Last night's box score says that Doug Fister pitched 7 2/3 innings, allowing 2 runs (both earned) and 5 hits (1 HR) throwing 108 pitches.
The box score is lie.
What Doug Fister actually did can be best described as two completely separate things:
Thing 1. Fister threw seven scoreless innings, allowing only 3 hits on 98 pitches
At that point, if the Red Sox had scored their usual 1-3 runs over those same seven innings, Doug Fister would have had butt firmly planted on the bench while the bullpen went out and tried to hold the lead. Amazingly, that didn't happen -- the Sox actually scored six runs. Shocking, right? Anyway, that put the team's win expectancy somewhere around 99.5%.
From there, John Farrell had to balance two completely independent goals: (A) Turn that 99.5% win expectancy into a real W, and (B) use the bullpen as little as possible to keep relievers fresh for future games.
That completely changes the calculus of what to do with Doug Fister in the 8th inning. No longer is the team's #1 goal to prevent runs. It's to obtain as many outs as possible until Farrell has to pull him -- basically, the baseball equivalent of running out the clock.
Thing 2: Fister pitched another 2/3 of an inning, giving up 2 more runs, on two hits, throwing another 10 pitches.
Any conventional statistical analysis would treat Thing 2 as a big negative when evaluating Doug Fister's performance. But was it, really? I think not -- win expectancy dropped from 99.5% to about 98.4%, but the bullpen only needed to obtain 4 outs rather than 6. Farrell-bashing aside, I think that's a trade-off that most managers would be comfortable with. So does it really make sense to treat Thing Two pitching stats the same as any other 2/3 of an inning when measuring Fister's performance? Or, alternatively, is this such an unusual situation that it's completely non-predictive of Fister's pitching going forward?
Remember why Defense-Independent Pitching Stats became so important over the past decade or two. The idea is to measure pitching performance separately from what the pitcher has no control over. Just as pitchers have no control over whether the defense successfully turns balls in play into outs, they also have no control over whether the manager leaves him in the game rather than calling on the bullpen. Shouldn't there be advanced metrics that account for this?
.
.
.
Anyway, the reason I'm posting here is that I haven't really seen any stats or analysis that adjusts for usage patterns this way. I'm interested to hear if anyone has some studies or articles that they can point to.
I can come up with two possible approaches here:
1) Split all starting pitching stats between outcomes that occurred in the first 6 innings (or 5, or 7) vs outcomes that occur in later innings. As the example above illustrates, performance in later innings is highly situational-dependent and doesn't necessarily give an valid measure of the quality of the pitching performance. You could even reclassify those as a kind of "relief" inning and compare those stats to relievers who might otherwise be pitching in those innings -- that has the side benefit of creating a new metric, albeit imperfect, to judge whether managers should leave starters in the game vs call the bullpen.
2) Give starters extra credit for additional outs beyond the 6th (or 7th) inning. That could more faithfully measure the additional contribution by starting pitchers in giving the bullpen a rest, as in the example above, but I can't think of a good way to do it.
.
.
.
So what say you, SoSH Nation? Am I already late to the party on this, or is this worth thinking about?
The box score is lie.
What Doug Fister actually did can be best described as two completely separate things:
Thing 1. Fister threw seven scoreless innings, allowing only 3 hits on 98 pitches
At that point, if the Red Sox had scored their usual 1-3 runs over those same seven innings, Doug Fister would have had butt firmly planted on the bench while the bullpen went out and tried to hold the lead. Amazingly, that didn't happen -- the Sox actually scored six runs. Shocking, right? Anyway, that put the team's win expectancy somewhere around 99.5%.
From there, John Farrell had to balance two completely independent goals: (A) Turn that 99.5% win expectancy into a real W, and (B) use the bullpen as little as possible to keep relievers fresh for future games.
That completely changes the calculus of what to do with Doug Fister in the 8th inning. No longer is the team's #1 goal to prevent runs. It's to obtain as many outs as possible until Farrell has to pull him -- basically, the baseball equivalent of running out the clock.
Thing 2: Fister pitched another 2/3 of an inning, giving up 2 more runs, on two hits, throwing another 10 pitches.
Any conventional statistical analysis would treat Thing 2 as a big negative when evaluating Doug Fister's performance. But was it, really? I think not -- win expectancy dropped from 99.5% to about 98.4%, but the bullpen only needed to obtain 4 outs rather than 6. Farrell-bashing aside, I think that's a trade-off that most managers would be comfortable with. So does it really make sense to treat Thing Two pitching stats the same as any other 2/3 of an inning when measuring Fister's performance? Or, alternatively, is this such an unusual situation that it's completely non-predictive of Fister's pitching going forward?
Remember why Defense-Independent Pitching Stats became so important over the past decade or two. The idea is to measure pitching performance separately from what the pitcher has no control over. Just as pitchers have no control over whether the defense successfully turns balls in play into outs, they also have no control over whether the manager leaves him in the game rather than calling on the bullpen. Shouldn't there be advanced metrics that account for this?
.
.
.
Anyway, the reason I'm posting here is that I haven't really seen any stats or analysis that adjusts for usage patterns this way. I'm interested to hear if anyone has some studies or articles that they can point to.
I can come up with two possible approaches here:
1) Split all starting pitching stats between outcomes that occurred in the first 6 innings (or 5, or 7) vs outcomes that occur in later innings. As the example above illustrates, performance in later innings is highly situational-dependent and doesn't necessarily give an valid measure of the quality of the pitching performance. You could even reclassify those as a kind of "relief" inning and compare those stats to relievers who might otherwise be pitching in those innings -- that has the side benefit of creating a new metric, albeit imperfect, to judge whether managers should leave starters in the game vs call the bullpen.
2) Give starters extra credit for additional outs beyond the 6th (or 7th) inning. That could more faithfully measure the additional contribution by starting pitchers in giving the bullpen a rest, as in the example above, but I can't think of a good way to do it.
.
.
.
So what say you, SoSH Nation? Am I already late to the party on this, or is this worth thinking about?