February 14, 2001
What's That Park Like?
Being Careful About Adjustments
One of the principal tenets of performance analysis is the idea that a player's statistics are affected by his home ballpark. That's why most serious performance metrics account for the environment in which a player plays using a park factor. This park factor is generally arrived at by comparing the number of runs scored in all games in team's home park and comparing that to the number of runs scored in that team's road games.
I have found that there is good reason to doubt the value of these park factors in a single season. I believe the relative ease or difficulty in hitting in a specific park has a large random aspect, and hence much less weight should be put on park factor for any single season.
This is important for people who calculate, for example, that the performance of Hitter A in any given year was more valuable than that of Hitter B, based on his Equivalent Average or Value Over Replacement Player (VORP). That is, if A played in a home ballpark with a calculated park factor of 105 and B played in a ballpark with a 95 park factor and their performances were comparable, many sabermetric stats will conclude that player B was more productive. But if there is a degree of randomness--call it an error rate--in ballpark factors of five percent or more, there is a chance that in fact each player played in a neutral park.
How can it be that park factors could be off by so much? The answer is limited sample size. A park factor for any given year is based on only 81 home games. In many seasons, some teams will play even fewer (due to non-rescheduled rain-outs and so on). Eighty or so games are not enough to eliminate a large error rate. A park with a park factor of 95 one year and 100 the next very likely cannot be said to be relatively easier to hit in the second year. Instead, it is fair to conclude that by random chance, there was a five percent differential in the two seasons (assuming that major changes did not occur with other home parks in the league).
Over 10 or 15 years, the error rate becomes much smaller as the sample size increases. Unfortunately, it is problematic to average out a park factor over more than a few years because the conditions of one or more of the ballparks in a league change. New stadiums are built, existing stadiums change their dimensions, and abnormal weather patterns have an impact. Nonetheless, a 10-year sample is likely to be more accurate than a one-year accounting.
Let's look at some examples. Excepting 2000, which looks like "an aberration," Oriole Park at Camden Yards appears to be a very average facility to hit in, with only a very slight edge to the pitchers. From 1996-1999, its ballpark factor averaged 98.25, almost exactly average:
1996: 99 1997: 97 1998: 98 1999: 99 2000: 95
But from 1992 through 1995, the first four years of its life, Camden Yards was actually a decent hitters' park, averaging a PF of 104:
1992: 103 1993: 103 1994: 105 1995: 105
What happened to change the personality of Oriole Park? The obvious answer is that after 1995, a number of new ballparks came on line and caused the relative ease of hitting in Camden Yards to go down. Now, The Ballpark in Arlington and Jacobs Field--both renowned as hitters' parks--came to life in 1994. Yet in 1994 and 1995, Camden Yards was much more friendly to hitters than were either. TBiA looked for its first three years almost like Camden Yards looks today:
1994: 100 1995: 98 1996: 100
The Jake looked the same for its first three years:
1994: 100 1995: 100 1996: 100
The only major renovation of an existing park that occurred in 1996 was to the Oakland Coliseum. But that change, which altered the Athletics' home park from being very pitcher-friendly (93.25 average PF) to only slightly less pitcher-friendly (97 average PF), could not have caused a major impact on Oriole Park for the next five years.
Kauffman Stadium in Kansas City was slightly renovated in 1995 and Edison International Field was remodeled in 1998. However, neither change seems to have affected the relative hitter-friendliness of either park. Kauffman Stadium (nee Royals Stadium) is still friendly to offense, and Edison (nee Anaheim Stadium) is still very average.
Starting in 1997, both Jacobs Field and The Ballpark in Arlington inexplicably adopted new personalities: going from average places to hit to great ones. The Jake:
1997: 105 1998: 104 1999: 104 2000: 105
And The Ballpark:
1997: 104 1998: 103 1999: 103 2000: 102
Why did they change? Chances are, they did not.The difference is likely due entirely to randomness, or "error rate."
