The Rule of Unintended Consequences

Gastric ReFlux · 03-27-2007, 09:54 AM

So far I'm enjoying the game a lot more than 2006. I've had more fun with it the last 5 days than I had with all of the play from 2006.

I do have an area of concern regarding the statistical output and AI player evaluation though, and worry that the rule of unintended consequences has reared its head.

An area of concern has been the AI's habit of sign and release, and relentless promotion and demotion. Back and forth, back and forth, over and over again. Things look much improved in this regard, but in my so far limited and unverified observation it has created an area of unrealism.

Too many 20 game losers.

I've been noticing this when at the end of the season I check the stats page of pitching leaders. I've even seen a fair number of times when the guy leading the league in losses had more than the guy leading the league in wins. I think I should definitely go in and take a count of things on games I've run so far when I get home tonight.

This is just an anecdotal unverified observation that I'm stating right now. Perhaps if there are some with time today they can take a look at their league histories and see if there are too many 20-game losers.

If there are, I suspect it's because the AI in being made more conservative about signing. releasing, promotion and demotion is being too conservative about trying out other pitchers.

This may be remedied I suppose by increasing the weight that the AI places on current season stats. Of course that could have its own run in with the rule of unintended consequences.

Curtis · 03-27-2007, 10:29 AM

That's a good catch. Are you playing fictional? Back in the deadball era twenty loss seasons were as common as dirt, and (of course) the seasons were shorter then.

statfreak · 03-27-2007, 10:32 AM

also, what are your player evaluations set at? I would assume if set to more heavily favor stats, the game would demote these starting pitchers earlier (unless of course the pitcher is 8-20 with a 3.69 ERA). Yeah, we'd need a ton more info to make an informed decision whether it is a problem or not...

Gastric ReFlux · 03-27-2007, 10:33 AM

Quote:

Originally Posted by Curtis

That's a good catch. Are you playing fictional? Back in the deadball era twenty loss seasons were as common as dirt, and (of course) the seasons were shorter then.

My obseravation has been largely with historicals, but I was noticing too many 20-game losers, I think, even well past 1950. Like I said prior, I may have the wrong impression and really need to go into the almanacs and take a count of how many there have been, then compare that to the number of 20-game losers the Major Leagues have produced over the years.

http://www.baseball-reference.com/fr...meLosers.shtml

The link above shows that most occurred prior to 1920.

But if anyone does have the time and inclination during today to take a peek at things, I'd appreciate it. It's certainly something that may need to be addressed in a patch if the game is producing well beyond reasonable amounts of 20-game losers.

Gastric ReFlux · 03-27-2007, 10:35 AM

Quote:

Originally Posted by statfreak

also, what are your player evaluations set at? I would assume if set to more heavily favor stats, the game would demote these starting pitchers earlier (unless of course the pitcher is 8-20 with a 3.69 ERA). Yeah, we'd need a ton more info to make an informed decision whether it is a problem or not...

That may be dependent upon what stats the game uses to evaluate its staff. If losses don't figure into the equation, changing the amount of influence stats have will not have any effect.

thbroman · 03-27-2007, 10:38 AM

Quote:

Originally Posted by Gastric ReFlux

This may be remedied I suppose by increasing the weight that the AI places on current season stats. Of course that could have its own run in with the rule of unintended consequences.

It's not clear to me that the weightings that are available to us for tweaking in the game setup affect more than the a.i.'s trading proclivities. Does anyone know whether they have wider effects? That would be useful to know.

Elendil · 03-27-2007, 10:52 AM

From what I understand, the solution to sign and release was essentially to force the AI to wait another month, then re-evaluate their roster. So since the AI can still change their rosters pretty frequently, I don't think that kind of fix would create this kind of problem, if a problem does exist.

Gastric ReFlux · 03-27-2007, 10:54 AM

Quote:

Originally Posted by Elendil

From what I understand, the solution to sign and release was essentially to force the AI to wait another month, then re-evaluate their roster. So since the AI can still change their rosters pretty frequently, I don't think that kind of fix would create this kind of problem, if a problem does exist.

Understood.

Mostly what I'm hoping for right now is to inspire a few people to go start taking a look at their almanacs. It may or may not be an actual issue, I'm just going off my impressions of what I've seen while trying things out in the game and doing general overview work, but if it is an issue, I think it's one that deserves some attention.

gordyhulten · 03-27-2007, 10:56 AM

Quote:

Originally Posted by Gastric ReFlux

Too many 20 game losers.

I've been noticing this when at the end of the season I check the stats page of pitching leaders. I've even seen a fair number of times when the guy leading the league in losses had more than the guy leading the league in wins. I think I should definitely go in and take a count of things on games I've run so far when I get home tonight.

This is just an anecdotal unverified observation that I'm stating right now. Perhaps if there are some with time today they can take a look at their league histories and see if there are too many 20-game losers.

If there are, I suspect it's because the AI in being made more conservative about signing. releasing, promotion and demotion is being too conservative about trying out other pitchers.

This may be remedied I suppose by increasing the weight that the AI places on current season stats. Of course that could have its own run in with the rule of unintended consequences.

I've noticed the same thing in some of my 50-year-test sims for my fictional universe. Over 50 years, I've probably had 50 20-game-losers. Taking a glance at them, however, a good proportion of them are decent pitchers on terrible teams, which I don't think is necessarily a problem. I don't want the AI to demote/release a pitcher who is nearing 20 wins if there are no better options available, and in many cases, for me, that seems to be the case.

(Aside: My player evaluation settings are set at a non-default 43/42/10/5, because that's what SkyDog had them set at in his template, and that was the baseline for my universe.)

