Proposed Fictional League Dev/Aging Modifiers

[Markus Heinsohn]

Originally Posted by injury log

No one presented any data to support this change, and it should never have been made, according to my analysis of real life stats of players before and after serious injury. This needs to be fixed before development as a whole has any chance of being right.

I disagree here. The main reason why players who are not aging (read: 36+ years of age) see their production go down the drain are injuries in real life.

[Markus Heinsohn]

Again, I would like to see a Tango-like study with a fictional 16-team OOTP league with all minor league levels. Start the league, sim 100 years, and then analyze the stats (not ratings). Sveein did that during OOTP 9 beta, and the results looked very good IMO. Also, my tests results showed that development is in great shape as well. I focused on: League average stats, league average age, league average career length, league average ratings, individual career leaderboards and single season leaderboards. Everything was extremely stable and feeling realistic, no matter if in year 1, 30, 50 or 100.

[injury log]

Originally Posted by Markus Heinsohn

It does nothing in OOTP 9 to offset the missing injuries. In real life, if players would not bet injured, the talent of the league would be higher as well, so I did not see a reason to artificially change the development routines when injuries are not set to default or switched off.

A very good point - why would we expect development to match Tango with injuries off? In Tango's real life studies, injuries are on; how can we even know what a benchmark should be for injury-free development?

Originally Posted by Markus Heinsohn

I disagree here. The main reason why players who are not aging (read: 36+ years of age) see their production go down the drain are injuries in real life.

I'll show you the data I've collected from real life, and hopefully this will convince you that OOTP goes much too far here.

[Markus Heinsohn]

Originally Posted by injury log

I'll show you the data I've collected from real life, and hopefully this will convince you that OOTP goes much too far here.

Have you got minor league data as well? Its especially young pitchers that suffer dropdowns after injuries, and OOTP 9 tries to mirror this...

[RonCo]

Originally Posted by Markus Heinsohn

Again, I would like to see a Tango-like study with a fictional 16-team OOTP league with all minor league levels. Start the league, sim 100 years, and then analyze the stats (not ratings). Sveein did that during OOTP 9 beta, and the results looked very good IMO. Also, my tests results showed that development is in great shape as well. I focused on: League average stats, league average age, league average career length, league average ratings, individual career leaderboards and single season leaderboards. Everything was extremely stable and feeling realistic, no matter if in year 1, 30, 50 or 100.

I understand the attraction of trying to use stats from OOTP to compare to stats that Tango used in creating his aging chains. Luckily, though, it's really not necessary since you have ratings that match the context of the stats you're trying to use. Why "luckily?" Well, because in order to get stats to work properly out of OOTP you need to adjust for ballpark--and you will need to adjust for ballpark in the same way as Tango did--not just apply OOTP park values. This task is very, very difficult.

In fact, if you use stats without adjusting them for ballpark (and possibly league totals .. though I need to think about that one a bit), and do the same study as Tango, your data will be wrong.

So, luckily, you don't need to do that because all that work Tango is doing to distill the data down is done _specifically_ to model what OOTP call "ratings." The one error in using OOTP ratings directly is that OOTP ratings are not quite used linearly in their mapping to performance rates. The error induced here is quite small, though, and if you get that close no one will really complain much. I have heard one valid complaint to the way I do the study, and that's that a pitcher's K-rate is not 100% equal to their Stuff rating. It is instead something like 85% stuff, 10% movement, 5% control (or whatever). Therefore, to be really perfect in my study I should add that weighting.

As far as taking injuries out of the study, I can see that having some issue, but since injuries are most likely fairly evenly spread out in the early years, the growth stage of the curve should be fine (since it's all relative--all early ages should get hurt at about the same rate). If there's an area of the study I would be concerned about regarding injury it would be the ages past 32 or so where I suspect injuries begin to happen more often--so I would expect players to fall off more rapidly after age 32-33 than my studies show.

[RonCo]

Originally Posted by injury log

A very good point - why would we expect development to match Tango with injuries off? In Tango's real life studies, injuries are on; how can we even know what a benchmark should be for injury-free development?

Given Tango's selection criteria (large numbers of plate appearances, for example), I suggest that the impact of injuries beyond day-to-day type bumps and bruises on the Tango numbers is minimal because injured players would not appear in the study very often. Just guessing, though.

[mikev]

RonCo, it's good to see you back.

[andymac]

Originally Posted by RonCo

Given Tango's selection criteria (large numbers of plate appearances, for example), I suggest that the impact of injuries beyond day-to-day type bumps and bruises on the Tango numbers is minimal because injured players would not appear in the study very often. Just guessing, though.

Wouldn't it show up in the absence of those players though? If Albert Pujols gets hurt and doesn't have the at-bats to be in the study, that means a less talented player is getting a lot more at-bats and therefore appears in the study when they wouldn't have had Pujols been healthy.

