Another Look at Starting Rotations

My pair of articles on starting rotation slots last month found a lot of readers, a lot of criticism, and a lot of misunderstanding. My analysis (along with Chris Jaffe’s, published at about the same time) looked at past data and determined—post facto—how rotations performed.

What these articles did not do was retain the standard usage of starter numbers. For instance, in the way we normally use the term, the “ace” of the 2006 Brewers on Opening Day was Ben Sheets. By the end of the season, the top performer had been Chris Capuano. In an effort to describe what happened in groups of 32 starts, my approach had to combine the two; Sheets had the better ERA, but he only logged 17 starts.

That method has little predictive value, but dramatically illustrates the effect of injuries. To recall, here were the ERAs amassed by the average MLB team’s first through fifth groups of 32 starts:

#1      #2      #3      #4      #5
3.60    4.14    4.58    5.10    6.24

In many teams’ cases, it wasn’t that their “#5 starter” turned out to be dreadful. Instead, because one or more starters got hurt, they were forced to turn to a sixth, seventh, or worse option. This table illustrates the point, showing the number of teams who require any given number of starters.

Starters Teams
7        1
8        2
9        4
10       7
11       4
12       8
13+      4

The White Sox were the most fortunate (or prescient, or skillfully managed—take your pick), using only seven starters. The Royals, by painful contrast, needed 17. Most teams used between 10 and 12.

For the vast majority of clubs, this meant that the #1 through #5 labels assigned in Spring Training quickly lost meaning. Here’s another way of looking at it: this second table shows the number of teams that got any given number of starts from the five pitchers who started the most games:

Starts  Teams
<=100   1
101-110 2
111-120 7
121-130 6
131-140 8
141-150 4
150+    2

Even that presentation is somewhat misleading. By including the five pitchers who started the most games, it might not be considering the "starting five" for each team. A single rotator cuff tweak or popup-chasing collision can result in #6 getting 25 starts, much to the chagrin of millions of Aaron Sele-hating fans.

In short, there's a big difference between the way we commonly use terms like "#3 starter" and the way I tried to redefine it (the third-best 32 starts) in my previous articles. To create a model that has more predictive value, we need to get closer to general usage.

The Model

To do so requires deciding who the official starting five or six was for each team entering the season. That's necessarily approximate, especially in situations where a team had to start the campaign with a replacement (for someone like Sheets or Roger Clemens) but nonetheless planned on getting that preferable starter back.

In order to come up with some version of MLB rotations, circa March '06, I went through a variety of sources and established each team's #1 through #5 starters. I didn't rely on anecdotal evidence to assign numbers to each starter, just to decide who made up each team's first five. I tried to stick as closely to that evidence as I could, but I didn't count pitchers who were slated to be in a rotation but got hurt before they made more than a start or two, such as Zack Greinke and Shawn Estes.

To assign numbers to each starter, I took the top five and ranked them by 2006 ZiPS ERA. For instance, in the case of the Cleveland Indians, it was clear enough that they started the season with Paul Byrd, Jason Johnson, Cliff Lee, C.C. Sabathia, and Jake Westbrook. They sorted out as follows:

#  Starter         Z-ERA
1  Sabathia, CC    3.79
2  Byrd, Paul      3.90
3  Westbrook, Jake 4.02
4  Lee, Cliff      4.30
5  Johnson, Jason  4.59

This way, the #1 starter was determined by an objective predictor of skill, not a mishmash of opening day starting assignments.

The Sixth Man

The trickiest thing to determine using pre-season evidence is the #6 starter. In fact, it's not only difficult, but probably worthless. While each team may well enter the season with a sixth starter in mind, the situation could be very different just a few weeks later. To take one example: the Brewers initially planned on using Rick Helling first if another starter was needed. But due to injury and ineffectiveness, Milwaukee used several pitchers before Helling finally got a big-league start.

In short, I'm convinced there's no single right way to pick a sixth starter. I went about it—in most cases, anyway—by selecting the pitcher who started the most games but wasn't part of the initial starting five. This resulted in choosing guys such as Elizardo Ramirez, Jeremy Sowers, Zach Miner, and Cole Hamels. It passes the sniff test, so the league averages will give us a reasonable approximation of who teams called upon the most when needed.

Finally, the Numbers

Using this model, here is the average starter in each of the first six rotation spots:

Spot       ERA     GS      IP
#1 Starter 3.91    28.5    181.2
#2 Starter 4.61    26.6    165.7
#3 Starter 4.74    27.6    167.5
#4 Starter 4.82    20.8    123.4
#5 Starter 4.96    20.6    126.8
#6 Starter 4.77    15.5    95.1

The average rotation got 124 starts out of its front five and another 15 from its top replacement, leaving 22 more for the rotating back end that most teams employed. The biggest difference is between the #1 and #2 starters, suggesting why having two aces catapults a team so far ahead of the pack.

A Hardball Times Update
Goodbye for now.

Notably, the number of innings and starts takes a big drop after the #3 spot. That's not so much due to injury, I suspect, as ineffectiveness. (The two, of course, are not wholly unrelated.) Especially in the #5 position, pitchers are hardly guaranteed their spot for as long as their arm stays attached.

The Sixth Man, Again

The most striking thing in this data is that the average #6 starter is nearly as good as the average #3 starter. Upon reflection, though, it begins to make sense. As I suggested in discussing the difficulty of choosing a sixth starter, there's plenty of bias there.

If Hamels had been ineffective in Triple-A, he would've been passed over for someone else. If Miner had gotten shelled in his first few big-league outings, a la Hayden Penn, he would've gone right back to Toledo, and we'd be using Wilfredo Ledezma in this sample instead. Despite the selection issues, there's a lesson here: the teams with the best sixth starters were the ones with large pools to choose from.

Francisco Liriano had a freakishly good season for a fill-in, but the Twins could also haved turned to Matt Garza or Boof Bonser. Scott Olsen was another superb #6, but the Marlins had more good young pitching last year than the rest of the NL East combined. I don't mean to take anything away from Liriano or Olsen, but the Twins and Marlins set themselves up to get good performances from their #1 fill-in, even if they didn't know exactly who would end up starring in that role.

Predict Away

This set of numbers, unlike the group I produced in my previous rotation analyses, are ones you can compare to projections for specific players. Keep in mind, though, that these aggregate ERAs are for a pitcher in a neutral park against average opposition—not precisely the case in very many instances.

One important way in which these conclusions dovetail with those of my previous articles is just how much teams have to rely on starters beyond their first five. The average team gets nearly 25% of its starts from pitchers who don't start the year in the rotation. Taking proper precautions to avoid disaster in those 30-40 starts can be just as important as making marginal improvements in the middle of your rotation.


Comments are closed.