December 11, 2009

Checking the Numbers

On the Swing

by Eric Seidman

Printer-friendly

Contact Author

It isn't exactly a secret that the roles of pitchers in baseball have evolved at an almost exponential rate over the last several decades. As we know only too well, long gone are the days when complete games were the norm, leaving in its wake the progressive idea of specialization, pegging those taking signs from the catcher not as pitchers, but as starters or relievers, with further binning based on their perceived duties. Consider George Sherrill, who has been referred to as either a lefty specialist, a set-up man, or a closer, all during the course of a career that isn't even over yet.

Research previously conducted by Rany Jazayerli suggests that specialized relievers did not begin their ascent until the late 1950s, before increasing substantially in each subsequent decade. As can be expected when starters are no longer relied upon to finish what they start each time out, the percentage of complete games shares a strong negative correlation to the prevalence of relievers. Scrutinized workloads and heavily monitored pitch counts preclude many starters from going the distance these days, especially when the role specifications have afforded certain pitchers the opportunity to work in the big leagues that would not have been available in the 1930s. Back in the early days, Cy Young could take over for other Cleveland Spiders starters should they need help, saving games (if not recording saves). Why wouldn't he? He was, after all, one of the best pitchers in the league, and the team certainly stood a better shot of winning a given game with his services in use. Nowadays, the idea of Cliff Lee potentially pitching in relief should this past World Series have been extended to a seventh game comes off as taboo enough to merit widespread national coverage.

The idea of starters occasionally relieving or vice-versa has not been entirely eradicated, which is interesting in a few ways, not least the sense that it serves as one of the last remaining ties to what now feels like a prehistoric pitcher usage pattern. Pitchers used this way will make spot starts throughout the season while jogging in from the bullpen a decent number of times, moving back and forth between roles. This is sometimes referred to as "swinging" back and forth between the rotation and the bullpen, thus we have swingmen. Should a starter get injured, or in the event of a doubleheader, a team can avoid using another starter on short rest or hauling some temp up from the farm system by simply dipping into their relief corps and plucking out a hurler capable of logging a majority of that game's frames. Swingmen tend to work as long relievers or mop-up men in the bullpen. They're the pitchers called on to eat some innings in blowouts in order to save other arms for tighter games. The current culture's obsession with categorizing and binning everything both quantitative and qualitative seems to have led to the increased prominence of the term "swingman," but there is no set definition for pitchers of this ilk.

Some pitchers swing within a season as part of a ballclub's specific plan for them. Take Max Scherzer upon first being called up, or Derek Holland last year-their teams strive to gradually increase the workloads of young pitchers by immediately installing them in the bullpen, all the while planning on eventually bumping them into the rotation. Afterwards, should the need arise, the young guns can be used in the bullpen. Chad Billingsley is another recent example of this, as he made 16 starts in his rookie season back in 2006, but then made 20 starts out of a total of 43 appearances the very next year; it wasn't so much that the Dodgers required his services in the bullpen, but rather that the team wanted to monitor his workload while simultaneously retaining his presence in the big leagues, deciding the best way to accomplish these dual goals involved an equal amount of starts as relief appearances.

In other situations, teams will employ someone like Mark Hendrickson, who can fit into a bullpen role as a second-rank lefty specialist and middle reliever, but who can also make spot starts if his team is in a pinch and does not want to either sign a free agent or recall a minor leaguer. Either way, there is no set definition for, say, classifying swingmen in a database. Though the swingman is not necessarily a precise role, Hendrickson's utility in this area is basically the only quality he provides to a team, and is likely the reason he continues to garner employment.

Jazayerli defined swingmen as pitchers with at least 25 starts and at least 10 relief outings in a season, finding that the role blossomed from 1910-1940 before drastically petering off after that. From 1974-2009-when Retrosheet data is complete and not missing any games-under 60 results surfaced for a query matching the aforementioned criteria. Suffice to say, for a meaningful look at swingmen, that definition requires some modification. For now, we will refer to swingmen as pitchers with at least 10 starts and at least 10 relief outings in the same season. Over the same 36-year span, this definition produced 897 pitcher seasons from 598 unique pitchers, substantially greater than the 55 or so names that were otherwise reported. Their prevalence has decreased the last few decades even with this definition, as evident by the chart below:

The sharp downward spike to the left occurred in the anomalous 1981 season, the year of the in-season strike. What I find odd here is the lack of an upward trend lately; while it makes sense to be less frequent than decades past, the early-season scheduling and its ample supply of days off feels like it should result in a few more swingmen; with so many days off in April, teams can get away with a four-man rotation and use their fifth starter as a swingman. The next logical question, given the history lesson and shift in usage patterns, asks why this matters, and why swingmen are important to evaluate, the answer to which deals with understanding what to expect when pitchers shift roles. Swingmen can also help us determine how a reliever would perform as a starter with the inverse true as well.

