THE BOOK--Playing The Percentages In Baseball

I posted this on their site:

Good job John.

There are two benefits here, that are huge:
1. the pitcher doesn’t hit.  As noted, that’s 0.42 runs gained.  Depends how good your pinch hitters are of course.  If you want to be conservative, let’s call it 0.25. 

2. the pitcher goes from a 90-100 pitch starter mindset to a 30 pitch relief mindset.  The Book shows that this is worth around 1 run saved per 9 innings.  Your 3 pitchers here would be pitching say 6 innings, so that saves you a good 0.65 runs.  That is fantastically enormous.

The big drawback is the depleting of your bench.  You need a good enough team of starting players to make this work.

And the other drawback is the extra innings.

Is getting 0.90 runs in 2 games out of 5 worth it?  I’d think so.

However, there is one time of the year where there is no drawback and that’s starting Sept 1.  You’ve got a basically unlimited bench and bullpen here.  Plus, you should have a bunch of 22-yr old firethrowers out there.  There is no reason for any below average pitcher to throw more than 2 innings, starting Sept 1.  Turning a guy with a 5.00 ERA as a starter into a 4.00 ERA as a reliever is so enormous.

***

By the way, you’d have to take the guy out after 30 pitches or so, even if he’s got a perfecto going.  We are asking the pitcher to no longer pace himself as a starter, but instead pitch as if he’s a reliever.  You’re not going to ask a true reliever to go more than 2 innings, so that’s pretty much a hard and fast rule.

Now, on the road, he may end up just pitching in 1 inning, when his turn comes up.  At that point, the manager makes the judgement call.  He lets him bat, then he’s got him for at least 1 more inning.  If he lets him go three fine.  But, he’s gotta talk to the pitcher about possibly throwing 45 pitches as a reliever.

Otherwise, if you let him go 2 innings in all, then you’re going to need a mopup reliever-type to come in to pitch the 1 inning, so we can get back on schedule.

You could also try it just for home games as well.  Or, at least start it with the first guy acting as a starter, and the other 2 as relievers, and then they rotate.

This is really not much different than when Maddux et al come in to pitch in relief in the playoffs on throw days.

IIRC, Doug/Bob Melvin (Brewers manager, can’t remember first name) said that he read The Book. One would think that he might try this strategy with the Brewers struggling. Might as well give it a shot after Sept. 1 as you said.

As I/we have said before, this is without a doubt the most powerful strategy almost all teams can use.  Why it has not been done is a mystery, other than trying to get a team/manager to do virtually anything radically unorthodox is like pulling teeth.

While Peterson has the right idea, I’ll make one correction to this statement:

Back-end starters are often effective early in a game but are “found out” by opposing hitters by their third time through the order.

ALL pitchers lose effectiveness as the game goes on, due to the familiarity issue.  Not just “back-end starters.”

And yes, for this strategy to “work,” managers have to resist the urge to keep in a starter who is pitching a good game.  That is a tall task.  You basically have to explain to the pitchers who starts the game that they are pitching as relievers, period, no matter how they end up pitching.  Not to pace themselves, etc.  Pitch exactly as if they were coming into the middle of a game and going 1-3 innings.

By far and way, the two biggest ways that managers give away WE is:

1) Letting pitchers bat (in the NL parks only of course).

2) Letting mediocre and bad pitchers pitch more than their “fair” share of innings/pitches (5/90) because they are pitching well.  In layman’s terms, or fanspeak, the idea is to take out your back of the rotation starters before they get hammered, not after.  The latter is more often the case.  In analystspeak, the idea is to simply get as few innings from your back of the rotation guys as you can, no matter how good or bad they are pitching on any given day.

There is no question that if I ran a team, that I would utilize this strategy. No question.  It is extremely powerful.  It is worth more than all of the optimal bunts, IBB’s, pinch hitters, bullpen changes, etc., chemistry, and leadership put together.  The Mets are certainly not the only team that can benefit from this.  I dare say that every team can use it.  I doubt there are any teams whose 5th starter is that good.

It’s funny, cause I was just watching the Mets broadcast, and both Ron Darling and Keith Hernandez were lamenting the decline of the complete game.  They both called it the biggest change in the game since they played.

