THE BOOK--Playing The Percentages In Baseball

I liked very much Kinkaid’s explanation of how rounding errors can introduce additional error in ROS projections, especially at the end of the year.  However, his article does not address the interesting question that he raises in this quote from the article’s seventh paragraph.

To illustrate, consider Ryan Ludwick’s ZiPS ROS wOBA projection. Right now, it is .375. Before the season started, ZiPS had Ludwick pegged for a .372 wOBA. He has since aged a bit, posted a .334 figure for the year, and moved to a worse park for hitters. How did his projection go up?

I believe the answer is that the ZIPS ROS projection is not really an updated projection that includes 2010 performance or age or park changes, but is simply the original preseason projection prorated for the number of PA left in the season in the component manner that Kinkaid describes.  This is completely different from the updated projections that Brian is providing at THT, a source that I wish Kinkaid had mentioned in his original article.

Peter, I don’t think that’s right:

http://www.baseballthinkfactory.org/files/oracle/discussion/2009_zips_in_season_projection_tool/

6. Dan Szymborski Posted: May 29, 2009 at 05:17 PM (#3199265)
Generally relative weights to current season mixed with previous projection (sorry, missed the q before).

It does appear that I am wrong.  My apologies to Dan.

Peter, I appreciate the support, but I did notice that Kincaid said “freely available sources” which by that definition ruled me out.

Kincaid was presenting the rounding errors as a possible explanation for the seemingly erratic ZiPS numbers, but I don’t recall any solid proof beyond speculation. I’d like to give Dan credit to not do this kind of mistake.

I keep all my rates as double precision throughout the entire process, only rounding or setting numbers of decimals in the final report to be published.

But Dan wouldn’t provide wOBA… David would calculate that himself.

I like Rally’s solution to all of this.  CHONE updated projections use whole season playing time projections throughout the season, but they are updated to include 2010 information to figure the number of singles, doubles, homers, etc, per PA.  This way you never get into rounding issues like Kincaid showed and they always work as an estimate of true talent.

@Brian/5, I’m glad that the wOBAs and such that you report with Oliver are based on non-rounded data, but if one uses the whole numbers to calculate one’s own values (like custom lwts or fantasy points), then the rounding issues become a big deal again. 

As a frequent user of in-season projections, I’d really just prefer that everyone just stick with whole-season playing time projections whenever they update their projections.
-j

The THT Forecasts include both rest of year and end of season projections, although each of the components and rates that are published are rounded. Perhaps we can have the downloadable version have extra decimals.

I’m down with that solution. As it is, I haven’t been able to use Oliver or ZiPS for at least a few weeks--less for guys with part time roles. 
-j

Kincaid has it right on.  I didn’t really design it to display a precise measure of true talent, just a simple best guess for the rest of the season.  It wasn’t really a mistake, so much as a lack of vision of what people would use ROS projections for - at the time I was just using it as an info box for transactions and I would get lots of emails for every single ROS projection because people seemed to have nothing better to do than “double-check” presented BA/OBP/SLG to the presented batting line. 

I’ll talk to David Appelman this winter about having the wOBA derived from unrounded numbers.