Park Effects Revisited Again

Dave Paisley

If you've read my prior articles about park effects, you'll know that I'm not a big fan of one-year numbers. Especially not when the park effects calculations are derived merely from runs scored. Statheads who go into incredible detail about the nuances of Runs Created per 27 outs (where you need a Cray supercomputer to see the details) blindly use the crudest possible technique when correcting for the park where a player actually plays his games. The more I look at park effects, the more I'm convinced that using runs scored is a really, really bad way to view effects, especially on a single-season basis.

My latest investigation took me to the National League for the last two years' worth of numbers. I was particularly drawn to the NL because of the weird things happening in San Diego this year. (Check out my previous article for more details.)

The first thing I did was to calculate OPS park factors for the last two seasons for each NL team, then correlate them with the Runs Scored factors. As you might expect, the correlation is very strong. OPS, after all, is a fairly decent measure of the fundamental building blocks of offense - total bases generated.

There are 30 points on the chart, representing the 14 teams' seasons in '97 and the 16 teams' seasons in '98. The two points way to the right are Colorado's '98 and '97 seasons. The point way down to the left is San Diego's '98 season. Another significant deviation is Atlanta's '98 season point. Plenty of raw offense generated at Turner field (OPS Park Factor of 105), but slightly less than average runs scored.

The final two points that are more than one standard deviation away from the line are the Cubs' and Astros' '98 seasons, both of which are significantly below expectations based on the OPS each team generated.

Of all these points, the two that point out the futility of using runs scored as a measure of park effect are the two Colorado points. With OPS park factors of 115 and 118 for the two seasons, the runs generated varied from 33% more than average in '97 to 60% more than average in '98. The '97 number falls perfectly on the correlation line, while the '98 number is way the heck out there. Admittedly, Colorado is an extreme case, but such wild fluctuations in run scoring from minimal variation in OPS (relative to league average) points up how difficult it is to get much meaning from Runs Scored Park Factor.

Note also that all of the oddball points are from the '98 season. '97 was very well-behaved, and it may be that expansion caused some strange effects.

Only time will tell if '98 was an anomalous year or not. But rest assured I'll be there to keep you informed.

 

about the author

Dave Paisley asked Santa for a Cray supercomputer for Christmas. Tell Dave that the existence of Santa is roughly equivalent to Otis Nixon being a great leadoff hitter at drdjp@strikethree.com.