Put in a historical perspective, you’ve got to think that this type of strategy (the 3/3 approach) will almost certainly happen one day.  The mindset of today’s managers, fans, players and sportswriters would seem to be moving in that direction.  I wouldn’t call it inevitable, but it does appear to be a natural next step in the game’s evolution.

BTW, if Randolph brings in Heilman one more time with men on base, I’ll scream.  Does he pay any attention to his team’s splits???

What do you mean by “team splits?”

As I say, you can’t have it both ways.  Do you want a manager to manage by sabermetric principles or by “fan” principles?

While we found a small talent component for pitchers pitching from the stretch and from the windup, it is small, like clutch, and does not really enter into the decision-making process.  Plus, all relief pitchers pitch from the stretch whether there are runners on or not (I think there is at least one who doesn’t).

Now, a pitcher’s value can change a little depending upon whether there are runners on base, based on his projected component rates, but I assume you are not talking about that.

So what “splits” are you talking about?

In the bottom of the 6th today, the Giants were losing 4-0.  They had a runner on 2nd and 0 outs and their starter, Misch, was due up.  He is a bad 5th starter, who probably won’t even last the season (maybe he is a prospect, I don’t know, but he has a bad projection).

He let him hit and sac bunted.  Either the manager, Bochy, is colossally bad or they don’t care about winning games and wanted to do something that I am not privy to.

As is often the case, Misch lasted 1/3 of an inning in the top of the 7th.

You do forget one issue. A starting pitcher who pitches only two innings cannot earn the win. This will matter to many egos.

FYI, Earnshaw Cook had a similar idea way back in about 1960: start a reliever, pinch hit for him, then let the next pitcher bat once, then pinch hit for him.

We’ve mentioned that before.  Actually, if one pitcher relieves for the other, like the 4th starter relieves for the 5th starter, and vice versa, each pitcher should get around the same number of wins anyway, maybe more.  If that is the only issue that prevents a team from implementing this strategy, well…

Right, the rotation system will give the win to the better pitcher (who did not start).  Stupid MLB rule of course.  Anyway, if you’ve got 3 guys sharing the duties of pitching in the other 60 games not pitched by the big 3 solo guys, that gives them 20 starts each, and 40 relief games each.  They can’t earn a win in 20 of the starts, but are eligible for the other 40.  In the end, it all works out to the same thing.  Somebody is getting those wins.

I think the only way this happens is with guys who are seen as relievers, not as starters:  there are just too many cw barriers to using starters like this.  my guess it’s gonna have to be done by a team that finds itself in April with three “name” starters, a closer, and basically nobody else of any proven worth…

Sounds like the Mets… and almost all of MLB.  How many “name” pitchers are #4, 5, 6 starters on an ML team?  Those pitchers are always competing in spring training for the last two spots anyway.

fair enough...what I meant, though, was guys who neither have any significant mlb record as starters nor are viewed as starting prospects.  in other words, guys for whom there’s no perceived downside (from the pov of management or the players) to Book-style usage...it just seems to me that otherwise the risk-adverse nature of mlb sets up huge barriers to this kind of strategy....

As I say, you can’t have it both ways.  Do you want a manager to manage by sabermetric principles or by “fan” principles?

Just because you didn’t find meaningful splits for pitchers as a whole with men on base vs. not, doesn’t mean it’s not a factor for individual pitchers.

Heilman’s career splits:

.214 with no one on
.284 with men on

In every single year of his career, same story.  Big enough sample size and very consistent pattern.

Should say:

“...doesn’t mean it’s not a factor for some individual pitchers.”

studes,

What would cause such a split?

Studes, you can say that about anything imaginable.  The issue is not whether the sample is meaningful for any individual player.  It is whether that sample is meaningful given the spread of talent in the population that the player comes from.

Once you find out or at least estimate the spread of skill is in the population, then and only then can you regress each player’s splits appropriately.

I don’t think there is any meaningful spread of skill in the population of pitchers for pitching with runners on or bases empty, other than for starters who switch from the windup to the stretch.  But I could be wrong.