RonCo · 03-27-2007, 11:00 AM

I'm guessing there are a few too many 20-game losers, but like Elendil said, I doubt that this is the result of the sign-and-release fix, but more related to either the GM AI or perhaps even the substitution/bullpen AI. I also would guess that the scope of the issue is not large.

RonCo · 03-27-2007, 11:02 AM

There are still stats gaps to fill, as in any simulation, but they are growing smaller.

Gastric ReFlux · 03-27-2007, 11:06 AM

Quote:

Originally Posted by RonCo

There are still stats gaps to fill, as in any simulation, but they are growing smaller.

I appreciate the work that has been done, but just had enough cases of where I kept thinking I was seeing too many 20-game losers on the stats leaders page.

It's a tricky issue to handle anyway. The AI may or may not use losses as a metric for staff evaluation, and at some level we the gamers may not want the AI to consider losses.

gordyhulten · 03-27-2007, 11:23 AM

Quote:

Originally Posted by Gastric ReFlux

It's a tricky issue to handle anyway. The AI may or may not use losses as a metric for staff evaluation, and at some level we the gamers may not want the AI to consider losses.

Not to start a statheadzzz vs. old-skool debate, but as you mention, I really don't want the AI to uses losses to evaluate the effectiveness of a pitcher. There are other, better stats available.

thbroman · 03-27-2007, 11:50 AM

Quote:

Originally Posted by gordyhulten

(Aside: My player evaluation settings are set at a non-default 43/42/10/5, because that's what SkyDog had them set at in his template, and that was the baseline for my universe.)

Thanks for the link to that thread. I hadn't noticed it when it first appeared and it's full of useful suggestions!

QuestGAV · 03-27-2007, 05:55 PM

I think it's pretty logical behavior- IRL the tendency would be to assume that the player's talent level had fallen below replacement level and that any sclub off the AAA roster/waiver wire would pitch more effectively. Since the AI can consider the player's actual talent level and realize there's no benefit to replacing him, they should be keeping the player around if no better options are available rather than paying that player's contract to release + potentially signing/promoting another player. I think the only real potential problem here is if the talent curve drops too quickly and globally the 4-5 starters don't perform as well as their real life counterparts.

03-27-2007, 09:54 AM	#1
Gastric ReFlux Hall Of Famer Join Date: Oct 2002 Location: Member #3409 Posts: 8,350	The Rule of Unintended Consequences So far I'm enjoying the game a lot more than 2006. I've had more fun with it the last 5 days than I had with all of the play from 2006. I do have an area of concern regarding the statistical output and AI player evaluation though, and worry that the rule of unintended consequences has reared its head. An area of concern has been the AI's habit of sign and release, and relentless promotion and demotion. Back and forth, back and forth, over and over again. Things look much improved in this regard, but in my so far limited and unverified observation it has created an area of unrealism. Too many 20 game losers. I've been noticing this when at the end of the season I check the stats page of pitching leaders. I've even seen a fair number of times when the guy leading the league in losses had more than the guy leading the league in wins. I think I should definitely go in and take a count of things on games I've run so far when I get home tonight. This is just an anecdotal unverified observation that I'm stating right now. Perhaps if there are some with time today they can take a look at their league histories and see if there are too many 20-game losers. If there are, I suspect it's because the AI in being made more conservative about signing. releasing, promotion and demotion is being too conservative about trying out other pitchers. This may be remedied I suppose by increasing the weight that the AI places on current season stats. Of course that could have its own run in with the rule of unintended consequences.

03-27-2007, 10:32 AM	#3
statfreak Hall Of Famer Join Date: Dec 2001 Location: Milwaukee, WI Posts: 2,434	also, what are your player evaluations set at? I would assume if set to more heavily favor stats, the game would demote these starting pitchers earlier (unless of course the pitcher is 8-20 with a 3.69 ERA). Yeah, we'd need a ton more info to make an informed decision whether it is a problem or not... __________________ Roll out the barrel!

03-27-2007, 10:52 AM	#7
Elendil Hall Of Famer Join Date: Dec 2003 Location: the dynasty forum Posts: 2,318	From what I understand, the solution to sign and release was essentially to force the AI to wait another month, then re-evaluate their roster. So since the AI can still change their rosters pretty frequently, I don't think that kind of fix would create this kind of problem, if a problem does exist. __________________ Heaven is kicking back with a double Talisker and a churchwarden stuffed with latakia.

03-27-2007, 10:29 AM	#2
Curtis Hall Of Famer Join Date: Jul 2006 Location: Watertown, New York Posts: 4,567	That's a good catch. Are you playing fictional? Back in the deadball era twenty loss seasons were as common as dirt, and (of course) the seasons were shorter then.

03-27-2007, 11:00 AM	#10
RonCo Hall Of Famer Join Date: Aug 2003 Posts: 10,386	I'm guessing there are a few too many 20-game losers, but like Elendil said, I doubt that this is the result of the sign-and-release fix, but more related to either the GM AI or perhaps even the substitution/bullpen AI. I also would guess that the scope of the issue is not large.

03-27-2007, 11:02 AM	#11
RonCo Hall Of Famer Join Date: Aug 2003 Posts: 10,386	There are still stats gaps to fill, as in any simulation, but they are growing smaller.

03-27-2007, 05:55 PM	#15
QuestGAV Hall Of Famer Join Date: Mar 2002 Location: Connecticut Posts: 2,008	I think it's pretty logical behavior- IRL the tendency would be to assume that the player's talent level had fallen below replacement level and that any sclub off the AAA roster/waiver wire would pitch more effectively. Since the AI can consider the player's actual talent level and realize there's no benefit to replacing him, they should be keeping the player around if no better options are available rather than paying that player's contract to release + potentially signing/promoting another player. I think the only real potential problem here is if the talent curve drops too quickly and globally the 4-5 starters don't perform as well as their real life counterparts.