In last week’s Lies, Damned Lies, I reviewed Adam Dunn‘s major league career one plate appearance at a time, in order to determine how his performance changed when facing the same pitcher multiple times. For those of you who, like me, did some damage to your short-term memory over the long weekend, the idea was to discover whether, per Michael Lewis’ discussion in Moneyball, Dunn is a hitter with a hole in his swing that gets continually more exploited in repeated trials.

In Dunn’s case, the answer was a tentative “no”, but a lot of people mailed me to ask that I broaden the scope of the analysis. As D.H. writes:

“I like your research, but my problem is that you’ve presented no baseline. It
reminded me of a STATS Baseball Scoreboard article on whether Greg Maddux did
better the more times he faced a particular batter because he’s so “smart.” The
data showed that the hitters improved as time went on. But, like in your study,
there was no baseline to compare against. Adam Dunn may show a drop-off the
more he faces a particular pitcher, but maybe all players exhibit identical
drops. Or, maybe all players exhibit more precipitous drops, and only the good
ones (like Dunn) stick around because they only lose 20% of their value.”

In other words, is there any systematic advantage to the pitcher or the hitter given repeated trials? Doesn’t seem likely, I wrote back, not if the league is going to remain at some kind of equilibrium for very long. But D.H. is correct that it’s a question that deserves further study, much like why on Earth I didn’t wear sunscreen to the ballgame on Sunday.

As I mentioned in the Dunn piece, there is publicly available play-by-play data for each season from 2000-2002. In order to make sure that the players we’re working with formed a closed system, I limited the analysis to players who made their major league debuts in 2000 or later. It was then possible to look at all possible ‘pairings’ of the batters and pitchers within this group–what happens when Billy Batter faces Pete Pitcher for the first time? For the fifth time? For the 20th time, after Bill Batter has dropped the -y from his name and grown a mustache, and Pete Pitcher is discovered to be three years older than listed and actually named Pedro Pichardo?

As a starting point, it’s necessary to evaluate the batters and pitchers involved in the pairings to see how their quality compares to that of the league as a whole. Since players who have recently made their major league debuts are almost always young, we’d expect at least some distortions in their performance as a group.

In last week’s piece, I introduced a method for adjusting for the strength of the pitchers that Dunn faced by evaluating their Batting Runs allowed per PA (an old Pete Palmer favorite). It’s a straightforward way to generalize this approach for all the hitters in our sample, and to apply it to the pitcher side of our pairings as well.

It wouldn’t be a Lies, Damned Lies column if not for a graph or two, right?

Though it’s hard to discern at first, there’s some interesting information contained in that chart. As we’d expect based on the natural tendency of baseball’s competitive architecture to select out the better players, the quality of both groups gets stronger as they stay in the league and face one another more often–the hitters produce progressively more runs per plate appearance, and the pitchers allow progressively fewer.

However, the comparison is not perfectly symmetrical. As a group, the pitchers who had made their major league debuts within the past three years were a little bit worse than those who had been around for longer. But the batters who had made their debuts in the past three years were considerably worse than their veteran counterparts. That makes sense, if you think about it: The performance of batters is more age-dependent than the performance of pitchers, and so the subset of younger batters that form our group of debutants suffers more relative to its competition. These results also conform with the notion that batters tend to have longer careers.

When combining the results of the two groups–e.g. the yellow line–it’s also apparent that batters take a little bit longer than pitchers do to get up to speed. Well, maybe I’m distorting the meaning here a little bit; the batters who receive very few plate appearances and subsequently drop out of the paired trails are relatively worse than the pitchers that do the same. Though it goes somewhat against the conventional wisdom, what I think that means is that there’s more to differentiate batters on the lower rungs of the talent pyramid than there is for pitchers. But that’s really a column of its own.

How then do the actual paired matchups compare to expectations? Here’s the raw data.

                                                  Batting Runs/PA
Trial#   PA      BA      OBP     SLG     OPS     Expected    Actual
1       3908    .234    .304    .363    .667       +.012     -.008
2       2442    .266    .330    .426    .756       +.015     +.021
3       1579    .277    .340    .438    .778       +.017     +.028
4       834     .279    .355    .429    .784       +.018     +.032
5       582     .235    .306    .371    .677       +.019     -.004
6       445     .224    .268    .334    .602       +.020     -.033
7       273     .282    .355    .445    .800       +.022     +.037
8       196     .233    .260    .354    .615       +.022     -.031
9       146     .195    .267    .308    .575       +.021     -.038
10      111     .238    .279    .448    .727       +.019     +.005
11      80      .303    .338    .408    .745       +.018     +.017
12      67      .333    .364    .476    .840       +.017     +.047
13      56      .264    .304    .377    .681       +.017     -.006

When mapped out, the results bear a striking resemblance to a Robert Downey Jr. cardiogram, bouncing back and forth wildly, but never really settling into a discernable rhythm.

Unless someone can explain to me why, for example, the batter seems to have the edge in the seventh trial against a given pitcher, but the pitcher has the edge in the sixth and the eighth, I’m going to assume that the fluctuations are random.

The one potential exception is the very first time that a batter and pitcher meet. Though the difference is small, amounting to 13 runs less than expected when extrapolated over 650 plate appearances, the pitcher does seem to have something of an edge, holding the opposing hitter to a .667 OPS. Keep in mind that this is the most robust part of our dataset, containing nearly 4000 observations. Even so, I’m not sure that the result has much of anything to do with game theory. Rather, the first time that a batter faces a pitcher is likely to occur early in the game, when the pitcher is still at his freshest, and tips the advantage in his favor.