In fact, you are welcome to prove me wrong about the spread of skill among pitchers with regard to runners on/off, but please don’t just quote me a split and assume it is meaningful because it is some particular sample size (I don’t even like to use the words “large” or “small”, because one skill’s large is another skill’s small).  That might cut it on a fan site, but not on this one.  We live and die with “spread of skill” and “regress toward the mean.” Without both, you have nothing, I am afraid.

For example, you can quote me a left handed batter’s (or pitcher’s) platoon split over a “large” number of PA and it means something.  You can quote me any pitcher’s platoon splits over a large number of PA and it means something.  On the other hand, you can quote me a RH batter’s platoon splits over a large number of PA and it means little to nothing.  You can quote me a batter or a pitchers day/night split over a large number of games and it means nothing.  You can quote me a batter’s clutch splits over a large number of PA and it means a little, but not much to speak of.

How do we know which category the “reliever with runners on or off” splits falls into?  We don’t, unless we first figure out what the spread of skill in the population is.  If it is little or nothing, then it doesn’t matter how large or small (within reason) the sample is for any individual player.  The caveat being that if there is ANY spread in the population, then for an infinite number of PA for any individual player, there is zero regression toward the mean.  But in general, if there is little or no spread of talent in the population, then we can functionally ignore all individual splits (until we get into REALLY large samples of data and even then, we still have to regress).

We’ve gone over this a million times on this blog.  Let’s not go over it again.

We’ve gone over this a million times on this blog.  Let’s not go over it again.

Heh.  I’m thinking you don’t quite have the temperament to host a blog.

Using a binomial distribution, the probability of the null hypothesis for a career split like that is 2.5% or so.  I take that as a pretty convincing argument that his splits are meaningful.  The fact that it occurs in every single year reinforces it in my mind.

Hate to tell you this, but your explanation is gobbledy gook to me.  If that disqualifies me from posting on this blog, let me know!

I’d like to point out that even if there is a 2 sigma difference between .220 and .280 (more accurately, between the populations whose sample means where .220 and .280), it does NOT suggest that that particular difference is significant.  It suggests that there is a non-zero difference in the means of the two populations.

It could very well be that the two pop means are .220 and .225.

That’s why you realllllly need to know the regression equation.

Studes, of course you are welcome to participate in this blog.  But please don’t tell me that I “don’t have the temperament to host MY OWN blog.” I can do or say anything I want.  You are free to participate or not, or challenge or criticize anything I have to say (which I guess you have).  There are only two people who have the authority to ask me to change my temperament, and they are Andy and Tango.

I am sorry that you don’t understand my explanation.  Maybe someone else can explain it to you better than I, if you are interested.  You are completely, 100%, wrong in your assessment/analysis.

But please don’t tell me that I “don’t have the temperament to host MY OWN blog.”

Sorry, MGL.  I was just kidding around, making fun of your well-known impatience when everyone knows you get all sorts of people posting comments on blog.  Ask Buzz Bissinger.  I’ll be more careful in the future.

I understand Tango’s response to my post, though I don’t understand the nuanced difference between his statement and mine.  I agree that the equation I used refers to the null hypothesis, not the absolute value of the difference.

But that doesn’t explain why my thinking is wrong, as far as I can tell.

Sorry, I didn’t realize you were kidding.  I just had a tooth extracted and its been killing me ever since the anasthetic wore off. 

I can accept that Heilman has a real difference, if what you say is true (it’s really irrelevant that it happens every year… we really only care about the career).

However, you haven’t specified how much the difference really is (as opposed to what the sample says it is).

That’s my only point.  This is similar to Olney showing a win% of .577 against an expected .539.  Even if we accept that the mean has changed, that’s the only conclusion we are making.  We are certainly not saying that the mean has changed by 38 points.  Perhaps the mean has changed by say 6 points.  If that’s the case (that the true mean has gone from .539 to .545, and that this true mean can produce, reasonably, a .577 result after 800 games), then, basically, no biggie.

***

I took studes’ comment in jest.  Easier to presume that when it’s the regulars commenting, that they mean it in, at worst, a collegial spirit.