Conventional sabermetrics wisdom suggests that starting pitchers can lop over a run off of their ERA with a sustained trip to the bullpen. Why? One suggested reason for the improved performance is summed up in a published exchange between myself and Andrew Gelman, a statistics professor at the University of Columbia. Professor Gelman called to question the "pinch-hitter theory" which helped to explain how one of his editors, trained to look for mistakes, proofread and offer suggestions, could miss errors that he himself caught upon further review. Gelman isn't a trained editor, after all, and certainly isn't as talented as the employee with that job title, but there had to be something that explains why he was able to pick up on mistakes the editor missed.

I chimed in, suggesting that the theory be renamed to reflect the differences between starters and relievers, since starters are more talented pitchers than relievers-otherwise the relievers would be starters (yes, there are exceptions, but let's stay focused here)-but as the starters circle through the lineup multiple times and fatigue sets in, relievers almost always become the better option. The comparison isn't between a starter and a reliever at 100 percent, but rather the former at around 55-60 percent and the latter at 100 percent. Overall, the starter is perhaps the better pitcher, but at this theoretical moment, the reliever is a better bet. Relating to the editor metaphor here, the starter has been editing the book or article all day long, and while he is more talented in this position than the writer, the writer enters the foray with very fresh eyes.

With this in mind, do the numbers reflect the theory? And if so, by how much? Sifting through the aggregate data for the entire sample of pitchers and pitcher seasons, and focusing on a few telling statistics, here are the splits:

           As       As
Overall  Starter  Reliever
FRA       5.07     4.39
ERA       4.61     3.83
K/PA      .131     .154
UBB/PA    .080     .082

The data suggests that, yes, pitchers perform better as relievers in the run prevention departments, with an ERA 0.78 runs lower and an FRA-which is the RA for a pitcher but with bequeathed baserunners factored in to score at a league-average rate-of 0.68 runs lower. Pitchers also strike more hitters out as relievers, by over two percentage points, which certainly jives with the idea of being able to rear back and let loose as opposed to staying true to a pace. Quite curious, however, are the approximately identical unintentional walk rates, as the numbers indicate that pitchers capable of missing more bats with perhaps increased velocity and movement do not boast improved control. Overall, the effect of shifting from the rotation to the 'pen is two-thirds to three-fourths of a run as opposed to one or more runs. Sure, certain pitchers may be vastly better as relievers, but the aggregate delta is not as drastic.

Not every pitcher is created equally, however, with certain pitchers boasting a greater propensity for striking hitters out, and others earning their keep by keeping the ball on the ground, for instance. What happens if we segregate the group with the categories of power, finesse, or neutral? Perhaps different trends will emerge, with certain pitchers performing better in a specific area than others. Baseball-Reference.com defines power pitchers as striking out and walking greater than 28 percent of their batters faced; finesse pitchers as the same sum amounting to less than 24 percent of their batters faced; and neutral pitchers comprise everything in between. Their splits are based off of all walks, so a very minor adjustment needs to be made to incorporate our usage here of solely unintentional walks. Literally just subtract one percentage point off of the floors, and we're on the same page. I am also going to classify pitchers based on the year in question, not B-Ref's process of calling on the three years before and after the target, since the classifications are rolling; someone might be a power pitcher in 2004, and a finesse pitcher in 2005, that the pre- and post-test will miss.

