Question

[David Watts]

I have a question for those that actually pay attention to ratings and such. Would love to read your thoughts on this matter.

I've been trying to get back into fictional lately and I will be finishing the 2nd season of my league this evening. One thing that bugged me for most of the 1st season was lineup construction (I don't manage or gm...I just watch). I was seeing so many teams with under performing players holding key spots in the batting order. I'm talking guys hitting .189 2 home runs and 12 RBI hitting cleanup halfway through a season. Meanwhile the team would have a guy performing much better buried in the 7 or 8 hole. I'd say around August or so of the 1st season, I made a change going with 10 ratings, 65 current season, 25 previous season and 0 for 2 years prior. I have to say, I'm loving the effect this has had on lineups. No longer do I see guys in a year long slump holding key batting order spots.

My question is, will going away from ratings lead to the AI making goofball roster/draft/trade decisions or will 10 percent ratings be enough to keep the AI from making poor decisions. I don't use scouting at all.

[jberg57]

I think you have a nice balance. That is the way a lot of people play...some play without ratings and totally turn them off.

After all if Player A was batting .355 and Player B was batting .212 theoretically Team A would not trade for Player B statistically, but Player B might have a much higher rating and the trade might go through on a ratings based model.

[NoOne]

by adding to the weight of stats, it will affect the overal more by the stats...

if the overall improves, they are more likely to be used / desired than previously to the AI. there is a limit to its effect, though. a 20-rated guy hitting .300-.400 for a month or 2 isn't going to supplant anyone rated highly even after his overall is adjusted and the other player is hitting .200 during that same time period.

one thng to note, when the current year is fresh, that percentage for current year is applied to the other 2 years (or possibly uses the 3 years ago while the current year is not viable - i believe the words were that they re-alot it though, which i'd assume just bumps it into the 1st, 2nd year and (mostly - may slowly increase to intended weight, too) ignores current year while it's not viable data).

so, bumping up current year may not have the intended effect that you want the first couple? months of the year. (they did not intimate when the current year's sample was largen enough to use as it normally would. i'd like to point out that any 1 year is rarely a large enough sample to be certain about anything. so, this is just based on potentially bad data. but, i'd probably add to "last year" stat weight to effect lineups at the start of the year.

i'd do some testing with the lineups that you see and do not like.... e.g. change the ratings, force immediate re-scout, then go to that team and "ask manager to set up all depth charts and lineups" ... you will immediately see the effects of your change, if any to the lineup.... obviously you'll have to make sure it doesn't have adverse effects on other stuff, if you find something that works well for you.. but before that, you need to find somethign that makes lineups that you prefer.

may want to look at sabermetric vs standard lineups - even if you don't like the syntax... labels don't matter, what results does matter.

while the opposite can happen, this is likely just a temporary disparity that you see... randomness making an inferior player look better than a more highly rated one.

ignoring other factors to make this simpler: if the guy's contact is 100/200, he's never going to be a consistently high average player no matter what his overall adjusts to due to stats weighting. they will always regress to the mean if given enough opportunities to play. always. since you can't predict when they'll play well for an extended amount of time, it's total luck when yo re-arrange a lineup due to recent results.

try turning off ratigns completely if you want knee jerk reactions in that league. it sounds a bit neagtive the way i say it.. but it's just an arbitrary preference and possibly a bit more realistic for the dumber gm's in RL (80-90% of them). so much of what is "normal procedure" is really just dogmatic practice.

anyway don't restrict yourself by wht you think it should be ... just focus on the results matching your goal for lineups - with a little give. don't expect all of them to make lineups you approve of.

[sprague]

Hey David.
I think it is a balance thing. Too much one way or another is a challenge.
Going with 10 is almost stats only. Good for lineups but ya ai can do odd things.
70 ratings you see the lineups.

Have you thought of maybe 35 40 25 0
It might be a good balance

[David Watts]

Originally Posted by sprague

Hey David.
I think it is a balance thing. Too much one way or another is a challenge.
Going with 10 is almost stats only. Good for lineups but ya ai can do odd things.
70 ratings you see the lineups.

Have you thought of maybe 35 40 25 0
It might be a good balance

Thanks, that gives me a direction to go if this off season proves freaky.

[RchW]

It has to be a bit of both. Too much ratings and you get what DW saw, however too much on stats and you can get a hot player considered equivalent to a better player and some AI trades show this.

Two questions;

Exactly what stats does the AI consider? It better not be counting stats.

Does the trade AI have a built in "stop" to trades that are driven by "this year stats? What I mean is there should be slightly different weighting given to trade eval vs lineup and play time eval.

Maybe there should be two sets.

[NoOne]

based on another thread, without being specific, the stats essentially influence how overall is added up.

e.g. contact * modifier based on statistical results (babip, ba, whatever)... each portion of the equation is modified by resutls and adds up as it normally does. it wouldn't even have to be a separate piece of code if they didn't want it to be.

other thread gave info:
defensive metrics affect overall - conclusion: so basically anything that makes sense to tie to each skill rating that's used in overall is possible and likely. there's a finite amount of info given, and it's only there becuae it is used somewhere (ui / feedback only type stuff excluded)... few things are for our eyes only.

so whatever the normal equation is for overall, it simply is applying a factor based upon statistical results - or roughly that, if not the exact math it's what is happening in essence.

i think it's a bit simple to think adding stats makes decisions better.. .they are small sample sets of data being used on purpose... randomness employed on purpose. they are less predictive than accurate ratings... that's why you want it in there... when a player is slightly underrated by a scout but his stats clearly show he's possibly better. it's a double-edged sword, though. it's likely going to work the opposite just as often. some udnerrated player will play over a better player just as often. humans make mistakes due to bad info, this is basically modelling that.