clock menu more-arrow no yes

Filed under:

Projecting Matchups, Prototype

New, 18 comments
Not your average lefty. (Photo by Jeff Golden/Getty Images)
Not your average lefty. (Photo by Jeff Golden/Getty Images)
Getty Images

Editorial Note: This tool is out of date and inaccurate. To find an updated and improved tool (made with help from Sandy Kazmir), go here.

Regressed batter L/R platoon splits are great. I have a tool to help me calculate them, and I use it often, enjoying the experience all the while. It's very handy for evaluating lineup construction, pinch hitting decisions, and bullpen management decisions. It's not, however, quite as handy as it might be. The problem is that when we project a batter's platoon split, we're talking about how we think he'll fair against an average pitcher of a given handedness. Sometimes this is appropriate, but sometimes it is not.

Consider a hypothetical situation when an opposing manager brings in Randy Choate to face Matt Joyce. My tool would tell you that because Randy Choate is a left handed pitcher, Matt Joyce should be expected to produce a .282 wOBA when facing him, which is pretty paltry. As Rays fans will know, however, Choate is not just any left handed pitcher. He's a specialist. He stands on the first-base side of the rubber and throws sweeping breaking balls from a full sidearm angle. He makes a living off of being able to completely neutralize otherwise scary left handed sluggers like Joyce. He faces the best left handed hitters in the game, and he dominates them, to the tune of a 2.41 xFIP. Joyce should be expected to do worse against him than a .282 wOBA, but by how much?

Enter Matchup Projection Tool (.xls version) (9.10.12). This is just a prototype, so I make no assurances that it's mathematically sound, but I wanted to put it out there for feedback and review (in the closed circuit of my own brain, the logic makes sense, but that's like editing your own writing -- everything seems right if you've read over it enough times).

The tool is just another layer on top of the batter platoon split tool. Values you need to enter are in peach. They consist of batter and batter's handedness, pitcher and pitcher's handedness, and whether or not the batter is pinch hitting (if he is pinch hitting, the tool assesses a 10% penalty after all other calculations are complete). From that point, the tool will supply information to my batter platoon split tool, and come back with the batter's projection versus the handedness of the given pitcher.

The other half is what's new. The tool will look up the pitcher's xFIP against the batter's handedness over the past four years. I chose four year numbers rather than career numbers because pitching splits are both way more stable and at times way larger than batting splits. This makes intuitive sense, because while the component skills of batting are things like bat speed and power, which are equally valuable against both lefty and righty pitchers, pitchers are for more individualized. Badenhop's slurve is very different than Rodney's changeup, and should be expected to produce very different results. When you combine this with the fact that pitchers can change their repertoire, I think that using full career numbers may do more harm than good. Consider the case of Joel Peralta. If you looked at his entire career, you'd think he was a mediocre pitcher with a very slight reverse split (basically no split). If you restrict your look to just the past three years, when he changed his approach to rely heavily on his splitter, he now looks like a good pitcher with an extreme reverse split.

After retrieving the pitcher's xFIP, it compares it to a league average xFIP, so as to determine what affect the given pitcher, facing the given handedness has on the outcome of the at bat. Back to our example of Choate, we see that his 2.41 xFIP is a 41% decrease from the league average 4.10 xFIP. I then apply this "pitcher affect" to the projection from the hitter platoon tool, and find that in a confrontation between Matt Joyce and Randy Choate, the expected result is a .166 wOBA. Poor Joyce.

One more thing. If you try to use this tool on a pitcher with less than 150 PAs over the last four years (roughly equivalent to one season from a reliever), it will return "SSS" rather than an xFIP. I'm not using any regression on the pitching side (that's something to be worked into future "production" versions), but I wanted to include some protection against returning completely arbitrary results. If you have ideas about how much to regress the pitcher stats, and towards what, I'd love to hear them.

So play around with this tool. Let me know if you think the logic and the math is right. Just a warning, it's not currently set up to survive copying and pasting into lower rows, so unless you want to add some "$" in yourself, you're limited to one result at a time.

All stats from FanGraphs.