Re: Carryover files, ratings, and game engine changes
When I set out (about 7 years ago) to learn this for myself I tested every conceivable statistic against their ratings to look for the highest correlations. I ran three year straight averages (after they changed from 25-25-50), eliminated players with few plate appearances or older players who were obviously nerfed. To my surprise, I concluded they are using only the most basic of statistics in assigning player ratings. They are what you would expect:
Contact and H/9...battting average
Discipline and BB/9...walks per plate appearance
Vision and K/9...strikeouts per at-bat (could be per PA...the numbers are close)
Power and HR/9...HR per at-bat (though I think they might be using ISO here for pitchers)
These statistics all correlate above .95 r-value to the ratings. Other stats which make sense correlate much lower. It is, however, beautiful in its simplicity...and goes back to what I say about the pitcher-batter matchup being a zero-sum game. Every hit, home run, strikeout, etc...is accounted for in BOTH a pitcher and hitter rating.
For this reason, the counterpart formula for any given rating is a simple inverse...and when you extend this out you learn that the rating scale actually extends to 124. So for contact, 124 would be .380 and 0 is .130, so the flip is true for pitchers. A pitcher with a .380 batting average against would be zero. This makes a 99 H/9 pitcher around .180. So, the only point on the scale where they are equal is 62...which is .255 I believe, which is roughly where the league average is.
For defense...I think it's conclusive those ratings are NOT calculate based off anything. I know, because I desperately tried for years to find correlations. Whereas the lowest correlations I found for the batter-pitcher ratings was about .8 (clutch), the highest r-value for any defensive statistic correlated to a rating was around .6, and that was defensive runs saved for second basemen correlated to reaction ratings (if memory serves). The defensive ratings are obviously subjective, and you can tell they're not calculated because most guys don't change from year to year or, if they do change, they jump from something like 60 to 90, which would be impossible if they are basing those values off statistical performance using multi-year averages.
It's more evidence that the fielding and defensive side of things is not much of a priority. It could be so much better...for one, their error rates at the various positions are all out of whack. Too few errors on the left side of the infield, and too many on the right. But that's a whole 'nother conversation.
When I set out (about 7 years ago) to learn this for myself I tested every conceivable statistic against their ratings to look for the highest correlations. I ran three year straight averages (after they changed from 25-25-50), eliminated players with few plate appearances or older players who were obviously nerfed. To my surprise, I concluded they are using only the most basic of statistics in assigning player ratings. They are what you would expect:
Contact and H/9...battting average
Discipline and BB/9...walks per plate appearance
Vision and K/9...strikeouts per at-bat (could be per PA...the numbers are close)
Power and HR/9...HR per at-bat (though I think they might be using ISO here for pitchers)
These statistics all correlate above .95 r-value to the ratings. Other stats which make sense correlate much lower. It is, however, beautiful in its simplicity...and goes back to what I say about the pitcher-batter matchup being a zero-sum game. Every hit, home run, strikeout, etc...is accounted for in BOTH a pitcher and hitter rating.
For this reason, the counterpart formula for any given rating is a simple inverse...and when you extend this out you learn that the rating scale actually extends to 124. So for contact, 124 would be .380 and 0 is .130, so the flip is true for pitchers. A pitcher with a .380 batting average against would be zero. This makes a 99 H/9 pitcher around .180. So, the only point on the scale where they are equal is 62...which is .255 I believe, which is roughly where the league average is.
For defense...I think it's conclusive those ratings are NOT calculate based off anything. I know, because I desperately tried for years to find correlations. Whereas the lowest correlations I found for the batter-pitcher ratings was about .8 (clutch), the highest r-value for any defensive statistic correlated to a rating was around .6, and that was defensive runs saved for second basemen correlated to reaction ratings (if memory serves). The defensive ratings are obviously subjective, and you can tell they're not calculated because most guys don't change from year to year or, if they do change, they jump from something like 60 to 90, which would be impossible if they are basing those values off statistical performance using multi-year averages.
It's more evidence that the fielding and defensive side of things is not much of a priority. It could be so much better...for one, their error rates at the various positions are all out of whack. Too few errors on the left side of the infield, and too many on the right. But that's a whole 'nother conversation.
Comment