Finally, the park factors
I have completed the outfield park factors to be used with zone rating. Some of them, I'm not totally comfortable with, but this is the best I could do with the data I have.
Here's how I did it: I looked at every outfielder at each position from 1987 to 2006, and totalled their chances (balls in zone) and zone rating runs. So, for example, Brian Giles has seen 394 chances while playing for a team in Jacobs field, with a total run value of 15.9. While playing left field for teams that do not play in Jacobs, Giles has seen 929 chances and a -31 run value. Note that I do not have home/ road splits, which would make the calculations easier and much more accurate. I compare Giles while playing for the Indians to Giles playing left for other teams.
The next step is the match innings. Reduce Giles' 929 'other' chances to 394 for purposes of the Jacobs field LF rating. Repeat for every outfielder for 20 years, and total. The matched innings method has some cool things about it, and some problems.
One problem is a player dominating the sample. Giles represents 2/3 of the sample for PNC park. Ken Griffey Jr. represents 3/4 of the Great American CF sample (amazing since he's missed seemingly 3/4 of the games). I chose to reduce their matched innings so that no one player is more than 50% of the sample.
A cool thing is that Manny Ramirez is exactly 0% of the sample for Fenway, left field. Manny is, more than anyone else, the player who inspired me to start this work. Its just that he's never played left field anywhere else, being a RF for the Indians. I like this because I'm using a diverse group of players and not Manny to figure the stat that will be used to evaluate Manny.
Ok, matched innings done, samples cleaned up. What next? Did someone say regress?
I thought so. But I'm not regressing to zero. There's no reason to regress a 37 foot wall like Fenway to the same value as an 8 foot wall. Fenway is regressed to other 37 foot walls, in other words, not regressed at all. There are a number of walls between 18 and 25 feet, and they are all regressed together, to their mean (which is less than Fenway), and finally all other parks regressed to what's left over. The run values to regress to are: (per 400 chances), -14 Fenway, -9 other big walls, and +1 other parks. Regression is 50% at chances = 470.
STATS inc. did some revisions to the zones they use for Coors field around 2001, so for Coors I only use 2001+ data. Before that, Coors severely depressed zone ratings.
The one that really sticks out at me is PNC left field, with a -13 rating. The big wall in that park is RF (also a -13 rating). If Brian Giles is excluded completely, the PNC LF rating would be -7. Giles was a good left fielder in Cleveland, and despite his age a good one in San Diego, but had poor zone ratings in PNC left. Whether this is a real park effect or a fluke I don't know.
In just a bit I'll present a link to the data. There are two ways to use it. One is to figure run value per 400 chances. The other is straight from raw zone rating. Example: Manny Ramirez has a .750 zone rating for -25 runs per 400 chances. Either:
A) Add 14 runs back to his total, now -11 or
B) Add .042 to his zone rating, now .792, and figure plays made and run value from there.
And here it is:
Outfield Park Factors