First, the 897 seasons broke down as follows: 609 finesse, 189 neutral, and 99 power. The table below breaks down their data similarly to the overall group:

          Finesse                Neutral                Power
       as SP  as RP           as SP  as RP           as SP  as RP
FRA     5.09   4.33    FRA     5.11   4.58    FRA     4.86   4.34
ERA     4.62   3.76    ERA     4.67   3.99    ERA     4.37   3.99
K/PA    .115   .139    K/PA    .154   .173    K/PA    .192   .214
UBB/PA  .072   .075    UBB/PA  .093   .093    UBB/PA  .108   .105

The first aspect of this data that sticks out to me is that the strikeout and walk rates all follow the same trend: for whiffs, they increase when starters become relievers, and they stay practically identical as far as free passes are concerns. The deltas between ERA and FRA also prove interesting-Finesse guys are at 0.76 for FRA and 0.86 for ERA, Neutral at 0.53 for FRA and 0.68 for ERA, and Power clock in at 0.52 for FRA and 0.38 for ERA. Finesse pitchers clearly benefit the most from the shift. That they comprise such a large percentage of the greater whole likely suggests some semblance of selective sampling, as they may be on the cusp of maximizing their abilities and being able to focus on a lineup one time through could get them over the hump. This does not suggest that moving someone like Jamie Moyer to the bullpen will gift him with an extra seven mph or a dramatic increase in effectiveness, but rather that the the finesse guy's guile works very well one time through the order.

Another potential bias in the data deals with relievers specializing; someone like Hendrickson may be used solely as a lefty specialist when in the bullpen, which could help beautify his pitching line in comparison to facing a mix of hitters multiple times when he's in the rotation. That power pitchers experience the least amount of improvement with role changes is not necessarily surprising, as they are likely already rearing back and letting it rip as starters. Put together, power pitchers are far less likely to be used as swingmen to begin with, and when they are, the deltas in run prevention do not improve as substantially as their finesse or neutral counterparts. Along the lines of improvement, another fascinating tidbit under this lens involves swingmen that change classifications upon moving to the bullpen.

Of the 897 pitcher seasons in the sample, 515 featured the pitcher retaining his starting classification as a reliever. That leaves 382 remaining pitcher seasons wherein a shape shifting in terms of performance between the two roles occurred, which broke down as follows:

Starter    Relief
Category   Category   #
Finesse    Neutral     137
Finesse    Power        91
Neutral    Finesse      44
Neutral    Power        83
Power      Neutral      15
Power      Finesse      12

Relative to the totals of each group, 37.4 percent of finesse pitchers changed, as did 67.2 percent of neutral pitchers, and just 27.2 percent of power pitchers. Passing the sniff test is how the neutral pitchers are more likely to become power pitchers when in the bullpen than to "downgrade" to the finesse category. The aggregate group data portends improvement in strikeouts with sustained walks, which in theory should lead to a change in classification. Scarce are the situations when a power pitcher becomes a finesse guy. To make a bit more sense of the classification shifts, we need to check out the strikeout and unintentional walk rates of the pitchers that changed; if the cutoff for being a finesse pitcher is 23 percent, and Jorge Sosa was at 22.5 percent as a starter but 23.5 percent as a reliever, the effect is nowhere near as significant as if he were to jump to 26.5 percent in the bullpen.

The finesse pitchers that retained their classification posted a .108 K/PA and a .071 UBB/PA as starters, and a .119 and .065, respectively as relievers. Their colleagues that jumped one notch to the neutral threshold chimed in with marks of .123 and .073 as starters, and .159 and .087 as relievers; when venturing all the way up to the power plateau, they produced marks of .129 and .076 as starters and .199 and .100 as relievers. Essentially, the group-shifting classifications as relievers were much closer as starters to the neutral category than those remaining stagnant, with very little difference between their strikeout and walk rates before changing categories as relievers. All told, this detracts a bit from the significance of the classification changes, as pitchers closer to the cutoff point should experience an uptick in strikeouts when in a situation enabling them to completely let loose without having to pace themselves. Plus, it isn't as if they skyrocketed to the top of the neutral category, rather safely settling slightly above the cutoff. Another line of research in this regard would involve further binning the classifications based on status within a category, something for us to look at the next time out.

There is much more to investigate with swingmen, and this is but a mere scratching of the surface. Next up, aside from the more granular bins, a look at swingmen under the PITCHf/x lens could be very telling, as could the incorporation of balls in play rate. Perhaps, regardless of the finesse/power/neutral breakdowns, the classifications based on fly-balling or ground-balling tendencies could prove more significant. So far, however, we can safely conclude that pitchers perform better as relievers than starters by around three-quarters of a run per nine innings, though the extents to which they perform better differ based on the various types of pitchers.

Eric Seidman is an author of Baseball Prospectus. 
Click here to see Eric's other articles. You can contact Eric by clicking here