Sunday, July 11, 2010

Does a "year of the pitcher" mean more 20-game winners?

"Pitchers Rule," declares the cover of the July 5, 2010 issue of Sports Illustrated. The accompanying story quotes some statistics about how pitching in MLB is very much improved this year.

"The 1968 season will always be the Year of the Pitcher ... Bob Gibson boasted a 1.12 ERA for the Cardinals ... Danny McLain won 31 games for the world champion Tigers ...

"[In 2010] there's an across-the-board dip in offensive statistics: through Sunday an average of 8.9 runs per game had been scored, down from the average of 9.3 through the same date last season. ... Home runs are in similar decline ...

"At week's end 15 hurlers ... were on pace for 20-win seasons; in the last six seasons there have been 12 20-game winners combined."

Ignore the obvious problem that there will always be more pitchers "on pace" for a goal than actually reaching it (for instance, there are lots of players "on pace" for a 162-home run season after one game). There's another thing there that didn't make sense to me: the idea that when pitching is up and batting is down, there will be more 20-game-winners than otherwise.

It makes sense for ERA, but not for wins. No matter how good or bad a league's hitting is, there is always the same number of wins to go around -- one per game played, or, in total, 162 times half the number of teams in baseball. Why should those wins be concentrated among fewer pitchers (which is what happens when you have lots of 20-game winners), just because the hitters are worse?

One explanation: perhaps hitters are worse because there are a bunch of great new pitchers and those are your extra 20-game winners. But that's not what's happening this year, is it? Even veteran pitchers are seeing their stats improve this year, aren't they?

So, maybe it's just a mistake, and SI didn't realize what they were saying. But, just to be safe, I checked. For every season since 1982 (except 1994), I counted how many pitchers won 20 games, and calculated the league ERA. (Actually, I mean the *MLB* ERA, not the "league" ERA. But that sounds awkward. Is there a better way to say it?)

Then, I ran a regression. And, to my surprise, the correlation was significant (p=.04). The higher the ERA, the fewer 20-game winners. For every 0.37 increase in overall ERA, there was one fewer 20-game winner that season.

Leaving out 1995 made the result a bit less significant (p=.06) but kept the effect size about the same (+0.42 ERA gives one less 20-game winner).

So what's going on?

Here's one possibility: managers are failing to adjust for the higher offense, and are going to the bullpen too soon. In a low-scoring league, a pitcher having a bad day might give up four runs in the first five innings, but stay in the game. In a higher-scoring league, he might give up five runs, and be lifted. The manager doesn't quite adjust enough for the fact that the two performances are equivalent. As a result, some of the wins that would go to the starter wind up going to the bullpen instead.

There's a better alternative, one that doesn't have to assume managers are irrational. It could be pitch counts. Suppose a starter stays in for 100 pitches. In a high-offense league, a starter's 100 pitchers lead to five runs in four innings. In a lower-offense league, the 100 pitches yield five runs in five innings. The low-offense starter gets the win; the high-offense starter doesn't.

In any case, I don't know which of the two explanations is right. It could be a combination of both. Or can anyone think of other explanations?

Labels: ,


At Monday, July 12, 2010 9:29:00 AM, Anonymous Guy said...

The problem with using regression over those years is that you have big changes in pitcher usage which may not be related to the change in offense. Offense spiked in 1993-94, then remained high. But pitcher usage was already changing prior to that. In 1982 11 pitchers had 37 or more starts; in 1992 no pitcher had that many.
In 1982 there were 17 CGs per 100 starts, but it had dropped to about 10 a decade later.

Since then there has been some additional drop in the number of starts, and a small reduction in IP/start. CGs have also continued to fall, to about 3 per 100 now. But it's hard to know how much if any of that change resulted from the increase in offense, since the trends were already apparent before 193.

At Monday, July 12, 2010 10:22:00 AM, Blogger Phil Birnbaum said...

Geez, you're right. I forgot about complete games. I used 1982 because I assumed the four-man rotation was pretty much universal by then ... but, of course, pitchers were left in a lot longer in the olden days of the 80s.

If I do just 2000-2009, the result is not significant, but it has the right sign and about the same magnitude (0.38 in ERA per 20-game winner).

It's probably that the pitchers-left-in-longer is responsible for almost all the effect, but, as you say, WHY pitchers are being left in longer isn't easy to determine.

I'm guessing that it's pitch count a lot more than it is ERA.

At Monday, July 12, 2010 10:57:00 PM, Anonymous Ken said...

Could it be that the lower era = less pitches = more innings pitched = more decisions for pitchers? (Less time for the bullpen to blow it, bullpen less likely to blow it due to lower era)

At Wednesday, July 14, 2010 10:02:00 PM, Blogger Martin Monkman said...

Phil, have you tried a multiple regression with some of the other variables that have been suggested? With 20 game winners as the dependent variable, and "league" ERA along with things like average number of innings pitched, complete games per start, and number of innings per start, you might get a still stronger model.

At Friday, July 16, 2010 2:04:00 PM, Blogger Phil Birnbaum said...

Hi, Martin,

Yup, I did. CG/start is not significant.

If I do Wins/start and ERA, the model improves, although neither is now significant.


Post a Comment

<< Home