Is the .42 figure the difference between the pitcher never hitting and the pitcher always hitting (when the #9 spot comes up?) Or is it the difference between pinch-hitting for the pitcher every time and the average current situation, where PH-ing for the pitcher starts in the fifth, sixth or seventh?

The Book (first edition), p.147:

The model says 5.25 runs if the pitcher never hits, 4.83 if he hits 65% of the time (which is the current reality). So, that’s a loss of .42 runs by having the pitcher hit…

OK. Some Metsgeek commenters assumed that the lesser figure was with the pitcher batting 100% of the time.

Ouch, MGL.  Sorry to hear about the tooth.  Totally understandable.

Tango, I agree with your statement, except for the annual samples.

I think another way of looking at this is to look at each individual year as a random sample (even though they’re not truly random, I guess).  If Heilman showed the same differential for six years in a row, and the probability of that differential in a given year is, say, 40%, wouldn’t the probability of six straight years of similar differentials be .4 raised to the power of six?  I find that more compelling than, say, half the years of a negative differential and half the years of a positive one, regardless of the career totals.

It does not matter whether it is every year or not.  The only thing that matter is career totals.  Plus, Studes, all of these things are Bayesian problems.  Which means that we have to look at both the entire population and any individual in question.

The best example is the coin flip.  If I told you that I just flipped a coin 100 times and it came up heads 65% of the time, which is 3 SD from the mean, would you conclude that I likely had a biased coin (that the true heads probability was not around 50%)?  No.  Why?  Because we know that it is extremely unlikely that I was flipping a biased or trick coin.  It is overwhelmingly more likely that it was just a fluke.

It is the same thing with samples of skill sets.  Just having one player with a very rare anomalous split does not tell us much about whether and by how much that split represents a true skill set, until and unless we know the spread of skill in the population from which the player comes.  In fact, if we are able to look at that population, which we usually are in baseball, and we find that there is little or no spread of skill, which is often the case, then no matter how rare or anomalous a particular player’s splits are we have to conclude that is overwhelmingly more likely that the split was a result of mere chance.  That is where the “regression toward the mean” formula comes from.  It is short hand version of what I am talking about.

So first tell me what the spread of skill is in the population of starting pitchers with respect to pitching with runners on or runners off, then, and only then, can we make any inferences about whether a particular split for any one pitcher, no matter how anomalous and no matter how large the sample, is meaningful or not.

Actually, as Andy explains in the appendix of the book, even without having the data to be able to ascertain or estimate the spread of skill in the population, we can still make some inferences about what a particular split means given the sample size.  As it turns out, it is much less meaningful than it seems.  That is because we absolutely know that even if there is no spread of skill in the population, absolutely none, such that all pitchers have exactly the same expected results with runners on or with runners on, that given all the pitchers in the league and given 100 years of baseball, there will be plenty, by chance alone, who have anomalous splits.

Is his split between pitching with runners on and with the bases empty all that germane to whether to bring him in with the bases empty or not? A lot of other things might be going into that split.

Here’s Heilman’s career, IBB excluded:
As SP, empty_: 324 PA, .347 wOBA, 4.59 szERA, .700 DER
As SP, run_on: 277 PA, .365 wOBA, 4.45 szERA, .647 DER
As RP, empty_: 607 PA, .241 wOBA, 3.52 szERA, .734 DER
As RP, run_on: 504 PA, .324 wOBA, 3.85 szERA, .693 DER

Here is Heilman’s career, broken down into games as a starter, games entering bases empty, and entering with runners on:

SP: 601 PA, .355 wOBA, 4.52 szERA, .675 DER
BE: 881 PA, .268 wOBA, 3.78 szERA, .740 DER
RO: 230 PA, .317 wOBA, 3.26 szERA, .623 DER

Now, here’s more numbers for the relief appearances. Broken down by PA first and then by whether runners were on when he entered the game:
K/PA = strikeouts per PA
W/PA = unintentional Walks plus HBP per PA
wOIP = wOBA on balls in play
XB/B = XBH per batted ball
HR/B = HR per batted ball
XB/I = XBH per ball in play
1B% = singles divided by (singles + outs in play - Reached on Error)
Home = % of PA at home

