Using ZiPS Hitter Splits for DFS

Utilizing splits in DFS research is obviously a big part of the decision-making process. The platoon split is all too real, and we’d be crazy not to factor it in to our process. I would imagine the most common way DFS players consider platoon splits is by looking at something like wOBA or wRC+ versus left and right handed pitching to discern which players have pronounced splits or even reverse splits on occasion (cough Adam Jones cough).

But looking at stats like wOBA or wRC+ doesn’t exactly translate to a specific daily site’s scoring system. If a player has a wOBA of .360 against left-handed pitching and a wOBA of .340 against right-handed pitching, how does that translate to the counting stats he might put up on a given day?

Plus, you run into all kinds of small sample size issues when looking at past splits. Younger players don’t have much data to rely on to begin with, but when you split up the data into two smaller subsets, it gets even less reliable. And how far back do you look with split data? Sample sizes against left-handed pitching are particularly small, so how many years are you willing to go back to get a decent enough sample size? And how confident are you in using three or four year old numbers?

I’ve long wished we had projections for splits that would address this problem. Projection systems like ZiPS can do the whole weighted averages thing and factor in regression and age. Splits projections can do more than we can by simply looking at a player’s past performance against pitchers of a particular handedness.

Thankfully, Dan Szymborski had made his 2015 ZiPS projections available, and they include splits projections. If Dan has been doing splits projections before this year and I missed it, that would be bittersweet. I’m glad if that kind of info has been available before but sad I somehow missed it. Either way, I intend to use it for DFS purposes this year.

I took Dan’s publicly available spreadsheet containing the splits projections and calculated how many daily fantasy points each player was projected to produce per plate appearance against pitchers of either handedness. I used the DraftKings scoring system just because that’s where I typically play. You can see the results of that calculation in this Google Doc (which is also embedded at  the bottom of the post). ZiPS doesn’t project runs, stolen bases or times caught stealing for splits since those stats are not accrued during the plate appearance (with the exception of runs scored on home runs). You’ll see a sheet in the Doc where I split up each player’s projections in those stats based on their projected plate appearances and on-base percentages against left and right handed pitching.

These per plate appearance projections I’ve calculated are obviously an average and don’t take into account many of the factors you’d consider in your process. To name a few: opposing pitcher, ball park, weather, Vegas lines. What I’d suggest is starting with those other factors to identify the teams in the best position to score runs from which you’d like to roster hitters that day. Once you have that smaller subset of hitters, you can then look at the projections to identify the best options and best values.

But for now let’s assume that on the first Monday of the season every team is facing a league average pitcher in a neutral ballpark in the same weather conditions. In that scenario, consider someone like Paul Goldschmidt. Since 2012, Goldy owns a .438 wOBA against left-handed pitching and a .373 wOBA against right-handed pitching. That split alone could lead you to only roster Goldy when facing a left-handed starter.

You might recognize the fact that Goldy is still very good against right-handed pitching. In fact, Goldy has the 15th best wOBA against right-handed pitchers over the last three years. But because he owns the third best wOBA against left-handed pitchers during that time span, you might assume his high price tag means he’s overpriced when facing a right-handed pitcher.

Goldy’s salary on the first Monday of the season is about two standard deviations higher than the average salary. His projected production per plate appearance against left-handed hitters is more than three standard deviations better than the mean. While not as far above the mean, his projected production per plate appearance against right-handed hitters is still well more than two standard deviations above the mean. Thus, he’s still worth rostering despite owning a wOBA against right-handers that is 65 points lower than what it is against left-handers.

Goldschmidt is a bit of an extreme example, but I’d much rather use splits projections to create site-specific projections as opposed to simply looking at the difference in a player’s past production with and without the platoon advantage.