Good Team Bullpen Games In 2010
Posted by Steve Lombardi on September 5, 2010
For this one, I started by asking Play Index to show me, in 2010 with games through yesterday, games where a team had 5+ relievers appear in a game where they each faced at least one batter and did not allow any runs.
Here's that answer:
.
So, to date, the Dodgers had 8 such games this season where the Rockies had 7 games. And, the Mets, Rays, Giants each had 6 of these games.
What does this all mean? Maybe something and probably nothing. But, it is interesting to see which teams this season had the most games where they really counted on most of their bullpen and the guys out there, for the most part, got the job done in the contest.
September 5th, 2010 at 11:26 am
The Mets-Cards 20 inning game shows up for both, but in the end both bullpens gave up runs.
September 5th, 2010 at 12:21 pm
Steve.... really interesting. Wonder how many of the games resulted in wins for the team using five pitchers?
Were the managers really "on their game" those days or just lucky? It seemed like they pushed all the right managerial buttons.
Surprising to see so many AL teams on the list without the possibility of double switches exept in NL interleague games. And Arizona and Atlanta holding down all the top spots.
The list must represent a micromanagerial in-game strategy! One could say... overmanaging?
September 5th, 2010 at 6:30 pm
Decided to stop being lazy and try to answer my own question, at least partially.
I looked at the first ten games on Steve's list. Not suprisingly, the teams making the multiple pitching changes were 10 and 0. Was half-hoping to find a game where the manager went through all those arms..... and still lost!
Four of the ten went to extra innings. When the first pitching change was made, the team making the multiple changes was ahead in 4 of the games, tied in 3, and behind in 3.
The starter was pulled after an average of 4.2 innings pitched. On unusual one was the Boston-San Francisco game where Bucholz was pulled after 1 inning. Must have come up injured or something.
The 3 come-from behind wins must have been particularly satisfying for the managers given all the substitutions.
September 5th, 2010 at 9:38 pm
Given the frequency of relief appearances lasting less than an inning, I don't really consider the number of relievers making a scoreless appearance of at least 1 out to be a great indicator of a good bullpen game. For instance, there have been 82 team games this year that featured at least 3 scoreless relief appearances of between 1 and 2 outs; those teams are 50-32. Minnesota lost consecutive games in that fashion on August 2-3; in the second of those, the last 4 relievers went scoreless, but the first guy
allowed 3 runs and blew the lead, en route to a Twins loss.
On August 11, Texas had scoreless outings of exactly 0.1 IP from 4 different pitchers -- the most in MLB this year -- yet the bullpen as a whole allowed 3 runs in 2.2 IP, and the Rangers lost by a run. Teams with at least 3 scoreless appearances of exactly 1 out are just 11-10 this year.
As great as the Play Index is, I can't find a really meaningful way to search for a "good bullpen game." No matter how many relievers made a scoreless appearance, you can't tell if someone else got hammered. And you can't do a Team Pitching Game search for just the relief innings. But given the limitations of the PI, I think a better measure would be at least 4 scoreless appearances of at least 2 outs. Those teams are 139-50 this year, a .735 W%.
September 6th, 2010 at 9:24 am
@5
"Given the frequency of relief appearances lasting less than an inning, I don't really consider the number of relievers making a scoreless appearance of at least 1 out to be a great indicator of a good bullpen game."
John A., I think this number of relief appearances correlates with exactly the opposite.... a poor bullpen, without good long relief men, a bullpen that is one huge committee!
September 6th, 2010 at 10:38 am
The 8/24 Phillies game was an extra inning affair that the Phillies eventually lost, even though they were the ones who forced extras after tying the score in the bottom of the ninth.. After all of those reliever who held the team scoreless, one finally came in and gave up not just one run but two in the top of an extra inning at home.
The 7/25 game, which the Phillies did win, may have been J.A. Happ's first since coming off the DL, and he may have been limited to five innings or a certain pitch count.
September 6th, 2010 at 6:33 pm
checking out the 20-inning game between the mets and the cards, something stuck out as funny. looking at the win probability chart, in the bottom of the 14th, a st. louis player went to second on defensive indifference, putting men at 2nd and 3rd with no outs. you can see the win probability move 1% in the direction of the mets. what accounts for this in the win probability formula? how could this possibly be favorable to the mets?
September 6th, 2010 at 10:26 pm
I'm surprised the Cards aren't on here a ton of times. La Russa is the worst when it comes to over-managing a bullpen.
September 7th, 2010 at 7:37 am
Kyle @7:
"checking out the 20-inning game between the mets and the cards, something stuck out as funny. looking at the win probability chart, in the bottom of the 14th, a st. louis player went to second on defensive indifference, putting men at 2nd and 3rd with no outs. you can see the win probability move 1% in the direction of the mets. what accounts for this in the win probability formula? how could this possibly be favorable to the mets?"
The WPA formula is determined by looking at who has won every game that's been played which had that exact situation. There are two main ways this algorithm can go wrong:
1. Variance in the sample -- this usually makes for very small and well understood errors, unless the sample size is too small for the subject.
2. Selection Bias
If the WPA algorithm is only looking at 14th innings, and not all extra innings situations, or even all potentially last inning situations, then the sample size would be very small, and I wouldn't expect it to give meaningful information about winning percentages. That said, this is a defensive indifference situation, so we don't expect it to improve the cardinal's win chance by much -- the algorithm here can't be off by more than a few percent, so it could easily be variance from a good sample (where all closing innings are used, not just the 14th). I don't know what the margin of error is on these calculations, maybe one of the blog authors knows, but my wild ass guess would be that it's in the neighborhood of +/- 1-2% for most situations.
I can also imagine a possible source of selection bias in this sample. We are looking at one set of games where there are no outs and players on first and third, then another set, with no outs and players on second and third. What kinds of things might be different about these two sets of games in terms of what the offense looks like? It seems plausible that the guy on third with runners on second and third is on average slower than the guy on third with runners on first and third. (one took two bases on a single, the other two bases on a double).
So with runners on second and third, the player on third might be less likely to score on an out, or more likely to be (or already have been) replaced with a pinch runner, which then presumably hurts their chances down the road if said pinch runner doesn't score, because a better overall player has been taken out.
That could skew the sample some towards the defensive team having a better chance with runners on second and third rather than first and third, even though in reality, given the same players on the bases, the win chance can't realistically go up from moving a player from first to second.
So I would guess one of those two things is what's going on there.
September 7th, 2010 at 7:42 am
Also, note that this is a very small error. There are probably similar size errors (from some combination of variance and selection bias) in many other situations in the WPA table, but you noticed this one because in this situation, it's possible to reason from clear and well understood baseball principles that a certain difference can't possibly be positive.
What this does tell you is that if you are making decisions about what to do using the WPA table for risk/reward, it's potentially going to give some flaky information in situations where what you do isn't likely to matter much.