The Skydome in Toronto has not been remodeled since it opened in 1989. Because it is a quasi-indoor facility, its annual PFs have been quite consistent:
1993: 102 1994: 100 1995: 100 1996: 106 1997: 101 1998: 100 1999: 101 2000: 102
1996 stands out, and no other attribution, save the fact that there is a randomness to park factor heretofore ignored, explains it. The Skydome's park factor of 106 that season means that stats like ERA+, EqA, and SNVA may all have been miscalculated for Blue Jays' players.
Take Woody Williams. In 1996, he posted a 4.73 ERA. The following season for Toronto, Williams improved to a 4.35 ERA. Yet because of the error rate in park factor, Williams's ERA+ declined from 112 to 106. Alex Gonzalez had virtually identical seasons offensively in 1996 and 1907: a 691 OPS in the first season followed by a 689 OPS in the second, with virtually the same secondary numbers. However, because of that erroneous park factor, Gonzalez's EqA rose from .229 in '96 to .243 in '97.
Getting back to what caused the "change in personality" observed at Oriole Park, we might find that part of it is due to the reconfiguration of the American League after the most recent expansion. County Stadium, home of the Milwaukee Brewers in the AL through 1997, was replaced by Tropicana Field, home of the Tampa Bay Devil Rays. Despite Tropicana's reputation as a homerdome, it has had a neutral impact on run scoring so far:
1998: 104 1999: 100 2000: 100
But if County Stadium was a very difficult park in which to hit, then that might go toward explaining why Camden Yards is now seemingly not so friendly to hitters. Unfortunately, County Stadium was actually a hitters' paradise compared with Tropicana Field:
1994: 105 1995: 108 1996: 105 1997: 101
So the change of leagues by the Brewers and the addition of the Devil Rays to the American League explains absolutely nothing about why the park factors of Oriole Park have changed.
Other explanations? The 2000 season was the first full year for two new American League ballparks, Safeco Field in Seattle and Comerica Park in Detroit. By reputation the old parks in those cities were hitters' parks, but let's look at the numbers. The Kingdome (100.7 park factor) was very average its last six full seasons:
1993: 100 1994: 102 1995: 103 1996: 100 1997: 99 1998: 100
Tiger Stadium (100) was exactly average in its last seven years:
1993: 99 1994: 100 1995: 101 1996: 99 1997: 99 1998: 101 1999 : 101
In 2000, Comerica (97 PF) and Safeco (91) were respectively somewhat unfriendly and very unfriendly to hitters. But neither explains in any sense what caused Oriole Park to become relatively tougher to hit in following the 1995 season.
If you only look at the numbers for Yankee Stadium the last two years, you might conclude that the park in the Bronx became easier to hit in at the same time Comerica and Safeco entered the American League:
1999: 91 2000: 104
However, for the nine seasons prior to 1999, Yankee Stadium was a very average facility, averaging a PF of exactly 99. Certainly weather may explain some of the variation over the last two years, but random chance run amok seems a much stronger argument, as the weather was in fact not dramatically different the past two summers in New York and the great decrease in park factor in 1999 came before Comerica and Safeco (mostly) opened.
As with the "changes" at Camden Yards, The Jake, The Metrodome, The Ballpark, and so on, I suggest that randomness and limited sample sizes are the best explanations of the dramatic changes in New York.
What all of this data underscores is that it is unwise to rely too heavily on single-season park factors in rating any one player or team. If a facility, year after year, appears to be very friendly to hitters (such as Coors Field) or to pitchers (such as Dodger Stadium), then some adjustment is rightly made to a players' stats. But to conclude that, say, Oriole Park is not friendly to hitters, based only on what has taken place there the last five years, is to ignore that there is randomness in what appears to be good or bad for hitters and it is to overstate the reliability of park-factor calculations.
(Ed. note: For more on the use of multiple-year vs. single-year park factors, see Clay Davenport's essay on the topic in Baseball Prospectus 2001.)
Rich Rifkin, 36, is an Oakland native and a lifelong fan of the Oakland Athletics who lives in Davis, Calif. He can be reached at email@example.com.