I'm not quite sure if this is kosher, but there were a lot of questions about ZiPS and Kevin Frandsen in a recent thread and I was out of town and unable to give responses. Well, I could have on my cell phone, but my phone doesn't have a swanktastic fold-out or slide-out keyboard and I'd rather take a bath with a toaster than write long sentences without one.
First off, one has to take into account (I'll drop David a note) that the rest-of-season projections on Fangraphs do not take factor in the seasonal minor league performance, simply because David doesn't yet have a mechanism for minor league translations. I recently provided him something for him to turn into such, but he's had a lot to do in maintaining the website before saving me my coding work. When I do RoS projections as part of entries on BTF or Twitter, I do use minor league translations. The Fangraphs RoS projection has no way of knowing that I have Frandsen's 2009 translation at 305/344/415.
There are also a few reasons why Frandsen's 2009 projection was better than his 2008 one. First, a missing season isn't as serious for a hitter as it is a pitcher. Also, SF's expected park factor went up a bit and some less-stellar performances, moved farther into the rear-view mirror.
Here's what I get for Frandsen's performances (minor league translations plus major league statistics)
Year Overall 2004 255/290/314 2005 292/337/372 2006 274/326/375 2007 275/341/381 2008 000/000/000So there wasn't a huge shift in his projection. It would have been higher without him missing 2008 (274/332/385 rather than the 269/325/370 actually predicted).
Majors Only Translation Only Majors and Translation 2009 071/161/071 305/344/415 247/295/344
It helps to think of a projection not just as a seasonal line, but an array of probabilities. ZiPS and other projection systems simply see the midpoint as the likeliest. ZiPS knows there's a chance that Frandsen will have a .900 OPS and also a chance that he'll have a .600 OPS. If you picture a large curve loosely resembling a bell curve, imagine every bit of new data hitting the curve from one side or the other, moving that curve to the right or the left or crushing it down. (This is essentially a really simple way to think of Bayesian inference).
Without the minor league data, all ZiPS on Fangraphs sees as new data is that 2-for-28 stretch. Small sample, sure, but Frandsen's missing 2008 already increases the uncertainty to begin with, enough to make the best guess move from 269/325/370 to 258/316/353. Add that to the numbers already "in the bag" and it's not a great season line.
It's less bleak when we see have his minor league performance for 2009 included. It's a better line, but a larger sample size, so the RoS projection becomes 262/316/361.
You can't not expect the projection to drop off a bit. You take an uncertain player with a .695 projected OPS and give them a little less than half a season of a .639 OPS and you have to lower your expectations (in this case, to .677).