This year all outfielders will receive an optional “lf/rf” rating. This is a combined rating based on all OF defensive range data. It is recommended that you use the combined rating as it is a more accurate representation of the player’s overall defensive performance. The individual position ratings are based on smaller data samples and frequently result in widely divergent ratings (e.g. a “G” rating in LF and a “K” rating in RF).
Before a combined rating could be created it was necessary to determine conversion rates, e.g. how does range in LF translate to range in RF. I collected the unadjusted (raw rating before playing time adjustments) range ratings for all outfielders from the last four years and calculated each player’s positional range differentials (e.g. lf-rf). An overall differential for the entire pool was then calculated by weighting each player differential by the lowest amount of innings played at either position.
Example: In 2013, Roger Bernadina had an unadjusted range of 0.656 (0.0 = K, 0.5 = F, 1.0 = A) in 146 LF innings, 0.352 in 139 CF innings, and 0.550 in 223 RF innings. This resulted in a lf-rf differential of 0.106 weighted by 146, a cf-lf differential of -0.304 weighted by 139, and a cf-rf differential of -0.198 weighted by 139.
Data for each year:
. lf-rf cf-lf cf-rf cf-of
2014 0.026 -0.129 -0.241 -0.165
2013 -0.012 -0.161 -0.146 -0.161
2012 -0.005 -0.152 -0.175 -0.182
2011 0.030 -0.159 -0.187 -0.179
It became clear that there is no functional difference in LF/RF range, so for the CF conversion rate I decided to calculate the differential between CF range and combined LF/RF range (“cf-of”).
As a result of this data analysis, the combined LF/RF range treats LF and RF range equally and a bonus of 0.175 is applied to CF range. In the case of Bernadina, his raw combined LF/RF range is:
(0.656 * 146 + (0.352 + 0.175) * 139 + 0.550 * 223) / (146 + 139 + 223) = 0.574
Standard rules are then applied for playing time adjustments. Since we are using aggregated data there is no need for a complicated bonus system.
Sean, if I understand your methodology correctly, I think this falls into the fallacy of “if you put everyone into a pool, the pool data washes out the individual differences.” In other words, it may very well be that *on average* there is no true LF/RF difference, but it doesn’t show that for individuals there aren’t.
To take a logical analogy, think of the assertion “There is no indication that certain individuals play better on Tuesdays” (which I agree is true). The way to determine that, though, is not by pooling data from everyone together and seeing that there is no discernible Tuesday/Wednesday difference overall, but by taking individual differences and see for how many players there are a discernible difference, and then see if that player population is greater than what would be expected from data noise. I think that may be what’s happening here.
The issue at question isn’t the possibility than an individual player could in fact be better in LF or RF. I did not set out to prove there are no such players, nor does it have to be true to justify creating a combined rating. There are hitters who truly have a reverse platoon split, yet we still regress them toward a common platoon split which is consistently observable in the entire population. Similarly, individual players could be uniquely affected by their home park yet we uniformly apply park effects. We make these adjustments because they produce more accurate results in aggregate, not because they are intrinsically “correct” for every individual player. What the above data shows is that there is no natural difference between range in LF vs. RF (for players who play both positions), whereas there is a natural and consistently observable difference between range in CF and the corner outfield positions.
Understood, yet at the same time I would say that you would agree that the split regression we employ do not then turn into a situation where everyone gets the same level of splits (and become even vs. LH/RH if switch-hitting), because we do recognize that some of it is small sample and some of it is true split difference based on individual performance. I think, at most, the same approach should be taken here. Perhaps if we wish to, normalize a bit using the other OF data, but I actually think little is gained by that. I feel we should retain the LF/RF differences.
OK, but with platoon splits there’s an identifiable characteristic behind the split (handedness of the batter) and consistent results for the entire population (e.g. LH batters collectively always hit worse vs LH pitchers). What’s your identifiable characteristic for the corner outfielders? If you break down the lf/rf delta based on whether the guy spent more time in LF or RF there is no consistency in the results:
lf-rf as calculated above
broken down whether playing time in lf>rf or rf>lf
......lf>rf..rf>lf..all..
2014 -0.168 +0.201 +0.026
2013 +0.016 -0.062 -0.012
2012 +0.139 -0.214 -0.005
2011 +0.123 -0.064 +0.030
4 yr +0.022 +0.020 +0.022
total
Constrast this with cf-of which is based on an observable characteristic (playing CF) and results in a consistently observable performance split.
(sorry about the table above, the formatting options in comments kinda suck)
There are at least some weirdness’s exposed by these formula:
AJ Pollock for example
CF 7/F/0 – lf 2/I/0 – rf 1/G/0 – lf/rf 7/D/0
Somehow his combined rating is better than any of his individual ratings.
What you’re seeing here is the result of Pollock having most of his playing time in CF (916 inn CF, 38 inn LF, 12 inn RF). He performed relatively poorly in small sample sizes in the corner positions (0.179 raw range in LF, 0.395 raw range in RF). I’d say Pollock is actually a poster child for how the combined rating is a better representation of the player’s defensive performance.
I would say that an even better example of the calculations having odd results is Brian Bogusevic: 9/E LF, 4/E cf, 0/H rf, 9/D rf/lf. I understand the math behind it, but it is so counter-intuitive to have a player achieve a better rating in rf/lf than any rating that he earned in any other single part of his card individually.
Yeah, I agree it is counter-intuitive. Unfortunately I didn’t have time to work on the combined rating until after I had already released the individual range ratings, so “lf/rf” was the only way I could get it on the card. This won’t be an issue going forward.
[…] combined lf/rf rating […]
[…] OF ratings are now based on composite data, if a player has a rating for a position he didn’t actually play in MLB it will be italicized […]