True story
A few days ago — on Memorial Day — I ended up stuck at the restaurant all day long (since we were getting murdered) and I showed up a few hours late for the barbecue at my parents’ house. Which was okay; the food was pretty cold by then, but everybody was still there, and the wife and I had a nice evening with my parents and the neighbours and friends.
In the middle of this evening, the subject of conversation changed to baseball. This was kind of a surprise to me, since, insofar as I was aware, I was the only person in attendance who gave any substantial portion of a rat’s anatomy about baseball, and I didn’t start it. But we’re talking about baseball nonetheless, and my father (out of nowhere) is all like, hey, fuck Jim Rice. And that didn’t sit well with me, as you might imagine.
My father’s argument about Jim Rice is that, hey, the guy may have put up good numbers, but fucking numbers don’t win baseball games, and he didn’t have the right stuff to hack it in real important situations, and that shit don’t show up in stats. Everybody else agreed.
Okay. I’m going to lay this out as plainly as I can. There are two points to deal with here, and I’ll take them in turn. First let’s talk about stats and what they do and don’t reflect. It’s important that we define, heading into this experiment, what matters in baseball, and I’ll define it like this: everything that matters in baseball is runs. Scoring runs and preventing runs are, ultimately, the only things that have any value, and, therefore, for any given event to have any significance whatsoever, it needs to modify the chance for runs to be scored. Can we agree on that? Nothing that has no impact on runs being scored is important in baseball? Good.
Now we get to the magic of statistics. Since we know how many runs are ultimately scored in any given game of baseball, and we have a very very large pool of data about games that have already been played, we are able to evaluate those runs and determine which game events had what degree of impact on them. And when I say "we," I mean "people smarter than me," because I’m quite frankly a complete moron who isn’t even qualified to evaluate the WPA I get for putting my pants on correctly at least 85% of the time.
Now, so far, this is nothing too shocking, right? Here’s the place where people get lost: if it’s the case that we know for a fact how many runs were scored in any given game, and it’s the case that, through statistical analysis, we can determine the relative value of game events, then it follows that there are exactly no events that cannot be measured statistically. As an example here, let’s make up an imaginary player — we’ll call him Derek Pope. Derek plays, let’s say… shortstop. As a shortstop, his primary contributions to the game would be his offense (both with the bat and on the basepaths), and his defense (both fielding and throwing). But wait, say some people, Derek has a magic talent in addition to these abilities! He has Calm Eyes and a Leadership Aura, and your fucking numbers can’t understand that.
Of course they can. If we see in the data — and remember, we’re looking at a very large set of data; we’ll say that (just throwing out a number here) Derek Pope has played in 2189 Major League games, which isn’t an unusual number for a guy whose career has lasted, oh, say, 16 seasons or so — an otherwise-unexplained increase in the number of runs Derek’s team scores (or a decrease in the number of runs his opponents’ teams score, if we allege his magic powers have defensive value), then we can attribute that to Derek. If we don’t see that sort of anomaly, then Derek’s magic powers clearly have no meaningful effect, since, as we’ve said, everything that matters in baseball impacts run scoring.
Now say we have another player. We’ll call this guy David Gritstein, and assume that he also plays shortstop. He has a magic power, too — giant brass balls full of hustle. If David has played, say, 1242 games, that’s enough of a data set that we can check and see if any anomalies show up in the run scoring. And by comparing the size of these anomalies to the size of Derek Pope’s, we can determine if Pope’s calm eyes aura is more powerful than Gritstein’s big balls or vice-versa.
Now. Don’t get the wrong idea. What I’m saying here is that everything can be measured statistically, not that everything has been measured statistically, or that existing stats are perfect and absolute. And if the only stats you know about are the dash line of AVG-HR-RBI, well, there’s a goddamn lot that isn’t reflected in the stats you have. But that doesn’t mean the information isn’t out there.
Now let’s talk about clutch. Clutch hitting certainly exists; there’s no question about that. Sometimes players come through when it counts, and sometimes they don’t. And, clearly, some players will do so more often than others. The question is: does this represent some actual repeatable ability on the part of those players, or does it represent essentially statistical noise? The Baseball Prospectus crew — people who are smarter than me — did a good deal of research on this subject, and Nate Silver writes about his findings in chapter 1-2 of the book Baseball Between the Numbers. What they found after a good deal of statistical prep work and a big plot of clutch data is that the coefficient of determination was .1. What this means is that the data shows that clutch hitting is only 10% repeatable phenomenon, and 90% plain, balls-out luck.
To summarise. There is absolutely nothing about a player’s character or makeup or class or spec or gear that cannot be reflected in statistics, and the data does not show large support for a "clutch" skill. I didn’t say this at the barbecue, mind you; I just let everybody talk about how it was obvious and proven that Jim Rice was not clutch because everyone remembers him striking out a few times, and how Adam Vinatieri is clearly the clutchest motherfucker who ever lived because, hey, remember that field goal in the Super Bowl?
Watch your sample sizes, guys. And control for confirmation bias while you’re at it, amirite?
