The superlatives and ways of breaking down R.A. Dickey’s fantastic 2012 season to date have pretty much been worn out. Consecutive one-hitters, a 2.00 ERA, 2.72 FIP, 2.6 fWAR, 3.5 rWAR over 99 innings tends to make that happen.
Not even the “mediocre pitchers are going to start climbing Mount Kilimanjaro every off-season” jokes are really available any more. Add in the mystique and fun of Dickey being a knuckleballer (not to mention the seventh-grade ["grade seven' for my Canadian readers] humor value of his name) and we have the perfect meeting place for everything we enjoy about baseball on an aesthetic level.
Some will not want his tremendous start to the season to be somehow “ruined” by discussing something boring like regression. Well, sorry, but regression is a real thing for any player who is incredible for part of a season. What is interesting about Dickey’s case is trying to figure out which group of players he should be regressed toward.
Regression to the mean has, for me at least, been a concept that, while complex in some ways, has been a simple way to understand many things about baseball analysis. You can read the article linked above for more information (see also the good introductions here and here).
Rather than re-wording something for the nth time, I will simply quote my own brief explanation:
When we use the term “regression” in a sabermetric contexts, we do not necessarily mean “get worse.” We mean that a player’s true talent is probably closer to the average of the population to which he belongs. We use it as a tool to estimate what his “true talent” is as opposed to his observed performance.
For example, when a player goes 0-4, we do not think he is “really” a .000 hitter. The same player could go 4-4 in the next game, and no one would think he was a 1.000 hitter, or, combined with the previous game, a true talent .500 hitter. We know this because baseball fans have a basic, intuitive understanding of things like “sample size” and “random variation,” even if we do not use those words. We understand that players will sometimes play above or below their true talent for period of time.
Hopefully you get the idea.
So back to Dickey: while 99 innings is a meanginful sample for a pitcher, it is hardly definitive. Given how far above his observed performance is above league average (and his recent past), we expect him to regress to the mean, that is, we expect his future performance to be closer to average (in Dickey’s case, worse than he has been so far) of the population from which he is drawn than it has been so far this season. But that is what is interesting to me about Dickey when trying to estimate his performance for the rest of the season: what population does he come from?
The simple answer is “major league starting pitchers.” Perhaps that is the best we can do, and a simple projection systems like Marcel actually do relatively well compared to the bigger models by doing little more than that. However, more sophisticated systems might actually want to narrow populations for specific players down a bit. Some random examples for pitchers that might currently be used (or could be used in the future given data from Pitchf/x and Hitf/x) would be handedness, fastball velocity, age, height, weight, and so on. One could also use pitch type, of course.
The problem is that Dickey is unique among current Major League pitchers because, well, you know. That’s not to say that pitch type is the only thing one should consider when putting a pitcher into a population. Leaving aside the real possibility that Dickey may himself be very different than past knuckleballers, we know that knuckleballers as a whole are really different.
For example, while not all pitchers have the same BABIP talent, the overall talent spread for BABIP among them is about half that as among hitters, so for an overall population of starting pitchers (starters and relievers should be treated separately, of course), you regress BABIP pretty heavily to the mean. However, we also know, historically, that knuckleballers tend to have a BABIP much lower than “standard” pitchers, which is why knuckleballers often give DIPS-type metrics such fits.
Still, it is one thing to say that we can generally expect that Dickey’s BABIP will not regress towards .295 or whatever league-average is at the moment, and another to actually make a claim as to what we should project to regress.
One might be tempted to say that we should look at what other knuckleball specialists like Phil Niekro or Tim Wakefield have done in the past. It is clearly advisable to look at past knuckleballers to gain insight into Dickey now. The way knuckleballers have aged, for example, is also very different from other pitchers, so for projecting Dickey’s “career path” down the road, this is extremely valuable information.
That is still not quite what we are looking for, though. How Wakefield, Niekro, and the other knuckleballers aged in their late-30s is important for trying to figure out what the next few years might hold for Dickey, but that is a different question than the one I am asking in this post.
If we are trying to figure out how good Dickey is right now — trying to estimate his true talent as opposed to his 2012 performance or his future career trajectory — we need to regress him to a group of his peers. And while Wakefield and Niekro and the rest are also knuckelballers, they are not part of the current grouup of major league pitchers, which is the relevant group here. We cannot simply regress Dickey’s strikeout rate to Wakefield’s from 2004, for example — that does not make any sense. At the moment, Dickey is not part of the same population as Wakefield in the relevant sense.
So does this mean we simply cannot project Dickey’s current true talent? Does it mean that (Mets fans understandably hope) that he is just so darn unique that he probably will not regress at all? No and no.
Dickey is still part of the population of major league starting pitchers. He is very different from pretty much all of them, but he is not a species unto himself. We still have to take into account his past performance and, yes, regress his various components to the average of the population of major league pitchers. That may seem to be the boring answer, but I am not sure that it is trivial.
ZiPS’ Rest-of-Season projects Dickey to have a 3.53 ERA going forward. Frankly, given Dickey’s past and regression, that seems reasonable in many ways. ZiPS is a very good projection system. However, we should have less confidence in projections for Dickey given how he stands in relation to the population to which he is regressed.
Maybe that answer is boring, but to me, the uncertainty makes watching how Dickey pitches the rest of the season that much more exciting.