ST LOUIS, MO - OCTOBER 12: Mark Kotsay #25 of the Milwaukee Brewers celebrates with third base coach Ed Sedar #6 as Kotsay rounds the bases on his solo home run in the top of the third inning against the St. Louis Cardinals during Game Three of the National League Championship Series at Busch Stadium on October 12, 2011 in St Louis, Missouri. (Photo by Jamie Squire/Getty Images)
It seems like it should be important to know how a certain batter has done against a certain pitcher. It is not. It is not important.
The other day, Brewers manager Ron Roenicke was drafting a lineup to put up against Cardinals righty Chris Carpenter. When it came to picking a center fielder, Roenicke had three options: Nyjer Morgan, Carlos Gomez or Mark Kotsay. He wound up going with Kotsay - not just because Morgan seemed to be in a bit of a slump, but also because Kotsay is left-handed, and was 4-for-11 against Carpenter in his career.
When I read that explanation, I sighed. I imagine that many people sighed. Of course, Kotsay would go on to homer against Carpenter and draw a pair of walks, but that's kind of beside the point. In retrospect, I probably should have chosen a better example.
Maybe it's my imagination, but I feel like I'm being bombarded with microsplits this month more than ever before. By "microsplits", I'm referring to any small-sample statistical split in general, but specifically, for this post, I'm referring to batter vs. pitcher data. Batter vs. pitcher data is everywhere I turn. It's all over Twitter, from reputable and less-reputable sources alike. It's being cited by managers. It's being discussed on the air. All the time, it's being discussed on the air. This guy has this many hits in this many at bats against this other guy. Over and over.
In a vacuum, that's okay. Batter vs. pitcher data is information. There's nothing wrong with information. The problem is how that information is interpreted. That information is treated like it's meaningful, where by "meaningful" I mean "predictive". A lot of people act like, because a match-up was so in the past, so it will continue to be in the future.
And that isn't true. With batter vs. pitcher data, that isn't true. Dave Cameron just wrote a good post about this at FanGraphs a short while back, and you should read it. In short: this data isn't predictive. Thorough examination has shown that this data isn't predictive. It's not like batter vs. pitcher data is completely, 100 percent irrelevant, but it has to be so heavily regressed that you might as well not have the data at all. You're better off looking at the observed overall performances by a given hitter and pitcher.
Okay, so for many of you, this isn't news. It isn't exactly an Internet revelation that batter vs. pitcher data is of little use. But I think it's worth considering why such data is still treated as significant, even though it's essentially been proven that it is not.
The first reason, and the main reason, is that, intuitively, batter vs. pitcher data seems perfect. It seems like exactly the data you should want. Let's say you're a manager putting together a starting lineup. The other team is starting a lefty on the mound. When you're making your lineup, you don't think about your hitters' overall performances - you think about their performances against lefties. You do that because it gives you a better idea of how they'll perform against this particular lefty. But what if they already have an established performance against this particular lefty? In theory, shouldn't that give you an even better idea of how they'll do? What better way to predict how someone will do against someone else than by examining how that specific matchup has gone in the past?
That isn't how it works. But it feels like that should be how it works. It makes so much intuitive sense that it can be hard to believe it doesn't make actual sense.
A second reason, and a lesser reason, is that I think people are wired to not care too much about sample size. It would be one thing if a batter had faced a pitcher 1,000 times. Then that information would be significant. More commonly, a batter has faced a pitcher 10 or 20 or 30 times, and so that information is not significant. The sample size is far too small, spread over too many years, for anything to be made of it.
But it isn't the instinct to worry that a sample is too small. People make quick judgments based on very limited information all the time. Think about your opinions of other people you've just met. Think about cities or countries you've visited once or twice. Yelp is a website built around members publishing reviews of establishments based often on one single experience. That's crazy! But we're always doing it. We seldom wait for a sample to be big enough in life, and many seldom wait for a sample to be big enough in baseball.
We want for batter vs. pitcher data to matter. It seems too perfect for it to not matter. It will never matter. Never, for as long as baseball is played as it's currently played. It's just a meaningless microsplit. There's that old joke about statheads worrying about how a batter does against lefties on Tuesday nights in domes between the fourth and sixth innings. The ingredients of the joke change, but the joke itself stays the same: statheads worry about ridiculous microsplits. In reality, it isn't the statheads who concern themselves with ridiculous microsplits.