Some days, we all wake up on the wrong side the bed. It starts to be a problem when your bed is against the wall. Slumps are a part of baseball (and life), because humans are creatures of cycles. Sometimes you’re caught in a bad one.
Slumping is the flip side of the “hot hand” coin, and so it’s going to get second looks and long stares around these parts. We know that there are situations in which a hitter goes 0-for-12 over the course of a few days, but is that “slump” a reflection of him having a few days in which he's actually a worse hitter than he normally is or is it simply a case that sometimes you get a few tails in a row even on a 50/50 coin? Could it be that sometimes it’s a little from Column A and others it’s a little from Column B?
Here’s the problem mathematically: Let’s assume for a moment that there really are days-long periods when a player just doesn’t have it. If we were to somehow do a blood test for hitting ability on those days, we’d find that he’s a .167 hitter that day even if he’s a .300 hitter overall. The next day, he’s still reading as a .167 hitter on the blood test and he performs that way.
Well, there are going to be other days-long periods when there’s nothing wrong with him and that blood test would say “.300 hitter” but he just happens to go 2-for-12. The next day, because nothing’s wrong, we continue to expect him to be a .300 hitter. There are going to be enough of each that we aren’t going to be able to sniff out which of those days-long periods are “slumps” and which are random noise.
If a slump were to go on for two weeks, we could start saying—mathematically—that it’s unlikely someone who was a true-talent .300 hitter could hit that poorly by chance for two weeks. But how many cold streaks actually last that long and where a hitter goes that cold? So, perhaps the problem with streakiness is not that it doesn’t exist but that it’s uniquely designed to avoid detection. We do have some evidence that these local variations can have some meaning and that it can actually tell us something about what to expect from a player going forward, but slumping is a tough one to pin down.
So, that brings us to fielding. Is defense slump-proof? There’s a general belief that while players might have weeks in which they are cold as liquid hydrogen at the plate, their defensive prowess is going to remain fairly steady over time. (For instance, this was what Cubs fans told themselves all last year about Jason Heyward.) A Gold Glover is always going to have a gold glove on his hand, right? And so, for players who are seen as streaky or volatile (and therefore, kind of a high-variance bet), the comfort is that even in those two-week stretches when they can’t hit anything it’s not affecting their defense.
Warning! Gory Mathematical Details Ahead!
Let’s ask two separate, but related questions today. One is whether fielding shows any properties of “streakiness." The other is whether fielding results appear to be tied to any streakiness on offense.
Now, we just need a data set.
All of the cool fielding data sets are proprietary. When one wants to do analysis on fielding and one has no budget, this can be a bit of a problem. But soft, what light through yonder Windows 95 breaks? It is the east, and Retrosheet is the sun. It turns out that Retrosheet data, from 1993-1999, has pretty complete hit location data. It’s not amazing hit location data, mind you, but it’s something. Balls were (manually) coded as landing in, or in the case of ground balls, scooting through the infield at one of several points on a grid.
We have no idea how hard the ball was hit. We have no idea where the fielders started, although since it’s the 90s we don’t need to worry about shifts. Is that great? No. Is it free? Yes! Also, since we’re going to be using data from the Clinton Administration …
(It was either that or the chicka-cherry-cola song.)
We’re going to focus on ground balls specifically. To get an idea of how likely a ball is to be fielded, we need to know a couple of things. One is generally how difficult a ball is to field. We can figure that out by looking at where it was hit (in this case, which zone) and how often the league as a whole got to that ball. We can also look at a player’s “success” rate in fielding the balls hit in his area.
I started with shortstops, and made sure that a fielder had been out there for at least 250 grounders hit near to him. I coded each for whether he made the play or not and used the “difficulty” factor and his overall success rate as predictors. Then, I looked at his success rate on his last 10 grounders. I ran it as both a raw success rate and one in which I adjusted for difficulty. If the “last 10” grounders have predictive power beyond how tough the play was and how good the fielder is in general, then we have evidence that recent events make a difference in his abilities. Sounds like a good definition of “streakiness” to me.
It turns out that the answer is “yes” that players who have converted more of their last 10 chances into outs are more likely to get to the 11th chance. But here’s the catch: While the effect was significant, it was also very weak. The vast majority of variance in predicting whether a ball will be converted into an out is explained by where the ball is hit. If you want to improve a player’s defense, hit a bunch of balls directly at him.
There’s about a five-to-one ratio between the player’s overall success rate for the season and his “last 10” success rate. That means for the most part, whether a player would get to the ball was based on his overall talent. Still, there was room in there for some amount of streakiness. I re-ran everything with the last 20 grounders as my “streakiness” indicator and continued to get the same thing. The same basic findings held for second and third basemen. Yes, there’s some non-zero recency effects, but they are small.
Now, does slumping at the plate portend a slump in the field? This time, I used a player’s on-base percentage in the last 10 plate appearances he had coming into the game. (Again, I tried 20 and got the same findings.) I looked to see, after we controlled for where the ball was hit and the player’s overall fielding ability, whether he was more (or less) likely to get to a ball in a game when he came into it scuffling at the plate. The answer to that one was no.
I think this one deserves a bit of a disclaimer. This isn’t the greatest data set and I’m only looking at grounders, so it’s not the type of pristine methodology that powers bold statements about the state of the universe. But I think it’s interesting nonetheless. If we’re willing to say that slumping exists for batters, even though the evidence is fairly weak, then this evidence suggests that slumping can exist for fielders. The evidence also suggests that a struggling batter is not necessarily a struggling fielder, which means either the two halves of the game slump because of different reasons or that perhaps players can compartmentalize their slumpage.
Maybe it’s just that I don’t have great data at this point and the two are related. So, if you see a guy who just looks lost in the field the same way he sometimes looks lost at the plate, maybe he is … hopefully temporarily. But that means if we have streaky hitters, we might also have streaky fielders, and the idea that a streaky hitter with a good glove is at least a slightly more stable bet than we had originally thought isn’t actually true.