A Baseball Weblog

Wednesday, December 2, 2009

Classifying relievers, part 2

Earlier this week, I did an introductory post on relievers and how I use AIPGR and Leverage Index to classify them. In this post, I’m going to look more closely at the members of the bullpen.

Defining relief roles

The closer of a relief staff is fairly easy to identify, since unlike other relievers, closers are judged by saves. So, by setting a baseline for a minimum number of saves, we can determine which pitchers are “closers” (for this post, I’ve set the minimum save number at 20). One-out-guys are also pretty easy to identify using AIPGR. Even though they’re often called that, these pitchers never actually wind up getting one out per appearance on average. Since 2005, only one qualified reliever has actually had an AIP under .5 (Mike Myers had an 0.49 on the 2006 Yankees). I decided I would set an arbitrary line at 0.80. I’ve always assumed that long-guys were used in low-leverage situations in which the only requirement was to eat up innings, and that primarily seems to be the case. Again, I've set an arbitrary AIP cut-off for defining long-relievers. I chose 1.40 as a reasonable number. Finally, we have our middle relievers, which don't fit any of the requirements of the other types of pitchers. I defined "middle relievers" as pitchers with AIP scores between 0.80 and 1.40 that do not meet the save requirement of closers. I split middle relievers into two categories, one with gmLI scores under 1.20, the other with scores at or over 1.20. Here are average numbers for qualified relievers since 2005 in each of the categories.


RoleAIPpLIinLIgmLIexLIWPAWPA+WPA-Adjusted WPA*
Closer1.011.871.641.682.001.759.49-7.751.22
Specialty0.691.040.871.251.320.123.79-3.671.02
Longman1.580.750.670.900.910.173.90-3.731.04
Setup11.021.371.191.411.460.877.24-6.371.14
Setup21.060.880.780.971.080.044.26-4.221.00

*WPA+ / |WPA-|. I use it as a rate stat in which 1 indicates a "neutral" contribution to the team.

Again, these are completely arbitrary cutoffs in terms of Leverage Index and AIPGR, but I think it gives us a good baseline nonetheless.

Are there any long relievers that actually pitch meaningful innings?

Long relievers have always been particularly interesting to me. While the numbers suggest pitchers with high AIPs are usually used in "mop-up" situations, there are some clear exceptions. Alfredo Aceves of the World Champion Yankees and J.P. Howell of the AL Champions Rays came to mind as examples from the past two seasons. I decided I'd look back at the numbers, set a minimum gmLI (I chose 1.20 because it seemed comparable to the setup and specialty relief scores), and see what came up. Here are the six pitchers that fit these criteria over the past five seasons:

NameYearAIPpLIgmLIinLIexLIWPAAdjusted WPA*
Travis Harper20051.411.111.490.911.17-1.020.86
Alfredo Aceves 20091.921.071.370.981.182.181.48
Zack Greinke20071.401.071.370.991.402.141.90
J.P. Howell20081.401.261.361.101.493.331.62
Pete Walker20051.710.971.230.871.36-0.100.98
Julio Mateo20051.540.981.200.851.140.321.06


That's a pretty small number of pitchers, and of the six, only Greinke, Aceves, and Howell enjoyed much success.

One final, brief discussion. Sometimes I like to use average pitcher entrance inning (or whatever you want to call it), which can be calculated by averaging the game state in terms of innings for each appearance by a pitcher. I'll use Mariano Rivera as an example. In 2009, Mo entered most games with no outs in the 9th inning, which would mean that his average pitcher entrance inning for those games is 9. He had some games where he entered in the 8th, some others where he entered with outs in the 9th, and one where he entered in the 10th. All of this averages out to a 8.99 average pitcher entrance inning score, which is the mark of a clear closer. I didn't make any charts for this since I don't have a lot of data on it, but it is a pretty good way of further distinguishing bullpen pitchers.

Data is from Fangraphs and Baseball-Reference.

No comments:

Post a Comment