A different way to look at the performance of starting pitchers

When baseball analysts talk about pitchers they cite a number of familiar statistics: ERA, FIP, WHIP and ERA+, for example. These are all valuable metrics, but they don’t accurately portray the modern game. With the exception of WHIP, these stats are scaled to nine innings. This made sense when pitchers were often called upon to pitch entire games, but today pitchers seldom pitch the full nine frames. As a result, a statistic such as ERA, probably the statistic in baseball cited to describe the quality of a pitcher more frequently than anything else, doesn’t entirely capture a starting pitcher’s effectiveness.

Let’s take two hypothetical pitchers as an example. Pitcher A, the good pitcher, averages exactly six innings each start, and has an ERA of 3.00. If he starts 34 games in a season he’ll accumulate 204 total innings and figures to be somewhere in the Cy Young conversation (although he won’t win it). This is about the level Cole Hamels performed at in 2010. Pitcher B, the mediocre pitcher, also averages exactly six innings each start, except his ERA is 4.50. He too makes 34 starts, accumulating 204 innings, except he’s considered a back-of-the-rotation starter at best because of his high ERA. Joe Saunders and Edwin Jackson performed to this level in 2010.

Using the current metrics, Pitcher A is a potential star, while Pitcher B is Joe Saunders. If these two pitchers were expected to toss nine full innings every five days then their ERAs would accurately reflect their relative performance. At the end of nine innings one pitcher has allowed only three runs to score, while the other has allowed either four or five runs to score. This is a material difference. In 2010 the Nationals scored almost four runs per game exactly. Over nine innings Pitcher A sets them up to win all his starts while Pitcher B leaves them in trouble every time he takes the mound. One is accurately perceived as being more valuable than the other.

The problem with this methodology is that almost no team expects its pitchers to pitch the full nine innings in today’s game. Teams have expert closers, set-up men and specialists to maximize the possibility of a win on any given night. As a result, the game shortens for a starting pitcher. It doesn’t make sense not to do the same thing with ERA.

Returning to our two hypothetical starting pitchers, on any given night Pitcher A leaves after six innings, having allowed only two runs. Pitcher B also leaves after six innings, having allowed only three runs. Pitcher A is still better, but he isn’t quite as much better as his ERA would suggest. Using the example of the Nationals one more time, in 2010 they were expected to have scored 2.67 runs at the end of six innings. In this new scenario, Pitcher A remains better. He leaves the team with a lead every time, but Pitcher B is no longer dramatically off the mark. On any given night, he gives the Nationals an excellent chance to win the game. Granted, the team is at best tied when he leaves, but on average it will score 1.33 more runs. Pitcher B has kept them competitive, and in this world where the offensive always performs to its average the bullpen is suddenly responsible for protecting the win, or letting it slip away. As a result, Pitcher A is no longer dramatically better than Pitcher B, and the magnitude of the difference has more to do with the strength of the offense and the bullpen than Pitcher B.

Let’s apply this slightly altered approach to the 2010 numbers of the projected rotation for the 2011 Yankees:

The data in this chart have been adjusted. The IP are only for games that a pitcher started. The same is true of the ERA figure. Relief appearances have been removed (CC Sabathia and A.J. Burnett only started). IP/GS is the number of innings each pitcher averaged in any given start. I’m calling the new stat that measures the number of runs the pitcher has allowed on average at the time he leaves the game aERA, or Adjusted Earned Run Average.

For example, CC Sabathia is a beast because he averaged exactly seven innings a start and had typically allowed only two or three runs at the time that he left the game. The 2010 Yankees scored 5.3 runs per game. At the end of seven innings the team had usually scored more than 4 runs. In his typical start, CC therefore didn’t just give the Yankees a chance to win, he almost gave them no excuse to lose. He typically handed the team’s lead and the game over to the Yankees’ two best relievers.

If we look at those numbers and consider the strength of the 2011 bullpen, the bad pitchers actually get better if they are pitching for a team like the Yankees. Specifically, A.J. Burnett isn’t the black hole he’s supposed to be. He’s by no means good, but on average he actually left the game tied, having given up 3.31 runs over the same period of time that it usually took the Yankee offense to score 3.34 runs. I’d rather have CC, but the 2011 Yankees will have many excellent options to get the ball into the hands of Mariano Rivera, options that will frequently be stronger than the opposing team’s starter, or its bullpen.

