The Astros bullpen through the win probability perspective

In a recent comment, I mentioned that win probability stats may be helpful in comparing relief pitcher performance.  ERA and RA fail to take into account leverage and allowing inherited runners to score.  FIP and x-FIP may be useful in projecting future performance but they don't tell us how well the bullpen succeeded in actually stopping the other team.  Since the Astros' bullpen has been a major weakness for this team--probably a prime contributor to the Astros' 7 game deficit relative to the team's Pythagorean record--we hear fans constantly asking, "why is the bullpen so bad?"

Advanced stats like FIP and x-FIP, and even shiny new stats like Fangraphs' SIERA (19th ranked bullpen), indicate that the Astros' bullpen isn't as bad as the results.  This may or may not indicate that we can look forward to improvement in the future.  Win probability stats can tell us something about what actually happened.  It's debatable whether WPA can tell us much about the likely future production of bullpen pitchers, but maybe it can give us some insight about the pitchers who have made positive contributions toward putting the team in a position to win games.  Win Probability Added (WPA) quantifies on a net basis the extent that each action by a player increased or decreased the team's probability of winning the game given the situation at the time.

The overall picture of the Astros' bullpen performance is shown by its WPA ranking compared to the other ML teams.  The Astros' bullpen cumulative WPA is -7.25, by far the worst in the majors--a full -4.6 worse than 29th worst team, and -15.7 worse than the best bullpen (Braves).  The 7.25 decrease in the team's cumulative win probability seems oddly similar to the Astros' 7 game deficit versus the Pythagorean record.  I don't know whether that comparison is analogous enough to ascribe cause and effect (probably not), but I'll leave it there if anyone wants to give their thoughts.

The relevant WPA stats for each Astros' reliever will be shown at the end of the article.  Before that, I will give my observations on each of the five win probability stats I used.  Also, as a cautionary note, like most statistics related to relievers, win probability stats over a partial season are subject to sample size concerns.

WPA

WPA shows the net accumulation of increases and decreases that the reliever's actions had on games over the course of the season.  Fernando Rodriguez easily leads Astros' relievers in WPA.  The only other relievers who have a positive WPA are Melancon and Valdez, with Melancon's WPA registering as barely positive.  All I can say about Valdez is "we hardly knew ye." Several Astros' relievers added more purely positive additions to win probability than Rodriguez, but Rodriguez gets better net results because he has had fewer negative events.  For example, Wilton Lopez has twice as much positive additions to WPA but nearly three times more negative contributions to WPA, compared to Rodriguez.

WPA/LI 

I'm not sure that this is a very useful stat, but it might be helpful if you compare it to a player's WPA.  This stat divides the player's WPA by the leverage index in order to remove the contribution of high leverage situations to the player's WPA.  A player can perform well in low leverage situations, but it will have less impact on WPA because late and close game situations provide more opportunity for the player to increase the team's chances of winning.  The WPA/LI shows the players' normal performance contribution without taking into account the leverage situation.

CLUTCH

This stat measures the extent that the player increased his performance in clutch situations.  We can debate whether "clutch" is a repeatable skill, but the existence of clutch performances during specific periods is factual.  Obviously clutch performance is very important for relievers, leading to labels like "fireman" and "gas can," depending on whether the reliever was clutch or not.  We would like to have relief pitchers who can bear down in tough situations.  Fernando Rodriguez and Valdez are the only Astros' relievers who were positive clutch performers in the bullpen.  Wilton Lopez has been the worst relief pitcher in the clutch.  This is because Lopez normally is very good, but he has been much worse in clutch situations.  Is this just bad luck?  I suspect that there is some luck involved, but who knows how much.  Mark Melancon's normal performance level also is good, but his performances haven't been as good in clutch situations.  Your view on how much this reflects response to pressure vs. bad luck might go a long way in projecting whether Melancon or Lopez should hold down the closer position  in the long term.

SHUTDOWNS MINUS MELTDOWNS

Shutdowns and meltdowns are two of the most interesting fangraphs counting stats for relief pitchers.  Meltdowns correlate highly with blown saves, but are a more encompassing statistic. Shutdowns correlate with saves and holds, but are not as inextricably linked to whether the action took place in the 9th inning. Like saves and holds, the SD and MD stats are somewhat dependent on the roles that the pitchers were allowed to perform. For purposes of this comparison, I have deducted meltdowns from shutdowns to determine the extent that negative outings overshadow positive outings. 

Melancon and Escalona are the only two relief pitchers to have a net positive in shutdowns.  Although Fernando Rodriguez fares well on most WPA stats, he has a net negative for shutdowns.  Considering that Rodriguez had 6 meltdowns in 30 games and Melancon had the same number of meltdowns in 51 games, I'm not sure we would be any happier if Fernando had been the closer.  Lopez has the most meltdowns, 12, which pretty much negates the fact that he had 12 shutdown performances.  Fulchino and Abad have the worst "net" number, which may explain why both relievers saw time in AAA.

This is how the bullpen got to where we are.  Any thoughts on how this bullpen fares in the future?