THE BOOK--Playing The Percentages In Baseball

Kind of makes sense.  Suppose the sluggers hit the ball an average of 380 feet, but the weak guys hit it 340 feet, and the fences are 360 feet away.

And suppose the park adds 20 feet to a fly ball (or moves the fences in 20 feet, which amounts to the same thing). 

Now the sluggers hit the ball 400 feet.  Since the fences are 360 feet away, going from 380 to 400 isn’t a big deal.  So not much gain.  But the non-sluggers now hit the ball 380 feet, moving easily past the 360-foot mark.  Big gain!

Put another way: Suppose every player hits the same number of warning-track balls in the old park—say, 8—roughly.  The 8 HR guys double their total, while the sluggers go from 32 to 40, for only a 25% increase.

If that makes sense.

I will run it as a differential and post the results

This is most clear with Barry Bonds at 3Com (or whatever it was called).  LHH were killed there, hitting one-third fewer HR there, than away.  But, Barry hit as many HR at 3Com as on the road.  Why is that?  Because he likely didn’t hit many “just over the fence” HR.  So, taking 10 feet off is HR just won’t do much.

I agree with Phil. It’s a matter of how many balls are hit betwen fence A and fence B, so that they would be homers in park A, but in play in park B.

I split hitters into 7 groups. C is average, 3 groups above, 3 below.

It appears that C and better, those with average or aboce power, all get about the same differential (trying to deal with sample size caused variance). It doesn’t matter how many you hit past the further fence, only how many fall in the space between the two fences.

Those haitters who are below average don’t hit as many balls even to the closer fence, so they see decreasing differentials, although even a +.005 is enough to double an F hitter.

So the graph of differentials for each ballpark factor would be increasing for F, E & D, then be flat C and above.

Rank F E D C B A AA
0.30 -0.004 -0.014 -0.018 -0.034 -0.037 -0.037 -0.055
0.40 -0.003 -0.009 -0.015 -0.023 -0.029 -0.034 -0.032
0.50 -0.004 -0.007 -0.013 -0.016 -0.025 -0.031 -0.027
0.60 -0.003 -0.007 -0.009 -0.019 -0.018 -0.023 +0.013
0.65 -0.001 -0.004 -0.011 -0.014 -0.019 -0.017 -0.019
0.70 -0.002 -0.005 -0.009 -0.013 -0.015 -0.015 -0.009
0.75 -0.001 -0.004 -0.006 -0.011 -0.012 -0.012 -0.021
0.80 -0.001 -0.003 -0.005 -0.006 -0.009 -0.009 -0.020
0.85 -0.001 -0.001 -0.005 -0.006 -0.006 -0.005 -0.003
0.90 -0.001 -0.001 -0.002 -0.001 -0.004 -0.007 -0.002
0.95 +0.000 +0.000 -0.001 -0.002 -0.001 +0.000 +0.000
1.00 +0.000 +0.000 +0.002 +0.002 +0.001 -0.002 -0.002
1.05 +0.000 +0.001 +0.003 +0.002 +0.003 +0.008 +0.004
1.10 +0.002 +0.002 +0.004 +0.005 +0.006 +0.005 +0.001
1.15 +0.002 +0.003 +0.005 +0.006 +0.010 +0.007 +0.010
1.20 +0.003 +0.003 +0.007 +0.012 +0.007 +0.011 +0.011
1.25 +0.003 +0.005 +0.008 +0.012 +0.009 +0.006 +0.019
1.30 +0.004 +0.005 +0.009 +0.013 +0.014 +0.023 +0.017
1.40 +0.001 +0.009 +0.014 +0.013 +0.023 +0.015 +0.015
1.50 +0.005 +0.011 +0.016 +0.018 +0.019 +0.008 +0.018
1.60 +0.008 +0.006 +0.018 +0.024 +0.014 +0.032 +0.052
1.70 +0.008 +0.009 +0.015 +0.019 +0.032 +0.036 +0.028
1.90 +0.006 +0.015 +0.029 +0.026 +0.046 +0.041 +0.027

Is the HR rate HR/BIP or HR/PA?

per balls in play

Brian, can you walk through a couple examples? I’m a little unclear on what the required inputs and the provided outputs are, but it’s interesting stuff…

I ran through the Retro pbp to calculate my park factors. These are multi-year factors, for the life of a ballpark “version”. I took the HR factor, in ratio form, for each ballpark version, then copied and rounded it, so that I could use it to group similar ballparks.

I then reran the factors, filtering for batters. I assigned seven levels of hr/cb, based on career batting totals. What’s the factor for AA HR hitters in Coors, what’s the factor for F HR hitters?

The first chart that appears at StatSpeak shows the factors expressed in traditional ratios. What I printed in this thread is the difference, per contacted ball, in the HR rates caused by the ballpark.

An average AA HR hitter has a hr/cb of .090. The average F is .005. In a 1.25 ballpark, the F will increase .003 to .008, a 60% gain. The AA will add .019 to .109, a 21% gain.

Greg (or anyone who wants to comment), are you worried that the high HR hitters will tend to come from the high HR parks and vice versa?  When I do studies like this, I always like to use out of sample data on one side.  For example, I might use data from the odd years to compute the park factors and then use data from the even years to group the players.  Or something like that.  You have 3 sets of data actually - one for the overall park factors, one to group the batters, and one to do the park factors with the batter filtering.  So I am not sure how I would do it, but as I said, I am always concerned about using “in sample” data and how it might affect the results.

MGL/9: right.

Brian shows parks that depresses HR at a rate of 70%.  And then he shows how much each class of hitter gets hit.  Well… it should be apparent who is going to get hit here… that’s alot of HR to remove and they can only come in one place.

I think what would be better is to take the “average” HR hitter, and establish the park tendency based on those guys.

Furthermore, the classification as to who is a power hitter and not should be based on those “average” parks, say the one-third most close to the mean.

Once you have that, you throw out all the samples that you used to establish your groups and look at the out-of-samples to tell you what’s what.

The method I described in this article uses the league average distribution of fly balls (well, a best approximation of that) by distance and direction to define a test set of trajectories to be imposed into each park, in that park’s average weather, to see how many balls cleared the fences.  Then the factors were normalized against the league average number of balls that made it out, and the results compiled in each of five wedges on the field, along with a method for adjusting the factor for weather.  To get a factor for a particular player in a particular park, you just need to modify the test set to be his spray pattern instead of the league average (maybe I’ll do that...)

Anyway, here’s the article link for anyone who’s interested, and I’m totally open to suggestions, comments, criticism, etc…

http://www.hardballtimes.com/main/article/home-run-park-factor-a-new-approach/

Greg, as I said in my article, I believe yours is the way to go in the future, where the data is available. I am trying to fine tune the normalization process for the Retro records, where all I know is who hit homers in which parks (and often to which field).

I will digest Tom’s comments in 10 and see what I can come up with.