Simulating Batting Order: Just Games?

Certain sabermetric analyses, such as those of batting order, are based on simulations. Some will argue that things based on simulations are less likely to gain practical acceptance in baseball. This may be true, but does it really make sense to rule out using simulations for baseball strategy given the prevalence of simulation in the contemporary world?

I’ve addressed batting order and related issues recently, and some may think it is out of proportion with their importance. Of course, that significance is relative to one’s perspective. Indeed, baseball and professional sports in general aren’t “significant” from a larger perspective, so why blog about then at all? More to the point, I have additional thoughts that were brought back to mind after reading Scott McKinney’s stimulating manifesto on sabermetric managing earlier this week. One common response to a posts like McKinney’s is to point out that managers will be reluctant to try new strategies because even if they are the right ones, if they don’t obviously “work” (or even if they do!) they are likely to face a backlash from traditionalists. Even if the front office backs them up, so much of the manager’s job depends on perception that it might be “not worth it” to step off of the beaten path. I’ve discussed the problem of player attitudes and reactions to applied sabermetrics before and what might be done about it. That is a separate discussion; here I will briefly discuss how one might respond to some (not all) predictable objections to using simulations to figure out the best batting order.

I won’t review the batting order recommendations made in The Book yet again, as they are outlined in some of the links above. While those recommendations are given as general rules, as the authors acknowledge elsewhere, those general rules (e.g., the best hitters should hit first, second, and fourth) aren’t necessarily always the case, they are just the most common results of Markov analysis and simulations. The results will vary given the projected specific abilities of the players. Why are simulations (understood broadly to include simulations proper as well as Markov Chains — strictly speaking there is a difference) necessary? By way of contrast, we can test the results of a model-based (e.g., Markov chain, Base Runs) way of deriving linear weights against empirically-derived linear weights. However, in the case of batting order, there isn’t an direct “empirical test,” since we can’t re-run the whole season with the same players against the same opponents all at the same true talent level but with different batting orders even once, let alone enough times to get an adequate sample. So we need to model or simulate different options based on probabilities.

Objections to making baseball decisions based on things that look like simply “games” are understandable (if a bit ironic). However, one could also note a similar thing about using a traditional batting order against just a random one — it isn’t based on a comparable empirical test, either. Moreover, in fields other than baseball (and to be fair, there are likely some sort of simulations and modelling used in baseball front offices, if not necessarily for traditionally “managerial” decision-making) simulations and modelling are used as the basis for decisions all the time. “Wargaming” is one obvious example going back in history before computers, but now adapted to computers. Contemporary non-sports uses of simulations can be found in fields from healthcare to urban planning to engineering. Of course, there is some empirical background to those simulations, but that is also true of simulating baseball — not only using the historical performance to estimate player true talent, but to estimate the various outcomes of base-out and game states.

Naturally, the simulation has to be constructed properly (not to mention having good player projections!), and there is room to debate just how to do that (e.g., Markov versus a strict simulation). I’ll leave those discussions to the smart people. In the meantime, if a manager or fan wonders whether or not it is right to use a “silly game” to help make a decision about how to play a totally non-silly game involving balls, sticks, gloves, managers in uniform (wish they had this in basketball, am I right Stan Van Gundy?), and hot dog-tossing lions, I would simply respond that those silly games aren’t all that different from those used in industries just as, perhaps even more non-silly than baseball.

We hoped you liked reading Simulating Batting Order: Just Games? by Matt Klaassen!

Please support FanGraphs by becoming a member. We publish thousands of articles a year, host multiple podcasts, and have an ever growing database of baseball stats.

FanGraphs does not have a paywall. With your membership, we can continue to offer the content you've come to rely on and add to our unique baseball coverage.

Support FanGraphs




Matt Klaassen reads and writes obituaries in the Greater Toronto Area. If you can't get enough of him, follow him on Twitter.

newest oldest most voted
Oscar
Guest
Oscar

The problem with using simulations is that the programmer’s assumptions about how baseball works will be “validated” by the program. A baseball sim will have to either include or not include a protection effect, for example. Using the sim to then test lineup orders to see whether or not protection exists will just reveal the rules of the sim, not of baseball.

You can’t use sims to test fundamental laws of baseball without being INCREDIBLY careful.

rogerfan
Member
rogerfan

You could try to measure “protection” in real life. Simply compare hitter A hitting in front of a good hitter B versus hitting in front of a average hitter C. I’m sure there are enough natural cases of this happening to be able to at least get in the ballpark estimate of the effects of protection.

AJS
Guest
AJS

A question about those protection simulations – do they find a reduced incidence of intentional walks? Even if a star player hits no better with another good player behind him, doesn’t the fact that he gets more at-bats (and more at-bats in high-leverage situations) mean that protection provides some value?

B N
Guest
B N

Not necessarily. If you make a simulation environment and then do a train/test approach, at that point you’re mainly capturing the assumptions implicit in the data. Certain categories of naive algorithms in machine learning literally require zero assumptions by the programmer. This doesn’t mean the answers are right, but it means that they’re the result of the data being biased- not the programmer seeing what they wanted to see.

I personally don’t think that naive algorithms are the way to go with baseball, but one could employ some basic rule and physical constraints and then use data to train into them, without introducing much (if any) assumptions. A constrained Hidden Markov Model is a simple example of this sort of thing.