Splits
Description:
When looking through a player’s statistics, it’s so very tempting to attempt and draw massive conclusions from their splits. This prospect hit for a .350 average last September? It must mean they’re set to break out this season! This hitter’s batting average is fifteen points higher against lefties than righties? Platoon them! This pitcher has a much lower ERA at home than on the road? Their home park must have a huge effect on their numbers! It’s a natural reaction in all fans, but it’s not always correct.
Split statistics are useful, but there are a couple of caveats to remember:
- Small Sample Size - Whenever trying to draw a conclusion about a player’s talent, large samples are always better than small. See the Sample Size page for exact information on how large of a sample is good enough, but some statistics don’t provide predictive value even after a full season. And when you look at splits, you’re almost always looking at small samples. Yuniesky Betancourt has a .350 batting average against Roy Halladay? Yes, but they’ve only faced each other 25 times and over the course of multiple seasons. Going forward against Halladay, Yuni is much more likely to perform closer to his career batting than at the .350 level. To avoid this, look at a player’s career splits and ignore anything with less than 500-1000 ABs. In other words, specific Batter vs. Pitcher and month-by-month breakdowns are fun to look at, but pretty much useless.
- Regression to the Mean - Say a .350 wOBA player has a big lefty-righty split, something like .300 lefty wOBA / .360 righty wOBA. Going forward, do you expect their true talent level against lefties to be .300 wOBA or .350 wOBA? The correct answer is somewhere in the middle, depending how large a sample you’re using. If it’s a large sample (~1000 ABs versus lefties), you can expect their talent level to be somewhere in the .310-.320 wOBA range. If it’s a small sample, though, you have to expect that their talent level is closer to their career numbers than their split numbers, so maybe something in the .340-.350 range. To determine what a player’s regressed splits are, you can download a calculator from Another Cubs Blog. For more information on regression to the mean, see Regression.
Links for Further Reading:
Estimating Hitter Platoon Skill – FanGraphs
