Printer-friendly

June 17, 2003

Rockies' #634

Park Factors and OBP

by Boyd Nation

Sometimes a big epiphany just leads you back to a better understanding of a mundane truth. Let me walk you through one of mine.

A few months back, I finally hit upon a useful algorithm for determining reasonably accurate park factors for all 287 NCAA Division I baseball programs. Given that, in any given year, a given team will only play 25-30 of the other teams, and that over half of those matchups will not be home-and-home contracts but will involve a smaller program playing only at a larger one (which is of no benefit in determining park factors), I was quite pleased with this discovery.

Current major league park factors, relatively speaking, are a little dull. Sure, you have Coors Field, which routinely comes in around 160. The rest of them, though, hover within about 20% of each other from top to bottom. It matters if you're picking at the fine details of performance analysis, but for a lot of fans it causes the issue to just resolve down to "the Rockies and everyone else." College park factors, on the other hand, have a good bit more range in them, from the lows down in the 60s up to New Mexico, at an astounding 211. In other words, a theoretical game played at New Mexico will produce more than twice as many runs as the same game played at a neutral park like Fresno State.

I then set about finding practical applications for these park factors. The most common use for park factors is to take performance metrics, both team and individual, and place them in a neutral context. So I began thinking of ways to park-adjust statistics and look for players and teams who were actually better or worse than they appeared at face value. Suddenly, it occurred to me that the park factor for runs scored was not the same as the park factor for OBP, and that the relationship between the two was not linear; it was exponential. In other words, if the park factor for OBP increased from 130 to 140, that would result in a greater increase in runs scored than 100 to 110.

Now, the notion that a given park has different factors for different statistics is hardly a new one. Although the most common use of park factors involves just using a single number (usually the factor for runs), more sophisticated investigators such as my hosts here at BP have recognized for years that a given park may increase homers and decrease doubles, for example. I had never seen a discussion of the relationship between increasing OBP and runs, though, although it's implied in the various component models such as Runs Created (RC) or Extrapolated Runs (XR).

To get a feel for the non-linear nature of the increase, consider a pathological case with one game for a team that produces only singles or walks and outs:

OBP    OB    PA   Runs
.250   9     36    2
.300   12    39    4
.400   18    45    8