The Universal DH Will Not Kill Your Fantasy Plans by Dan Szymborski May 26, 2020 Amid the difficulties that need to be hammered out before a theoretical 2020 season gets going, probably the easiest to sort out is the universal DH. Baseball has been inching closer to this outcome — which I’ve felt was inevitable as soon as daily interleague play became a thing — for a while now, and instituting it for an oddball 2020 season is probably the least controversial decision to make. But while it registers as easy when compared to the other issues facing players and league decision makers, for projections, it opens up a whole new can of worms. When ZiPS projects pitchers, it knows the team and (so it believes) the general league structure. Every club plays 162 games, mostly against teams in their own league, and in interleague play in AL parks, the NL uses the DH. Those things have been thrown into disarray by most of the proposed 2020 changes. 82 games instead of 162 is fairly easy to deal with; you just have to realize you’re going to be inaccurate. Swapping out pitchers for designated hitters is a little different. To get an idea of what offense will look like and who it would affect, which is important for both real life and fantasy purposes, let’s start by looking at non-pitcher offensive numbers for both leagues from 2008-2019: AL Rate Changes Without Pitchers Hitting League Year HR BB SO SB CS AVG OBP SLG RC/G AL 2008 0% 0% -1% 0% 0% .001 .001 .001 1% AL 2009 0% 0% 0% 0% 0% .001 .001 .001 1% AL 2010 0% 0% 0% 0% 0% .001 .001 .001 1% AL 2011 0% 0% 0% 0% 0% .001 .001 .001 1% AL 2012 0% 0% 0% 0% 0% .000 .001 .001 1% AL 2013 0% 0% 0% 0% 0% .001 .001 .001 1% AL 2014 0% 0% 0% 0% 0% .001 .001 .001 1% AL 2015 0% 0% 0% 0% 0% .001 .001 .001 1% AL 2016 0% 0% 0% 0% 0% .001 .001 .001 1% AL 2017 0% 0% 0% 0% 0% .001 .001 .001 1% AL 2018 0% 0% -1% 0% 0% .001 .001 .001 1% AL 2019 0% 0% 0% 0% 0% .001 .001 .001 1% AL Total 0% 0% 0% 0% 0% .001 .001 .001 1% The AL, unsurprisingly, sees little change in the overall league level of offense. Road games against NL teams represent a very small percentage of the schedule, so the overall league level of offense changes by very small amounts. For the NL, the differences are much more significant: NL Rate Changes Without Pitchers Hitting League Year HR BB SO SB CS AVG OBP SLG RC/G NL 2008 5% 3% -5% 6% 5% .007 .008 .013 13% NL 2009 5% 3% -5% 6% 5% .007 .008 .013 13% NL 2010 5% 4% -4% 6% 6% .007 .008 .013 13% NL 2011 5% 3% -4% 6% 5% .006 .008 .012 13% NL 2012 5% 4% -5% 6% 6% .007 .008 .013 14% NL 2013 5% 4% -5% 5% 5% .007 .008 .012 13% NL 2014 5% 4% -4% 6% 6% .007 .008 .013 14% NL 2015 5% 4% -5% 6% 5% .007 .008 .013 13% NL 2016 5% 4% -5% 5% 5% .007 .008 .013 13% NL 2017 5% 4% -4% 5% 5% .007 .009 .014 13% NL 2018 5% 4% -4% 5% 5% .007 .009 .014 13% NL 2019 5% 4% -5% 5% 5% .006 .008 .014 12% NL Total 5% 4% -5% 6% 5% .007 .008 .013 13% Conveniently, pitcher-hitting hasn’t changed much over the last dozen years, so the changes in offense tend to be consistent. We’re almost always talking about six or seven points of batting average, somewhere around 20 points of OPS, and a 5% bump in homer rate with a similar decline in strikeouts. After all, pitchers are lousy hitters and unless they’re true two-way players (Shohei Ohtani) or extreme outliers (Wes Ferrell), a good hitting pitcher is a lousy hitter, in league terms. Madison Bumgarner doesn’t get his half-win a year on offense because he’s a good hitter, but because he’s less horrifying at the plate than most of his peers. Bumgarner’s career 45 wRC+ has been enough to eke out an additional five wins of value, or just under 15% of his overall career value. But is it as easy as simply adjusting pitchers by those changes across the board and calling it a day? Not really. We would still need to know if there’s a type of pitcher who gets a larger benefit than usual from facing pitchers rather than actual hitters. To answer this question, I looked at all pitchers who faced at least 200 pitcher-hitters from 2008 to 2020. That number might be smaller than you think; only 103 pitchers have done so. We’re going to be firmly in small sample size territory, unfortunately, but there’s no way around it. It’d be nice if we could get a million years of baseball in DH leagues and a million years without, but reality insists on rearing its usual, obnoxious head. From there, I looked at who had the largest difference between pitcher and non-pitcher OPS against. Let’s start with the pitchers with the largest and smallest differences: Pitchers with Largest P/Non-P Splits Player P OPS Non-P OPS DIFF Derek Lowe .194 .781 .587 Robbie Ray .198 .778 .580 Mike Leake .240 .786 .546 Clayton Richard .265 .811 .546 Chris Volstad .298 .832 .534 Charlie Morton .223 .747 .524 Andrew Cashner .239 .761 .522 Jordan Lyles .305 .819 .514 Wade Miley .265 .770 .505 Jason Marquis .317 .821 .504 A.J. Burnett .251 .750 .499 Anthony DeSclafani .284 .781 .497 Chase Anderson .292 .788 .496 Roy Oswalt .254 .749 .495 Jordan Zimmermann .266 .760 .494 Tyson Ross .242 .727 .485 Mike Pelfrey .320 .804 .484 Madison Bumgarner .210 .692 .482 Matt Harvey .258 .738 .480 Ryan Vogelsong .295 .775 .480 Pitchers with Smallest P/Non-P Splits Player P OPS Non-P OPS DIFF R.A. Dickey .456 .732 .276 Jacob deGrom .348 .625 .277 Jeff Locke .486 .779 .293 Patrick Corbin .437 .741 .304 Clayton Kershaw .285 .601 .316 Ricky Nolasco .448 .778 .330 Jon Lester .372 .705 .333 Mike Minor .394 .735 .341 Julio Teheran .383 .724 .341 Jason Hammel .429 .772 .343 Zach Duke .446 .792 .346 Travis Wood .405 .753 .348 Livan Hernandez .478 .833 .355 Tim Lincecum .353 .715 .362 Ubaldo Jimenez .375 .739 .364 Dan Haren .352 .721 .369 Carlos Martinez .323 .693 .370 Johnny Cueto .342 .714 .372 Johan Santana .323 .700 .377 Kyle Hendricks .304 .681 .377 These are interesting lists, but as is typically the case with these sorts of things, the differences between the groups are non-obvious to the naked eye. We can’t really say “OK, Madison Bumgarner is hurt more by the presence of a DH than Clayton Kershaw” since we don’t actually know whether these are predictive. And from a year-to-year standpoint, they’re not. The year-to-year r-squared for pitcher vs. non-pitcher difference is 0.002. That number’s plagued by even more inadequate sample sizes, of course, so if we’re going to find out which pitchers are likely to be hurt more by no longer facing pitchers, we’ll need to look at the characteristics of the groups. So, as is my wont, I did some exploratory data analysis. I won’t go too deep into the craggy details, but I had to do some dimensionality reduction. In simple terms, if we’re making a predictive model for pitcher-vs-non-pitcher splits, which is necessary for our purposes, we have several techniques to defenestrate the explanatory variables that, well, just aren’t very explanatory. Using our 103 pitchers, I tested every variable I could think of, and each one went out the window, whether they were traditional rate stats (HR/9, K/9, etc.), pitch usage stats (fastball velocity, breaking ball percentage, etc.), or plate discipline stats (Zone%, SwStrike%, etc.). General measures of quality such as overall ERA or OPS against also got the axe. So did things like pitcher-handedness or more out-there things such as age, pace, or height. Except for one thing. Where everything else had no value, there was one stat that actually had relevance to our terrible model. Since guessing games are fun, look at the first list of pitchers and ask yourself what they do have in common. And if you look at the next chart before doing so, you’ll have to live with the realization that you’re a dirty cheater! Pitchers with Largest P/Non-P Splits Player P OPS Non-P OPS DIFF GB Percentile Derek Lowe .194 .781 .587 100% Robbie Ray .198 .778 .580 15% Mike Leake .240 .786 .546 85% Clayton Richard .265 .811 .546 95% Chris Volstad .298 .832 .534 87% Charlie Morton .223 .747 .524 93% Andrew Cashner .239 .761 .522 68% Jordan Lyles .305 .819 .514 70% Wade Miley .265 .770 .505 77% Jason Marquis .317 .821 .504 90% A.J. Burnett .251 .750 .499 83% Anthony DeSclafani .284 .781 .497 28% Chase Anderson .292 .788 .496 8% Roy Oswalt .254 .749 .495 55% Jordan Zimmermann .266 .760 .494 16% Tyson Ross .242 .727 .485 94% Mike Pelfrey .320 .804 .484 75% Madison Bumgarner .210 .692 .482 32% Matt Harvey .258 .738 .480 34% Ryan Vogelsong .295 .775 .480 22% There are a lot of extreme groundball pitchers on this list. Overall, the percentile a pitcher ranks for GB% only explains about 20% of the variance of pitcher vs. non-pitcher discrepancy, but it’s the only thing that proved to be even slightly useful. Now, what does this change for fantasy purposes? In the end, our noise still remains stronger than the signal, and it’s only enough to gently nudge a few stats slightly over the course of the season. Our very simple model would project Derek Lowe to be hurt by only one point of BABIP, one walk, and half-a-homer per year versus an extreme fly ball pitcher. In other words, it matters a skosh, enough that you maybe take the fly baller over the groundballer, all things being equal otherwise, but almost every decision is one in which all things are not equal otherwise. Knock NL pitchers down a peg in your valuations, but don’t sweat it beyond that.