One of the things that came up in the lineup smackdown post (yes, that's still in the works--sorry it's taking so long, folks) is that if we're going to propose lineups vs. LHP and lineups vs. RHP we should really use split projections, rather than overall projections, for our hitters. Otherwise, many of the things that folks may try to adjust for in their lineup--sheltering Jay Bruce lower in the lineup against lefties, for example--would not be rewarded.
Here's the thing about left/right splits, though--the numbers we often hear are, even at the career level, based on sample sizes so small that they are imperfect (at best) predictors of future performance. In other words, just like BA/RISP, you really can't just look at someone's seasonal--or career, even--left/right split and make a conclusive statement about how they will perform in the future. There's just too much noise and volatility in the data.
But if we account sample size and we know past performance, we can make a reasonable estimate at a hitter's true (i.e. future) split size by employing regression. To do this, I've copied Matt Klaassen's methodology described here, which, in turn, is based on the recommendations put forth in The Book by Tango et al. This approach takes a hitters' multi-year left/right split totals and regresses them, based on the sample size, toward the league mean split (6.1% of wOBA for RHB's, 8.4% for LHB's). The more PA's you've had against LHP, the closer your projected split will be to your career totals. But with only a few PA's, you'll be hovering right around mlb average.
Below are the results. I'm listing this in terms of projected wOBA, which is an overall measure of offensive production on the same scale as OBP (it accomplishes what OPS tries but fails to do properly, integrating both on-base and power data into one number). Splits are based on regressed career average splits, centered on each player's CHONE projections for this season. I was using FanGraphs data for the splits, so career split totals "only" go back to 2002.
(table is sortable if you click in the header)
|Name||Bats||PA vs. LHP||CHONE wOBA||Career Split||Regressed Split||Proj. vRHP||Proj. vLHP|
By far and away the largest recorded split (Career Split) on the club has been that of Jay Bruce, who has been good against right-handers over his short career (.354 wOBA), while seeming completely incompetent against lefties (.269 wOBA). He only has 273 PA's against LHP so far, however, which means that we have to regress his split significantly. Based on this methodology, we should probably cut the observed split in half....but even then, it is the largest projected split on the ballclub. Nevertheless, even after accounting for his above-average split, he still projects for a .357 wOBA vs. lefties--5th on the team.
Other large projected splits came from Dickerson, Votto (who actually has a below-average split for a lefty hitter of his quality), and...Brandon Phillips. In fact, the Phillips split surprised me: CHONE + Matt's method only projects him as a .332 wOBA hitter vs. righties, but a .362 wOBA hitter vs. lefties. A .332 hitter is about league average...not bad, but probably not a guy you want hitting cleanup for you either. Against left-handers, however...given that Phillips primary offensive skill is power, and the fact that we have few other power sources against lefties, I have no problem with him hitting cleanup for the team.
This exercise helps show the advantages of a Dickerson/Gomes platoon in LF. Overall, Dickerson and Gomes project to .342 and 0.358 wOBA hitters on the season. But if you leverage them against opposite-handed pitchers, those rates bump up to .347 & .376, respectively. Similarly, a platoon between Stubbs and Dickerson in CF makes a sense.
Overall, it is an interesting exercise, and gives us what I think are much more realistic expected splits for players moving forward. I'll be interested to hear what you think of the numbers.