The numbers for Ivan Nova and Sergio Mitre are a small sample size, but using this approach only one of these pitchers augurs to be a disaster, and that’s Mitre. While Nova will most often leave the game tied if these numbers carry over into 2011, Mitre will leave in the fifth inning, and the Yankees will be trailing. The offense will probably plate some more runs, but the real pressure falls on the bullpen that needs to account for 4.1 innings. On any given day that will call for every available arm, a luxury no team can afford on a regular basis.

Using this approach, Brian Cashman may actually be committing malpractice if he doesn’t sign a veteran arm. While it would be nice for the Yankees to find a pitcher of Cliff Lee‘s caliber, that’s not going to happen, but between the team’s strong offense and bullpen Cashman really only needs to add a pitcher of A.J.’s caliber to give the team a lead on any given night. He’s a bad pitcher, but take Kevin Millwood, for example. In 2010 his ERA was 5.10, accumulated over 190.2 innings. That translated into just about six innings per each of his 31 starts, during which time he allowed 3.4 runs. A pitcher of Millwood’s caliber may not help a team like the Nationals, but he would help the 2011 Yankees. Assuming the 2011 offense is about the same as the 2010 vintage, the team would theoretically have the lead each time this pitcher left the game, plus the bombers would be adding a starter who has proven consistently that he can accumulate innings in the big leagues.

I’m not advocating for Millwood. He just serves as an excellent example that there is no excuse for the Yankees to start the season with Mitre in the rotation, and every reason for a team with so many strengths to go out and get a pitcher who can be relied upon to give the team six innings every five days, even if that pitcher has an ERA of 5.00. Sure, that’s terrible over nine innings, but over six its only about three runs before the game is handed over to the bullpen, thus bringing down the team’s ERA over the last three innings tremendously. Andy Pettitte is, of course, the best option, but in a pinch Edwin Jackson would also do.

6 thoughts on “A different way to look at the performance of starting pitchers

  1. I like the idea of an adjusted ERA, but to make the number more meaningful, I think some kind of penalty factor needs to be applied for pitchers that don't pitch at least the league average number of innings per start; those pitchers kill a bullpen over the course of a season and the residual damage from their short starts can adversely affect a team's bullpen for 3-4 days. That warrants an additional penalty.I have no idea what that penalty should be or how it should be calculated, but I shudder at the idea of a guy like Mitre going into arbitration and saying "Hey, I should get $8 or $10 million because my aERA was only a little higher than CC's aERA." And you know something like that is going to end up happening if aERA becomes popular.

  2. The statistic is only relevant when paired with the average number of innings the pitcher makes per start. An aERA of 2.50 is awesome, if the pitcher lasts at least 7 innings. Its catastrophic if he lasts 2. The idea is to do a better job of presenting the situation a given starter leaves the rest of his team. On any given night that means the inning when he leaves, and the score. So, just saying that a player's aERA is X.XX is not complete with the innings pitched per games started (IP/GS). I didn't do a good enough job in the post linking those two statistics. So, returning to your Mitre example, I believe Cashman would respond as follows: "True, but you only lasted 4.2 innings per start. For an aERA of 3.08 that's terrible. Now get out of my office."

  3. Sorry, the second sentence in the second paragraph should read, "So, just saying that a player's aERA is X.XX is not complete without the innings pitched per games started (IP/GS)."

  4. Thanks for the additional information Mike. The linkage between IP/GS and aERA seems to address the issue I raised.Now, my only questions is this: how to we get Cashman to say your hypothetical quote to Mitre now, with one minor word change: replace the word "office" with the word "stadium."I can't believe I'm saying this, but I actually think Millwood could be a vast improvement over — uh, oh . . . excuse me, Mike, I had to run to the toilet. My stomach revolted over any sentence that included the words Millwood, Mitre and vast improvement.

  5. Hi Wayne, I hope your stomach has improved. You've hit the point, nail on the head. Although many of the starting pitchers being mentioned are terrible, on a team capable of providing the offensive and bullpen support that the Yankees have these pitchers become much stronger options. I don't like the idea of Kevin Millwood in pinstripes either, but he is, in fact, an improvement over Sergio Mitre.