It’s Time To Be Smarter About Bullpen Usage

When I saw that the White Sox had signed Zach Duke for three years and $15 million, my first thought had nothing to do with Duke or Chicago. It was, “wow, Andrew Miller and David Robertson are going to get paid.” Though Duke was obviously a much different pitcher in 2014 than he was in previous years, he’s still a guy who has one good year on his resume after nearly a decade of mediocrity. If he gets three years, it seems clear that the more accomplished Miller and Robertson are going to get at least four (though Robertson’s qualifying offer will hurt, somewhat).

This isn’t necessarily about whether they will or should get multiple-year deals; they clearly will, even though Dave has written here numerous times over the years about how poorly long-term deals for relievers tend to work out. Part of the reason they’re going to get paid is because the Royals just made it to the World Series based in no small part on having a shutdown bullpen of Kelvin Herrera, Wade Davis, and Greg Holland, and success will always breed copycats. Part of it is because of the perception that with starters going shorter and shorter into games, it’s more important to have a talented bullpen.

But is that second part actually true?

Check out the chart below, which shows the percentage of innings pitched by relievers since divisional play began in 1969. As you might expect, there’s been a general upward trend, especially since the low-water point of 1972 (22.92%), a season in which 25 different starters went at least 250 innings (!), four topped 300 (!!), and Wilbur Wood put up 376.2 (!!!) Though it hasn’t always been a smooth trend — 1988 was an outlier dip — it’s hardly surprising that relievers throw more innings now than they did decades ago. (Or, if you prefer to think about it the other way, starters throw fewer innings.)

That’s not news; it’s been well-chronicled over the years. If I’d pushed this chart back earlier than 1969, the upward trend would have been even more stark. (To pick two random years, relievers threw 17.9% of innings in Jackie Robinson’s first year of 1947, and only 8.3% of innings 30 years earlier in 1917.)

But what is interesting is that contrary to the narratives, this isn’t a trend that’s still increasing. For most of the last two decades, things haven’t changed that much around some yearly ups and downs. In 2014, relievers threw 14,620 innings, or 33.52% of the total. In 1995, relievers threw 12,142.2 innings, or 33.65% of the total. (1995 was of course a strike-shortened season, which is why a similar percentage comes out to more than 2,000 fewer innings.)

This is a number that once rose by eight percent in 11 seasons from 1972-1982, but has since stayed relatively steady, never going up or down by more than two percent for the last 20 years. Put another way, the five highest reliever-inning seasons since 1969 all came between 2004-2009. 2014 was merely 14th on the list, almost equal to what we saw in 2000 and 2001, years generally accepted as being in the heart of the steroid era.

So we can see that the upward trend in this department really took place in the 70s, 80s, and early 90s. Why did it stop? And perhaps more importantly, should it have stopped? Name your reason for “why,” really. Though bullpens are larger, the rise in specialization that has produced more one- or two-batter appearances puts an end point on how short managers can trim starting appearances, at least if you want your bullpen to survive all season. Easier access to strikeouts has kept offense down and allowed starters to get more outs over the last few years. Pitchers still put a personal value on being “durable,” and while we’ve all moved beyond thinking pitcher wins are meaningful, we know that many players don’t agree, and so we still see managers trying to “get the pitcher the win” to keep the player happy. (The average start was 5.96 innings in 2014, meaning the general starter is right on the border of qualifying for a victory.)

As for should it have stopped, well, it’s pretty easy to see how differently the game is managed in the playoffs, and how attitudes towards that have changed. For example, in 1996, 25% of playoff starts lasted five innings or fewer, less than the 31.58% of regular season starts. This year, just over 39% of playoff starts were under five innings, far more than the 27.08% in the regular season. You probably already know the reason why, if you’d been reading FanGraphs through the postseason — because the “times through the order” penalty is a very real thing.

You’ve seen variations of  this chart before, no doubt, but here it is again, this time showing Baseball-Reference‘ times through the order stats from this year, 10 years, and 20 years ago. It’s uncorrected raw OPS, so don’t worry too much about comparing one season to another — obviously, offense is down right now — but the point is to show that the idea that pitchers get worse the deeper they go into a game is real, and there’s few times that a tiring starter is more effective than a good, rested reliever. It’s true now, it was true years ago, and it’s likely always been true:

This is why the game is so different in the postseason. Managers and teams know this to be true, but it’s a lot easier to implement in October, when days off allow for more creating pitching usage than during the season. As I mentioned earlier, you can’t have endless five-inning starts, lest your bullpen explode by June — unless you have a few quality relievers who can go multiple innings, which is very much an old-school approach that has largely disappeared.

It’s an idea that’s ready to come back. Earlier, I said that this year’s percentage of relief innings was the same as 1995’s, but what I didn’t point out is that this year’s percentage of relief appearances was nearly five percent higher. That is, in order to get the same percentage of relief innings, teams are using far more relievers for fewer outs, a figure which makes sense in the era of endless pitching changes and relief specialists. But using the bullpen like that leads to more times getting warm and more appearances without getting more innings out of those relievers, which is a straight line towards overuse. Innings from relievers haven’t changed in decades; numbers of relievers have.

Back to this year’s free agent pitchers, it’s certainly not that pitchers like Duke, Miller, and Robertson won’t be useful, because clearly they will be. It’s that they’re all now short relievers — Duke and Miller, each somewhat recently big league starters, threw two full innings only a combined five times this year — in a pitching world that demands longer relief stints to satisfy the dual masters of “don’t extend your non-elite starters” and “don’t overwork your bullpen with too many times warming and short appearances.”

We’ve seen failed starters like Miller, Duke, Wade Davis, Eric Gagne, Mariano Rivera, etc., revitalize their careers by becoming dominant one-inning relievers. The team that manages to do the same to create a relief ace, someone who can go two to three innings two or three times a week in big spots, will be the team that creates an advantage. Maybe that’s a starter who should no longer be starting for performance reasons (Tim Lincecum, obviously, or perhaps Jacob Turner?) or maybe it’s one with talent but enough health concerns to make shorter stints preferable (Brett Anderson, Chad Billingsley, maybe even CC Sabathia if the Yankees felt adventurous, etc.). Obviously this is an incredibly talented outlier, but just look at what Madison Bumgarner offered San Francisco in Game 7 of the World Series. Now imagine having something like that (obviously, not as dominating) semi-regularly.

The game has changed. Pitching roles haven’t quite kept up with the times. Relievers aren’t throwing more, contrary to popular opinion — they’re just throwing shorter. It’s time to bring back an old idea to solve a new issue.