THE BOOK--Playing The Percentages In Baseball

Part of me wants to include something like this in my Marcels forecast - a player who pitches as a starter gets regressed to the average starter, a pitcher who pitches as a reliever gets regressed to the average relieiver, and a pitcher who does both gets regressed to a combination mean based upon their split.

Of course, then they wouldn’t be Marcels, and I am trying desperately to stay out of the projections business right now, mostly because I don’t want to build my own MLEs.

All the above becomes the following:

% of innings as starter
= GS/G * (4.74 / IP * (G-GS) + 1)

So for Masterson, we get:
= 9/36 * (4.74 / 88.333 * (27) + 1)
= .25 * (1.4488 + 1)
= .6122
= 61%

The “4.74” is the difference between number of innings as a starter (5.81) and number of innings as a reliever (1.07). 

Simply a matter of coming up with the best number for each season.  Just picking a season at random (1966), 6.28 minus 1.66 is 4.62.  Picking another season (1979), and we have 6.35 minus 1.70 is 4.65.

Seems pretty constant.  We can just use 4.7 and be on our way here.

And replacement level would be set as:

.470 - .09 times percentInningAsStarter

So, in Masterson’s case, that’s:
.470 - .09*.6122 = .415

That’s the win% against which he is compared.

yes, that is the way it should be done.  For my projections, I use the PBP data to determine exactly how many innings a pitcher pitches in relief or as a starter and then I adjust all the stats to a common baseline or I adjust to the pitcher’s primary role (either a reliever or starter). Then if a pitcher is used in a role which is not his primary role or his role completely changes, then I have to adjust the projections.

But you definitely want to keep these things in mind as there is a huge difference between when a pitcher starts and relieves (over a run per 9, we think).  That is one reason why pitching projections seem worse than they are at times, and another thing we have to be careful about when evaluating pitching projections.  For example, is Pecota or Chone projecting Joba as a starter or reliever?  And did they do the proper adjustments to his historical stats, since he pitched as both.  What about Dempster?  Etc.

Hi Tango,

Thanks for thinking about this problem, as it is something that I essentially punted on.  Yours is a nice workaround, and I agree is much better than just forcing someone into one category or another. 

I’ll probably try to institute this adjustment next time I post my numbers.  It could potentially make a big difference for someone who had a substantial number of starts and yet still fell below my starting pitcher criteria (15 starts or 50%+ of games as a starter iirc).
-j

Also, 4/mgl, while I don’t know about specifics, PECOTA (at least) does include an adjustment based on whether a player was pitching in the past as a starter or reliever.  It was implemented and discussed a few years back (maybe in the 2006 BPro annual?).  Not sure what’s used for the projections, though I imagine it’s based on some combination of past using of the specific player and the history usage of comparable players...with some manual tweaking if needed.
-j

wow, that last sentence didn’t make much sense.  Should read:

though I imagine it’s based on some combination of past usage of the specific player and the historical usage of comparable players...with some manual tweaking if needed.