THE BOOK--Playing The Percentages In Baseball

I did some similar work with this here:

http://stats.mostvaluablenetwork.com/general/expected-park-factors/

I have since been working on something that does a much better job of predicting park factors. The problem is that there are variables we simply don’t understand that have a pretty large effect on how a park plays. Also, some data is tough to find. Your equation correlates pretty well (r = .69) with actual fair territory, though it consistently underpredicts (I had to add 4,868 square feet so that they would match up) the actual amount. I may use it to fill in my missing data points (nine of thirty). Still, I have my doubts that we can ever predict how a park will actually impact play until major league teams actually play games in it.

I have fooled around with some regression equations as well for predicting overall park factors (run scoring factors).

The relevant variables, in rough order of importance (I think), are size of fair territory, altitude, average temperature, size of foul territory, and average fence height.  Other minor variables are average prevailing wind, lighting (visibility), and “smoothness” of outfield walls.

I do think that we can come up with some pretty good park factors before having any actual run scoring data.

David, since you did not use altitude as one of your prediction variables, how in the world did you come up with a predicted HR factor of 1.08 for Coors Field, which has the LARGEST dimensions in ML baseball?  In fact, many of your numbers do not seem right assuming that you used outfield dimensions as your only dependent variables.

One of the things I’ve been a bit disappointed by over the course of this season is the lack of (what I perceived to be building) progress in turning lefty-righty split data into split park factors.  The apex seems to have been an article at Prospectus that was fairly eye-opening to me.  It examined Minute Maid’s HR factor, which was something like 40 points higher for righties than lefties (around 120/80).  Then you look at Clemens and Pettitte and the fact that Pettitte faced a very high percentage of righties, far more than Clemens.  I forget if it was put out there explicitly, but certainly the implication was that Pettitte’s season may have been as impressive as Clemens’, a statement you could only make by breaking into these split factors.

We’ve got loads of split data nowadays.  More importantly, we’ve got a fair number of asymmetrical parks.  It doesn’t just seem natural to move to split park factors; it almost seems silly not too.

Mickey,

My coefficients were probably way off. I can’t find any other way to explain it. I’ll see if I can find the actual numbers I used for the study. Maybe I used some other variables; I don’t remember doing so…

As a point of historical interest, Earnshaw Cook in Percentage Baseball (1964, 2nd ed 1966) used a similar approach for devising an equation for park home run factors (computing surface area of field, treating as equivalent to pi x r-squared / 4, then solving for r to get an average effective distance to the fence).

I believe what Dan is talking about is multiplicative park factors…