How are Overall and Potential Ratings (Stars) calculated?

[justafan]

There doesn't seem to be a field in the CSV files for this value, so I guess there needs to be a formula to calculate this value. Anyone know?

[NoOne]

each ratings is likely used with some weight applied to it....

it's easier to explain contact, because it is similar and simpler.

The contact ratings you see is actually caclulated from 3 ratings.

power+babip+avoid k's.... each is weighted by a factors, each of these factors add up to 1. so, you could easily deduce the exact equation they use for contact, because it doesn't involve many things.

overall is just like that but it likely uses ALL of the ratings a player has.. .they aren't there for window dressing. (some things are, but common sense should make that easy to delineate)

So, overall can be misleading in certain situations... this is why most people in the forums suggest ignoring overall for any major decision... i use it for filtering and that's about it.. i will never pick one plyer over another due to their overall rating (potential is different animal, of course).

if you cannot understand why some palyer with the exact same contact ratings have drastically differen results over the course of their careers, this is one major reason why. (things like personality, coaches and other factors have a smaller influence than the skill ratings).. it's not a 'pure rating.' it is a conglomerate of 3 other ratings. so, different combinations of avoid k, power and babip that result in the same Contact will have different probabilities of results, therefore different results that are statistically significant.

it's possible they have it perfectly tuned and the differences are merely random... if that were the case why have 3 ratings combinded into 1, if those 3 ratings are completely unused on their own. you could simply just have a contact ratings and save all the headaches...

so it's likely to be relevant what each of those 3 are even if they add up to the same Conctact value. same with pitch selections adding up to overall stuff. a 90 stuff is not always the same as another player's 90stuff.

contact, overall, and some other ratings are merely to sum multiple related factors. therefore the resulting values of a conglomerate rating like contact likely does not have a 1:1 relationship with results (poor choice of words...) - or at least nowhere near as strong of a correlation as "avoid k's" rating and resulting k's.

from my experience this holds true... i can make multiple ptichers with the same stuff and other ratings and only change the pithc selection and get drastically different results with the same ratings... Contact is similar, but a bit simpler. overall doesn't tell you anything definitively, but the "better" players are still way more likely to have a decent overall rating. some combinations add up to lower overall but end up with better resutls... it's not to rank or compare players in a specific way, but rather it's to sum them up on an individual basis.

[justafan]

Thanks for your in-depth analysis. I guess without a formula, I will not be able to duplicate the ratings from available CSV values.

There are a few other fields missing from CSV's. I'm disapointed that the exports don't include everything.

[NoOne]

overall would be more difficult, but you could do it, if you wanted to.

i wouldn't suggest it because of ROI isn't worth it. (return on invested effort and time)

i'm sure a spreadsheet could hammer out the factors and weights that result in the overall column (you'd have to type overall in, but the rest of the data is available in exports.. you don't need to do all players in the analysis.. just a handful)

[justafan]

Actually, not using a spreadsheet, I'm writing my own league browser. I'm not sure if it will turn in to anything, just needed something to try out some new controls with. It's nowhere near done, but this is how far along I am...

[NoOne]

Originally Posted by justafan

Actually, not using a spreadsheet, I'm just writing a little league browser. I'm not sure if it will turn in to anything, just needed something to try out some new controls with. It's nowhere near done, but this is how far along I am...

i meant to hammer out the actual equation they use, which you could incorporate into what you are working on as intended. it's possible with a spreadsheet...run a "multiple regression" or somethign liek that? use a short list of players and their ratings with overal being what they 'predict.' (you won't need many because we know the ratings actually add up to the overall in a very consistent manner)

i would think the weights would all add up to 1 or you can use them to figure out each ratings weight relative to each other (use as ratios, that add upto 1) - rounding could cause some amount of inaccuracy.

use editor values (1-200) and 20-80 scales for overall/potential with 100% scouting accuracy, since that overall isn't in the editor - this will minimize the effect of rounding, even if you prefer to use a different scale when playing the game. avoid players with >200 ratings in the sample data you give to the spreadsheet. >max may be handled differently, not sure. maybe it's ignored, maybe it's still in proportion.. either way remove it to learn the weights for normal ratings, then test that equation with someone with >max rating(s).

you'll know immediately if it doesn't predict the overall correctly. and if it does, it works for those guys, too. (obviously test the data-mined algorythm for overall on a normal player from outside the sample data set, too).