Nowhere to Go but Up

In a recent post, I looked at the 2008 Rays' bullpen as a case study in bullpen success and how a bullpen can contribute beyond its ability in a single season to distort its value. In this post, I'd like to return to an earlier subject I've written on here, the 2008 Cardinals, to explore the opposite: how a bullpen can go wrong.

The Cardinals led the Majors in blown saves last year with 31 (tied with Seattle), and only San Diego lost more games in the bullpen, with Baltimore and Seattle tied with St. Louis at 31 bullpen losses. The NL average for blown saves was 23 per team (MLB average, 22), and for bullpen losses, 27 (MLB average, 25). The Cards' bullpen rated a little better in other areas, though still badly, but the blown saves and losses were the most visible numbers to the fans and the most intuitive for translating to win value. Unfortunately, these simple measures don't actually translate very well to win contributions. Instead, here's how they rated in some other areas (REW is Run Expectancy Wins, essentially an alternate WPA formula based on run probabilities instead of win probabilities):

ERA: 20th
FIP: 25th
K/9: 26th
BB/9: 23rd
K/BB: 25th
WPA: 22nd
WPA/LI: 21st
REW: 18th

As expected, the Cards' bullpen fared better in the metrics that don't differentiate between pitching and defense-ERA, WPA, WPA/LI, and REW-than in the ones that isolate pitching ability-FIP, K/9, BB/9, and K/BB-due to the good defense the Cards fielded behind them-but in both types of measures, they rated above the bottom 5 in baseball. Of interest to us here is how they ended up with all the blown saves and losses beyond where they rated in more advanced metrics (WPA had them as 1.92 wins below average, REW at .57 wins above average).

First, though, we might want to see how the Cards' bullpen ended up above average in REW when they were nearly 2 wins below in WPA. A number of pitchers varied a bit in these two categories, but by far the biggest discrepancy was from Jason Isringhausen, who posted a WPA of -2.97 but only a REW of -0.77, a difference of 2.20 wins. How can such a discrepancy exist? To answer that, we have to understand how the metrics are different. One key difference between WPA and REW is that the former incorporates Leverage Indexes while the latter does not: a 1-out solo home run, for example, will change the win probability of a tie game much more than of a blowout, but the run expectancy will change the same in each situation. There are more complicated implications of that, like how various hit-types compare to each other in different situations, but the basic differences more or less stem from that one idea. The Cardinal's bullpen WPA/LI (-0.42) reflects this and settles between the two measures. For Izzy, we can see this quite clearly in his gamelogs: the following table shows how he fared in his 13 appearances with a LI under 0.50 and how he fared in his 14 appearances with a LI over 2.00.

Jason Isringhausen, high and low leverage situations in 2008

	G	IP	R	ER	R/9
LI<.5	13	15.2	2	2	1.14
LI>2.0	14	10	20	19	18.00

In addition to this, Izzy also gave up 3 more runs in his next highest leverage appearance, putting all but 5 of his 28 runs allowed in the 15 highest leverage appearances of his 42 games. One of the two runs in his 13 lowest leverage appearances shown above was in the highest leverage of those games, and he didn't allow a run in his 7 lowest leverage appearances of the year. This fits well with anecdotal evidence, as Izzy was horrendous to start the year before losing the closer job and being shut down with a hand injury that some questioned as an excuse to get him off the roster. He then returned in a middle relief role and pitched much less abominably. Obviously, with the diminished role came lower leverage situations.

This seems to be one of two things, or likely a combination of the two, causing the discrepancy. It could just be a fluke, as such results in small samples tend to be, or Izzy may have really been pitching hurt early in the year as he was in his bad 2006 campaign. It is likely not any indication of nerves or other true skill of Izzy's, considering he has been better in higher leverage situations over his career than in less consequential situations, and he has a reputation, like a lot of closers, of not being completely right when asked to pitch without the game on the line.

In my last article on the 2008 Cardinals, I mentioned that it was debatable whether considering LI was appropriate in evaluating bullpens. In theory, using LI-weighted stats like WPA accounts for the best pitchers being used in the most crucial situations and the reality that not all situations a bullpen faces are of the same value, and such stats more closely approximate the actual contributions of a bullpen. There is enough noise in the random variation across different leverage situations, however, to counteract those benefits in simple WPA. It would take a more sophisticated method of compiling WPA (like weighting the closer's or other bullpen ace's peripherals more heavily in computing projections to account for their higher leverage usage) to take advantage of this. As it is, both WPA/LI and REW in one season are better predictors of a bullpen's WPA the following year than WPA based on the correlation exhibited in paired seasons since 2003. This means the true value of the Cards' bullpen, and what they should expect from the same cast next year, should be closer to the more optimistic values than the low WPA they posted.

This still leaves us with our initial question. We've looked at where the additional losses in WPA came from, but what about the additional losses in the bullpen's actual record? The blown saves can be partially explained by the Cardinals having the second most save opportunities in baseball and by the disproportionate number of losses distributed into blown saves (note how the blown saves compare to the League average and how the losses compare to the League average), but the Cardinals still lost 4 more games in the bullpen than the average NL team and 6 more than the average ML team. Why is that more than the WPA figure, and why can't we just use that as the bullpen's value?

Bullpen losses do not isolate the work of the bullpen in the same way as the other stats above. Losses are influenced by the situation inherited by the bullpen as well as the run support provided by the offense. A bullpen that comes into a tie game in the fifth and gives up 1 run with no run support will take the loss, but one that comes in down by 1 in 8th and gives up 5 runs to prevent a 5 run rally in the 9th from winning the game will not, even though the latter did much more to hurt its team and contributed much more to the loss. For the Cards' bullpen this year, run support in particular was an issue.

The Cardinals led 87 games through 6 innings last year and were tied through 6 in 17 more, a good 9-10 games better than their Win-Loss record. The difference between the first 6 innings and the last 3 was not all in the bullpen, though. The offense was scoring on a pace of 5.10 runs per game through 6 innings in games last year, but only 4.33 runs per game in innings 7-9. In general, teams score less in the final 3 innings, but even after accounting for that, the Cards scored .27 runs more per game than the average team through 6 but .06 runs fewer per game than the average team in the final 3 innings. The offense's skew toward helping the starters more than the relievers redistributed some of the losses to the bullpen. The difference in run support was substantial enough cost the bullpen about 2 extra losses that should have been pinned on the starters instead. Again, this fits the general perception, this time in regard to the over-performance of the rotation.

In this case, there really is nowhere to go but up. This won't all be of benefit to the Cardinals, because a good chunk of the improvement in the bullpen is bound to be in areas that didn't actually hurt the team, namely the redistribution of losses among the rotation and the the bullpen, but some of it will. This also doesn't account for changing personnel: Izzy will likely be gone, Perez and Motte should be up the whole year, Kinney should be a much bigger contributor, and Trever Miller will shore up the left side of the pen. Russ Springer, the Cards' best reliever, is gone for 2009, and the second left-handed spot is still uncertain. So the focus here is less about projecting for next year (an area the Rays article dipped deeper into) than about looking at what went wrong this year. Still, it would seem highly unlikely that the bullpen will look so ugly again next year, even without the personnel moves, unless they do something absurd like give Jose Oquendo a set up role or keep letting Randy Flores pitch over better options (like, oh, say, Aaron Miles).