My ideal set of rankings has input from both human and computer sources. In our rankings, we have one set of ranks that are built on projections, meaning that overall our rankings are 25% computer. That’s low for me. But I know that our other three human rankers also use the computer projections to varying degrees, so the number is probably closer to 50% computer, which I like.
But, ostensibly, we have one computer ranker. Jeff Zimmerman uses a proprietary mix of ZiPs and Steamer projections, and then takes the plate appearance projections from our depth charts. Every once in a while a player won’t be updated on the depth charts — human error — but if we stay vigilant, this mix should produce the best computer-generated rankings you can find. The addition of depth chart information can do a lot to undo the biggest flaw in projections: they don’t always know who has the job.
One remaining avenue to deal with with respect to the computer rankings is what to do with the missing plate appearances due to projected injury. Zimmerman has done a lot to advance the state of injury projections, and some of that is now built into the various projections he uses. We know that past DL stints predict future ones, for example. And that age is a heavy factor. But once those PAs and IPs are gone, they can be replaced through the waiver wire.
Read the rest of this entry »