Subjectivity in MLB
Baseball is widely considered the most predictable sport over the long-term; with 162 games and circumstances that vary only slightly from game to game, we see incredible consistency over the course of a season. The players who are “supposed” to hit .300 usually do, the pitchers who have dominated in the past typically continue to excel, and so on.
Some have taken that to mean that there’s very little use for subjectivity in the game. While we need to make subjective judgments all the time in a sport like football that’s volatile and in which players are so dependent on one another, baseball lends itself well—probably better than any other sport—to a Sabermetrics-driven approach.
However, I’m going to disagree with the popular opinion that the process of creating daily fantasy baseball projections and lineups is a completely objective one. While I’m fully in support of analytics-based methods, we don’t always need to blindly follow the numbers. Here’s why.
Baseball Is a Different Sport on the Daily Level
Baseball is incredibly predictable over long stretches, but it’s highly volatile on a per-game basis. Think about how often star players come up empty-handed; with just five or so plate appearances per game, there are many times when studs don’t even get on base. Compare that to a sport like basketball, which has very high night-to-night consistency. We know LeBron James will score a certain number of points, but we can’t guarantee anything from a baseball player, even a pitcher.
You can and should still emphasize long-term numbers in your day-to-day decisions. If you’re continually making smart long-term choices, you’re going to come out on top after a while. But it’s important to realize that a high level of daily volatility affects how confident we can be in our projections and lineups.
That last sentence is key. Imagine you could hypothetically build a supercomputer that could perfectly predict MLB games—every inning, every hit, every pitch. In that case, you’d never want to make subjective judgments; you just follow the numbers. On the other hand, if baseball were completely random, it really wouldn’t matter who you pick or how you do it.
Well, baseball isn’t completely random by any means, but we certainly can’t be so confident in our projections that there’s no room for interpretation. I’m not saying there’s not one true optimal lineup out there—because there might be—but we need to be concerned with our ability to actually identify that lineup.
A Low Deviation in Expected Output
In some sports, there’s a massive gap between the stars and scrubs. Calvin Johnson might average over 100 yards per game, while other No. 1 wide receivers could check in at half that amount.
No such gap exists in baseball. The best batters in the world might get a hit 35 percent of the time, while well below-average ones still get a hit 25 percent of the time. That difference will manifest itself over the course of a season, but it’s going to be really difficult to predict on a nightly basis.
That doesn’t mean we should ever be favoring sub-optimal players, but just that we need to account for the fact that we’re not always going to be able to identify the best options on a given day. The low deviation in stats makes it more difficult to accurately project players from day to day, which increases the potential for going against the numbers at times.
Fragility in Projections and Lineups
Ultimately, this all comes down to the inherent fragility of projections and lineups. Let’s assume we’re projecting the hits for three players:
- Player X: 1.51 hits
- Player Y: 1.50 hits
- Player Z: 1.49 hits
The question is “how confident can we be that Player X is really the best bet to get the most hits?” Well, not very. Since there’s a lot of noise in daily MLB results and a very small deviation between these players, it would be a mistake to blindly pick Player X due solely to the hit projections. He might indeed be the best option and, if all things are equal, we should pick him.
But what if there are other factors at play that can’t be easily quantified? Maybe Player Z is playing at Wrigley with the wind blowing out straight to center, but you don’t account for wind speed in your MLB projection model. Is Player X still the best bet? Probably not.
The point is that, while we should certainly use analytics to help us project players and build lineups, we shouldn’t pick the top-projected players, no matter what, without considering the fact that we could just be wrong. If you can legitimately identify the difference between 0.01 hits per night, you possess a skill that no one else in the world has.
Whenever we’re using projections to build lineups, we should ask ourselves “Do these projections/rankings get thrown off with very minor changes?” If the answer is yes, it’s a sign that the projections are fragile and, although still useful, shouldn’t be taken as “absolute truth.”
In summary, using analytics or a projection model to pick players is a good thing. Selecting the top values, no matter what, from an inherently fragile system, however, requires a level of faith that runs counterintuitive to why the model would be created in the first place.
Stacking in MLB
We’ll have plenty of information and data on stacking teammates in MLB, but it has a lot of implications on daily baseball strategy. Specifically, the symbiotic relationship between teammates—the fact that they’re dependent on one another for production—can change how you analyze your projections and build lineups.
If you’re building lineups based only on projections without any kind of subjectivity in the process, you’ll need to select players based solely on their numbers, ignoring teams altogether. We know that’s not a smart strategy in GPPs, where stacking is almost a prerequisite for winning.
But can we completely quantify the positive effects of stacking? Take a look at this example.
- Player A on Team 1: 12.5 points
- Player B on Team 2: 12.5 points
- Players C and D on Team 3: Both projected at 12.0 points
Assume we’re deciding between a duo of Player A and Player B, projected at a combined 25 points on separate teams, versus Players C and D who play on the same team but are projected at only 24 points. Which combination is best?
For safety, it’s probably the first one, for upside, the latter. Can we completely quantify this effect? What if we wanted upside but Players C and D were projected at a combined 23 points?
The point is that the dependent nature of teammates can alter optimal decisions in a way that isn’t reflected in traditional projections and, at least at this point in time, requires some subjectivity.
Again, I’m not at all advocating that you choose players based on your gut feel—math should be the backbone of your approach to any sport—but blindly sticking to the numbers in every case, especially when they’re extremely close, is counterproductive.