[RonCo]

Originally Posted by andymac

Wouldn't it show up in the absence of those players though? If Albert Pujols gets hurt and doesn't have the at-bats to be in the study, that means a less talented player is getting a lot more at-bats and therefore appears in the study when they wouldn't have had Pujols been healthy.

No.

The aging chains are player-specific. The process assumes the mythical "average player" ages the same as any other player. They may peak at different values (Pujol's HR-rate will peak at 7 gabillion times that of Juan Pierre), but on average they will both peak at 27 and they will both grow to that peak at the same _rate_. That's the beauty of the chain process. It's attempting to model how the average human being ages, so data from a group can be successfully used to predict how any individual will progress.

So, when you take Albert pujols out, you're still using a big collection of players that consist of Juan Pierre, Barry Bonds, WIllie Mayes, Roger Metzgar, Harmon Killabrew, Lonnie Smith, Dave Parker ... blah, blah, blah, ... to guess how any othe human will grow (given their unique starting point).

[RonCo]

Or, maybe it's better to say yes, but when you take into account the number of players in the Tango study, and realize you're missing data from injured Pujols's and missing Metzgar's at like similar rates, your overall error rate is probably not very high due to this factor. But all this data is subject to question...it is, though, the best data that I know of that is publicly available.

[andymac]

Yea, my post was a bit of a brain fart.

[injury log]

Originally Posted by RonCo

Given Tango's selection criteria (large numbers of plate appearances, for example), I suggest that the impact of injuries beyond day-to-day type bumps and bruises on the Tango numbers is minimal because injured players would not appear in the study very often. Just guessing, though.

I don't know - think of the player with chronic injury, but who plays through it. He'll show up in Tango. I'm thinking of someone like Eric Chavez, who at a glance probably fits Tango well (peak at 27, then decline). In its current version, OOTP would probably model Chavez by knocking down his ratings a bit each time he suffers a minor injury. Take those injuries away from OOTP, and Chavez probably maintains his peak longer.

Of course, there are lots of other potential effects that might disappear when injuries are turned off, and maybe it all irons out in the wash, but to test OOTP for a Tango match, I'd think you'd want to try to simulate real life as closely as possible, and part of that is keeping injuries on.

Anyway, if we can agree on a methodology, it'll be easier to test all of this in beta this year and get a good result for the next game.

[injury log]

Originally Posted by Markus Heinsohn

Have you got minor league data as well? Its especially young pitchers that suffer dropdowns after injuries, and OOTP 9 tries to mirror this...

Not much, as it's time consuming even to collect MLB data. I also suspect there's a feedback type effect in real life, especially for pitchers- from one bad arm injury, pitchers seem to have somewhere around a 90% chance of recovery, but I suspect that a second bad arm injury has a much lower recovery rate.

(by 'recovery', I mean 'coming back at a level reasonably close to, but not equal to, their previous level').

[Markus Heinsohn]

Originally Posted by RonCo

Luckily, though, it's really not necessary since you have ratings that match the context of the stats you're trying to use.

That is true for hitters, but not pitchers. For example, for calculating if there is a strikeout, the game uses 95% stuff, and 2.5% movement+control each. To see if it's a HR, it uses 90% movement, and 10% control. For a BB it's 100% control.

[RonCo]

Originally Posted by Markus Heinsohn

That is true for hitters, but not pitchers. For example, for calculating if there is a strikeout, the game uses 95% stuff, and 2.5% movement+control each. To see if it's a HR, it uses 90% movement, and 10% control. For a BB it's 100% control.

I just ran the data with these figures. It nudges the development lines a little, but not enough to appreciably change anything.

It would be more appropriate, IMO, to make:

Stuff = K-rate = .6 * velocity + .2 * movement + .1 * control + .1 * intelligence

or something like that. So, make stuff a composite rating like contact is today.

[RonCo]

Originally Posted by injury log

I don't know - think of the player with chronic injury, but who plays through it. He'll show up in Tango. I'm thinking of someone like Eric Chavez, who at a glance probably fits Tango well (peak at 27, then decline). In its current version, OOTP would probably model Chavez by knocking down his ratings a bit each time he suffers a minor injury. Take those injuries away from OOTP, and Chavez probably maintains his peak longer.

Of course, there are lots of other potential effects that might disappear when injuries are turned off, and maybe it all irons out in the wash, but to test OOTP for a Tango match, I'd think you'd want to try to simulate real life as closely as possible, and part of that is keeping injuries on.

Anyway, if we can agree on a methodology, it'll be easier to test all of this in beta this year and get a good result for the next game.

As I said, injuries could shift it a little, but my _guess_ is that since injuries happen fairly randomly at young ages, the effect is equal (therefore you can safely ignore it) at the young/growth curve. If injuries negatively influence the study it would be at the older ages, where I assume OOTP injuries are more prevalent (hence not random). Still, my guess overall is that OOTP injuries wouldn't change much, but I've been wrong before.