So it’s pretty safe to say that there isn’t any systematic advantage with respect to repeated trials for either batters or pitchers. Does that mean that individual batters and pitchers don’t exhibit any such patterns? Well, not necessarily.

I ran a regression analysis for each pitcher that debuted in 2000 or later and had faced at least 250 opposing batters through 2002. Similarly, I ran a regression analysis for each batter that debuted in 2000 or later and had at least 250 plate appearances. The idea was to determine whether there were any players who exhibited a statistically significant relationship between number of trials and productivity, as measured by Batting Runs.

There wasn’t much going on with the batter data. Out of a sample of 69 qualifying batters, there were five–Bobby Kielty, Brian Schneider, Andy Tracy, Chris Truby, and Ramon Vazquez–who displayed a statistically significant improvement in their performance in repeated trials. Unless I’m missing something, there isn’t much of anything in common between those players, and so I’m inclined to that this is strictly a random phenomenon. If we set the confidence level for statistical significance at 95%, as I have here, there is still going be that lucky 5% of monkeys who bang out a verse or two of Shakespeare. Well, maybe not, but you know what I mean. Interestingly, there weren’t any batters who displayed a statistically significant decrease in their performance over repeated trials–if those guys exist, I suspect they don’t survive in the big leagues for even 250 PAs.

On the pitcher side, the results were more intriguing. Out of a group of 123, there were no pitchers who systematically improved in repeated trials, but 12 who systematically declined. They were:

1       Sean Douglass
2       Adam Eaton
3       Travis Harper
4       Jason Jennings
5       Joe Kennedy
6       Jason Marquis
7       Jose Santiago
8       Gene Stechschulte
9       Brian Tollberg
10      Josh Towers
11      Dave Williams
12      Matt Wise

The ratio here–about one pitcher in 10–is hardly of the beyond-a-reasonable-doubt variety, but there’s some solid circumstantial evidence that we might be on to something. That is, the pitchers listed here have a lot of commonalities between them (no, not just that most of them suck). Here are a few snippets from the ESPN / STATS Inc. scouting reports on them:

  • Travis Harper: “Harper doesn’t dominate with any one pitch, but rather is successful by using his full repertoire…low-90s fastball, curve, changeup, slider…hitting his spots and changing speeds.”

  • Jose Santiago: “An extreme groundball pitcher without a strikeout pitch, he’s averaged fewer than 4.5 whiffs per nine innings for his career.”

  • Brian Tollberg: “Because he lacks exciting stuff, Tollberg has been overlooked his entire career. He throws a fastball that rarely reaches 90 MPH and mixes in a nice curve and changeup.”

  • Dave Williams: “Williams has a lot of moxie for a young pitcher, enabling him to get hitters out despite not having overpowering stuff. Although his fastball rarely touches 90 MPH, it looks quicker because of his deceptive motion. He also throws a decent changeup, and his big breaking curveball is effective…when he keeps it in the strike zone.”

  • Joe Kennedy: “Kennedy…mixes two- and four-seam fastballs that regularly are clocked in the low 90-MPH range and seem to sneak in on hitters, a big-breaking curveball and a consistent changeup. …He also was working on a slider.”

  • Sean Douglass: “Douglass has to mix pitches successfully because he’s yet to develop one dominating pitch. His fastball has been clocked in the low 90s, but it lacks enough movement to be an out pitch.”

  • Gene Stechschulte: “A successful closer in the minors, the jury is out on whether Stechschulte has the out pitch to be employed in the same role in the big leagues.”

And so forth. None of these pitchers has a single, dominating pitch. None of them except Marquis throw particularly hard. All of them except Santiago consistently mix in at least three pitches. They rely, to borrow the euphemism from the Dave Williams comment, mostly on “moxie.” And the more often they’ve faced a particular hitter, the worse off they’ve been.

It’s easy to construct an argument for why this should be the case. Changing speeds, location, and delivery will likely be a successful strategy the first couple of times that a hitter sees you–he hasn’t seen your full repertoire, allowing you to get the most out of your bag of tricks. But after the entirety of your arsenal has been exposed, you’re vulnerable without an out pitch, especially if you lack Greg Maddux‘s Vegas-bred sensibility for avoiding recognizable patterns when mixing up your stuff. The surest route to sustained success is to have a pitch that a hitter will have difficulty with even if he knows it’s coming.

Makes enough sense, right? What’s interesting, though, is that this line of reasoning runs counter to at least one long-standing bit of conventional wisdom. Namely, you’ll hear all the time that a pitcher is best suited for the bullpen because he doesn’t have enough good pitches to be a successful starter, the idea being that you can fool a batter once with stuff alone, but he’s sure to catch up to it the second and third times through the lineup. This is a maxim that has affected the career paths of literally thousands of young men. And yet, like so many other pieces of conventional wisdom, it has never been proven. In fact, I’ve presented some very, very scant, very, very preliminary evidence that the opposite might be the case–given the choice between a pitcher with several average pitches, and a pitcher with one or two dominant pitches, it’s the latter that may hold up better in repeated trials.

I suspect that this is a question that hasn’t been researched very thoroughly because it involves the use of two types of information–play-by-play data and detailed, reliable data on pitch types–that are notoriously difficult to come by. But man, if I was trying to figure out how make optimal use of my stable of million-dollar arms, it’s something I’d sure as hell want to take a closer look at.