot of the ratings. Hopefully everything will look much more consistent and clean

01-29-2018, 06:52 PM	#3
actionjackson Hall Of Famer Join Date: Apr 2009 Location: Toronto, ON Posts: 6,181	I hope this helps clean up the issues raised in this thread, which are not limited to seasons where you tie the stats to one particular historical season, but persist as well in straight historical replay, whether it's a regular historical game or a random debut historical game, as evidenced in this post, and this post from this thread.

01-31-2018, 11:14 AM	#4
Garlon Hall Of Famer Join Date: Jun 2004 Posts: 4,273	This has the potential to cause simulation problems in historicals. This is really not how the system was designed to work in conjunction with the league totals modifiers...I provided the design for the LTMs many years ago. They work because the absolute ratings are derived from the raw stats compiled from the league environment they were produced in by that player. This is then compared with the engine totals to produce results and then modified with some algebra to tune the league to perform very accurately. In addition this will also make things more confusing in the dead all era. HRs are not a significant contributor to offense in that era, so the rating will not be informative. I hope there is still a way to use the old rating system. Players like Cobb and Ruth are performing at a level so much above the league average that they may not perform properly with a ratings system like this.

01-31-2018, 06:25 PM	#7
JaBurns All Star Starter Join Date: Sep 2015 Location: Kelowna, British Columbia Posts: 1,295	I stand corrected, the “running hot thing” is bothering me in 18 because I try to introduce random results by moving the TCR to 140. This allows me to think that players will not precisely follow there historical path and allows for some deviation. I have just started a new random debut league. Now for example if Tony Gwynn hits .400 I will question whether it is my TCR setting or the league running hot. I really enjoy historical random debut and was trying to find my niche with this setup inside this amazing game. But now I have questions and doubts; hopefully I am over reacting and the design team will find and fix the problems.

01-31-2018, 06:34 PM	#8
r0nster Hall Of Famer Join Date: May 2006 Location: Michigan Posts: 3,053	definitely interested in the scouting features and many others as well

01-31-2018, 08:45 PM	#10
hjrrockies Minors (Single A) Join Date: Mar 2016 Posts: 78	Standard deviation (the square root of the variance) is still well-defined for distributions other than normal.

02-01-2018, 09:24 PM	#13
Garlon Hall Of Famer Join Date: Jun 2004 Posts: 4,273	As for random debut leagues, the game was never designed to handle that. You'd really need to translate the player ratings to the new league environment to do that. This standard deviation stuff will not work for that either. Baseball statistics simply do not work that way. The Davenport Translations will allow for play with players from different eras but we do not have the Davenport Translations in the game database. Playing with Tony Gwynn in any other era than his actual career is meaningless in terms of results. He compiled his stats under a given set of playing conditions, so if you put him in the deadball era there is no real value to that. Batting .393 in 1994 is completely different than batting .393 in 1905. If the way that the ratings are actually made from the stats is not different than in previous OOTP games then historicals are not going to work because the ratings will not be consistent with how the outcomes are generated by the game engine. The absolute values are the ratings the players need because that sets them relative to each other properly. The standard deviation system or setting these rating to 50 as league average or something will cause issues. If you look at major league ERA's there are many more pitcher below the median than above the median because pitchers who do not perform well will get sent down to the minors or will not accumulate as many innings pitched...it is not a normal distribution. Another issues with using some sort of standard deviation is that you need to consider the fact the pitchers have different amounts of Batters Faced and the positon players have different amounts of PA. In 1921 the average major leaguer hit 6.05 HR per 550 AB. So are we now going to see this as an "average" Power rating of say 50 on the 250 scale? What is Ruth going to have as a rating as he hit 59 HR in 540 AB, nearly 10 times greater than the league average. The scale only goes to 250. So if there is not the same separation between Ruth and the average player in the league when using the absolute ratings then Ruth and other star players will always perform poorly. With absolute ratings this is not an issue because the average player in the league will have a power rating worth about 6 HR per 550 AB and Ruth will be 10 times higher and be much more likely to hit HRs. Whats the standard deviation for HR per 550 AB going to be in 1921, 3 HR (unlikely that the SD ifs going to be more than half the average probably)? Then Ruth, who averaged 60 HR per 550 AB is about18 standard deviations above the mean...how will the rating system be able to handle that? It can't, and therefore the great players will no be so great when you play. Everyone will get pulled toward the mean. The League Totals Modifiers will keep the league average rates correct butt getting realistic player outcomes requires that the player ratings be relative to each other properly and that is done with the absolute values. This pretty much compromises historical play.

02-01-2018, 09:43 PM	#14
Matt Arnold OOTP Developer Join Date: Jun 2009 Location: Here and there Posts: 15,902	Just a quick note - under the covers the ratings will work as they always have, It's more of a display issue that changed. And we'll make sure that things look good overall- nothing says that we can't tweak things to make sure they look accurate

02-01-2018, 10:21 PM	#15
Garlon Hall Of Famer Join Date: Jun 2004 Posts: 4,273	Ok, that's good. I hope we have the option to switch between these new ratings and the traditional OOTP absolute ratings when playing.

02-02-2018, 06:20 PM	#17
Reed Hall Of Famer Join Date: Oct 2013 Posts: 2,353	It’s my understandiding that neutralized stats is NOT to make players stats neutral across season or eras. It’s to take account of ballpark factors so you can make a comparison between a player that played in Coors field and a player that played that year in Dodger stadium.

02-02-2018, 11:23 PM	#20
Garlon Hall Of Famer Join Date: Jun 2004 Posts: 4,273	It does not neutralize stats across eras. Even though th DB is based on a 750-run environment, the way those 750 runs are accumulated are vastly different throughout history. The neutralized stats really adjust for ballpark factors. If you want realistic play for players from different eras then you need translated statistics and even then you have to decide what era you want to translate to. The Davenport Translations translate player stats from every year to a standard environment. This is done using EQA. I have tried to find the formulas used to do this but I don't see these published anywhere. I was planning to do a couple roster projects using these as the basis. One project with the 25 best players from every franchise, and another project with the single best team for every franchise. This of course will take quite a while to put together and can only be used for a single season.