There will be a very short planned maintenance outage of the site tonight (7/22) at 11 PM ET
April 4, 2013
In A Pickle
Can You Buy What You Can't See?
Baseball knowledge expands rapidly, inside the organized professional realm and out. We know things about outfield defense and batted balls and catcher pitch-receiving and pitcher skill and the best way to score a run that we did not know 10, 20, 50 years ago. There is also plenty we do not know, sometimes particular to baseball and sometimes dealing with general human knowledge as applied to baseball. (Think about questions of psychology, for instance.) The question, or one of the questions, if you're in a front office, is how these areas of knowledge intersect with your willingness to pay D dollars for player P.
By which I mean this: we have a solid handle on what offensive skills are worth and can probably, by combining stats with scouting, do a reasonable job of projecting a hitter into a particular situation and understanding which hitters are more valuable than other hitters as hitters. Given a player and his statistical line and his mix of physical skills as relayed by trained professionals, a team can put a number on his offensive value. The error range around that number might be unfortunately large, especially compared to the relative certainty that other industries deal in, but there is a number nonetheless. Similarly, teams can put a number on a pitcher, though perhaps the error bars are larger. Defense is harder, but using scouts and the advanced ball-tracking data that teams have at their disposal, they can figure out how many balls Larry the Left Fielder is likely to catch vs. Lonnie the Left Fielder. They can put a number on that.
But what about the less easily quantifiable aspects of the game? What about team chemistry? Let's say a team hasn't followed Russell Carleton's advice and modeled the networks in the clubhouse. They're choosing, say, between Jonny Gomes and Ichiro Suzuki, players who will cost a similar amount of money and (let's just suppose) are valued very similarly in terms of projected offensive and defensive performance. (And, in Ichiro's case, in terms of projected pitching performance for that one blowout you're going to throw him into in July.) Should the team, because it can't put a very precise value on Gomes's clubhouse presence versus that of Ichiro, simply ignore that variable?
It isn't, after all, like we're chasing unicorns. We're very sure that unicorns do not exist. The effect of personal interactions and friendships on our ability to perform is, by contrast, both within the lived experience of many of us and a topic of research. And even if that effect is tiny, a team as a whole deals with such scale over the course of a season or multiple seasons that the tiny effect can easily add up. Read Russell:
If a team is willing to pay some large sum to win one more game than they would if they did not spend that large sum (and we don't need to get into dollars-per-WAR calculations to understand that teams of course do spend large sums on such things—Mike Pelfrey is making $4 million), then a team that is not willing to gamble on a player's ability to add to the bottom line by methods not measured in the box score or the FIELDf/x database is missing out.
"Gamble" might be the key word. If someone gives you a 10 percent chance to win $101, then you should be willing to bet $10 on that proposition. So suppose one extra win is worth $6,000,001 to a team. Suppose further that a team doesn't fully buy that a single player can be worth 10 wins based on chemistry and leadership, but thinks there is actually a 1 percent chance that the effect could be that big. That's 0.1 wins, and that means the team should be willing to go $600,000 more on its offer than it would if all players were robots. That's not much in the scheme of things, but it's not nothing (and it intentionally puts to the side the rest of the range of outcomes—what if there's a 50 percent chance that Gomes can add a mere half-win to the roster via his Magic Mohawk? That's .25 wins, or $1,500,000).
What about clutch ability? It would be highly unreasonable to think that humans do not have different abilities to perform in high-pressure situations. It would be less highly unreasonably but still, it seems to me, unreasonable to think that major-league baseball players all have the same ability to perform in high-pressure situations. I don't think anyone actually thinks this, though. The stathead orthodoxy appears to be along the lines that clutch ability at the major-league level exists in such a narrow range and/or is so hard to measure that teams paying for supposed clutch ability are more likely paying for mirages born of bleeps and bloops that happened to fall in for one player but not another.
But if teams can get some handle on who actually is likely to be at the top end of that narrow range? Then, by similar reasoning as above, it might be worth paying a little extra for those players even if they're not certain that they're right about the size of the effect. If they can find a way to make some reasonable guesses (and here, the reasonable guesses are a lot easier than in the chemistry scenario, since we're essentially talking about an individual player increasing performance in a small number of high-leverage situations) and if they can through psychological testing combined with some observation (of the scouting and statistical variety) put a flag on some players who they think fit the clutch profile, then there could be an advantage to be had.
There are caveats, many of which I'm probably not thinking of, but here's one I have thought of: teams have to be able to identify players who actually are good at these intangible or semi-intangible skills, by which I mean that if you're throwing money at players who have a symmetrical risk centered around zero in whatever skill we're talking about (e.g. the likelihood of adding X wins via clutchiness is the same as the likelihood of subtracting X wins via lack of clutchiness), then in the long run, you're losing money. The point here is just to say that if you think you have a good handle on which players are adding chemistry or which players really do have more clutch ability or some other intangible skill I'm missing, the fact that you can't peg those skills to numbers in the same way that you can hitting or defense doesn't mean you shouldn't pay for them.
And here's the other thing: Teams probably are paying for clutchiness and chemistry and leadership and makeup and grit, so there may not be much gain in spending time and effort and money in identifying the players who have those skills in semi-rigorous ways if you think that these elements are already being built in to evaluations because, basically, it's baseball, and scouts and front office types and coaches have been looking for and prizing those skills forever and ever. Even if you think that rigorous identification of intangible skills can do a better job than classic nonrigorous identification, maybe that nonrigorous identification, which is essentially free in that your organization is already doing it simply by being a baseball organization, is good enough. Maybe the marginal gain simply isn't worth the expense.
The flip side of this identification issue, though, is that it's possible that the existing modes of identifying clutch players are, to the extent they're missing, not just missing randomly but missing in biased fashion. Maybe players who appear calm under pressure and thus who gain a reputation actually perform worse. (I have no reason to think this. It's just an example of how large segments of the industry could be doing it wrong.) In that case, it could in fact be worth putting money into developing techniques to find these players.
Alternatively, it could be worth putting money into untraining systems and pushing valuations to a place where clutch or chemistry or what have you is expressly not considered. Note that even this situation fits with the thesis overall: the scale on which baseball operates makes it valuable to explore the ways wins are created from as many angles as possible, to bring rigor to the field beyond the areas in which sabermetricians have typically operated. If that rigor results in understanding that one's own processes cannot, given the current state of technology and knowledge and training, make a valid and unbiased assessment of a given aspect of the game, and if that understanding results in dumping that aspect from your models, then you will, overall, be better off, particularly if you can find a way to reallocate resources to other areas.
Putting aside the specific area of study, I don't think anything above requires teams to change the way they understand how to bid on baseball players. They're already projecting players based on a range of possibilities, acknowledging uncertainty due to skill changes, luck, injuries, and the looming specter of total ecological collapse. Adding chemistry or clutch ability or anything else into the equation may be more difficult and it may be more uncertain and it may have less effect on any given game than simple pitching, hitting, and fielding, but declining to pay for it at all because of those factors could be a significant mistake.