The way the regression works, it takes the career split, regresses it against a certain number of PAs at a league average split, and then distributes that split around the projected wOBA, based upon the career PAs against each handedness.

When I was looking at this, Luke Scott was coming out as the better option than Pena against LHPs, and it’s entirely due to me giving both Scott and Pena an overall projection somewhere around a .350 wOBA.

I’m not saying that I’m right on this (I’ve put very little thought into that), but it’s important to remember that the split is only one component, and the overall projection that you use has a big effect on them.