THE BOOK--Playing The Percentages In Baseball

Yes, might interesting.  I’ll comment as I read.

1. I agree with Dan, that it’s either the umps getting tired, or the back of the bullpen being used.  Easy enough to check.

2. We already knew about the HFA for walks and Ks.  Now we’ve got proof on the “mistake” pitches.  Could be a combination of the two, whereby the umps give the mistake pitches more to the home team, plus the home team is adjusting, however subconsciously, to the expected favorable treatment, and therefore approaches the PA a bit differently.

3. Wow, old pitchers are treated differently.  I’m getting confused on the signs already.  I thought the negative number was good for the pitcher.  Now, Dan is saying the positive number is good for the pitcher.

***

MGL makes a comment on Dan’s blog that one characteristic to look for is at pitchers who live on the edge.  If you don’t make alot of edge pitches, then you have less chances to have a mistake made on the edge. 

So, this should be broken down into two steps: look only at the “edge” pitches, and see what percentage of the time you get the favorable call.  Then, what percentage of all your pitches are “edge” pitches.  If the difference is purely in the second group, then it’s really jsut a question of pitching style.  This may be why the old pitchers shows such a gap, that they must live on the edges, and so, have more chances to earn mistakes.

***

A fascinating article…

Thanks Tom,

I’m planning on doing the breakdown that MGL suggested, and your suggestion should follow pretty easily.

I agree that the signs are confusing - especially for the HFA breakdown, because it’s presented as the batting team - not the pitching team.  So in all cases, a positive is good for the pitchers (i.e. more strikes called than expected), but for that particular split, a positive number under the Visitor batting means good for the home team pitchers.  I definitely should have made it more clear by tying it to the pitching team and not the batting team.

Dan, can you explain exactly how you computed the run value for each pitcher (or group of pitchers), with an example or two? Thanks.

Sure thing.

The first thing I did was determine how many called pitches there were that were measured by PITCHf/x in 2007 (166433).  I then looked at which pitches were called balls but were inside John Walsh’s strike zone definition, and which were called strikes but were outside his definition for both left handed and right handed batters.

Using walk rate as an example (for walk rate I had to eliminate rookies since they didn’t have a pre 2007 walk rate, so the sample is only 148550 called pitches):

Pitchers with a walk rate of less than 2.5 BB/9 had 21241 called pitches.  614 of those were to RHB and were called balls when they were actually in the strike zone.  1091 were to RHB and were called strikes but were outside the strike zone.  471 were to LHB, called balls, and inside the zone.  982 were to LHB, called strikes, and outside the zone. 

Crediting the pitcher with the ones called strikes and debiting the ones called balls (1091 + 982 - 614 - 471) gives the group of pitchers 988 more strikes than would be expected if Walsh’s strike zones were cut and dry.

Of course they’re not, so I took the numbers for the whole league - which was 5793 more strikes than expected from Walsh’s zones in 148550 pitches, or .038997 extra strikes per pitch.

Based on that, you’d expect the low group to have roughly 828 more strikes than expected from Walsh’s zones.  But it really had 988 more, or 160 more than would be expected from the league aggregate numbers (I know I’m using two different meanings of expected here - one based on Walsh’s zone and one based on league aggregate numbers).

Finally, I scale that 160 number to be per 150 opportunities (chosen by Jonathan Hale to represent one game worth of opportunities) - 160/21241 * 150 = 1.127.  Multiply that by the .161 runs per opportunity (figured here) and you get .1815 runs per game.

Someone over at Beyond the Box Score suggested I may have calculated the .161 runs per opportunity incorrectly by weighting by the total number of plate appearances at a given count rather than just those that included called pitches, so I’ll be working on figuring that out.

I know there are plenty of concerns with this type of approach - including sample size, data quality (I assume that measurement errors from PITCHf/x cancel out), selection bias and interdependence, but I’m not entirely certain how to correct for most of them.  Any suggestions would be greatly appreciated.