__,__,K/PA,W/PA,wOIP,XB/B,HR/B,XB/I,1B%_,Home
PA,BE,.231,.079,.246,.060,.012,.048,.216,53.4
PA,RO,.196,.097,.294,.104,.039,.067,.252,51.6
Gm,BE,.212,.092,.246,.082,.026,.057,.206,55.2
Gm,RO,.226,.070,.350,.074,.019,.057,.331,42.5

So, comparing his PA with the bases empty to PA with runners on, he has been worse across the board with runners on, and the sample size is 607 against 504. If we break it down by when the game was entered, he seems to have pitched better in every respect but one (and with a big disparity in home/road) - singles allowed. The sample, once again, is 881 PA against 230 PA.

Breaking it down by batted ball types:
PA,BE: 1.33 G/F, 18.1% LD
PA,RO: 1.29 G/F, 17.7% LD
Gm,BE: 1.32 G/F, 17.8% LD
Gm,RO: 1.29 G/F, 18.5% LD

Here’s the wOBA by batted ball type:
PA,BE: .689 LD (1 HR), .204 FB (4 HR), .143 GB, .600 Bu
PA,RO: .826 LD (1 HR), .353 FB (13 HR), .219 GB, .053 Bu
Gm,BE: .727 LD (2 HR), .275 FB (14 HR), .153 GB, .000 Bu
Gm,RO: .834 LD (0 HR), .257 FB (3 HR), .267 GB, .675 Bu

If the HR is excluded, the LD’s are .672/.807/.703/.834 and the FB are .155/.162/.157/.161.

With runners on, 23% of his LD, 16% of his FB, and 2% of his GB have been XBH; bases empty, it’s been 16%, 8%, and .5%. His 1B% with runners on is 76 LD, 10 FB, and 22 GB against 62, 10, and 16 bases empty.

Entering games with runners on, the XBH %s are 17, 10, 1 against 20, 12, 1. The 1B%s are 83, 10, 29 against 64, 10, 16.

So, while I wouldn’t rule out some possibility that Heilman is better with the bases empty, I’m not really seeing a compelling argument in the numbers. His performance has been better, but it’s not an expansive sample and the primary differences are in sectors largely impacted by luck.

And even if we are to assume that Heilman has been better, talent-wise, with the bases empty, it does not necessarily follow that that talent split would continue into the future. And in any event, the consideration should not be whether or not to bring Heilman in with runners as a rule but rather how much our estimate of his talent should change. Studes, are you defending that Randolph use a runners-on penalty in deciding whether to bring in Heilman (for example, that he’s a .280 wOBA against entering bases empty but .305 with runners on) or that he simply not use AH with runners on if he can use anyone else?

Tom, I’m not sure what all your data means (hard for me to read the tables, and I don’t fully understand your splits), but it sounds like you have better, more detailed info than my simple look at batting average.  I’m sure it’s more useful and insightful.

Regarding your last question, obviously the better approach is the former (a “penalty” instead of a yes/no decision).

I just took all of his PA from B-R PI and parsed them. Make of his numbers what you will:

http://tinyurl.com/3v7tka

His splits for K and BB are just not remarkable and actually have been better in games he entered with runners on base. His szERA (ERA estimator using (K-BB)/PA) is 3.85 runners on and 3.52 bases empty. His distribution of groundballs, fly balls, and line drives is stable splitting it either way. Most of his split has been because of the results of his batted balls.

His HR split is seemingly extreme, but we are only talking about 19 of them. Four of them were bases empty after entering with bases empty (all four have been since 7/1/07). Twelve were with runners on that he had put there after entering with bases empty. One was bases empty to ARod to start the 8th after he had entered with runners on in the 7th and got the job done. And then two were when he came into a game with runners on and let them score on the HR, and we remember those since they happened six days apart this April.

The split in singles has the biggest impact on his overall production and it looks like he’s been lucky on balls in play with no one on and unlucky on BIP with runners on. While I’m open to the interpretation that Heilman has been serving up better pitches to hit with runners on, I don’t think there is any significant predictive value to be gleaned from the observed split.

I mentioned this in another thread, but there are generally two potential sources of random noise.  One is when a player has been “genuinely bad or good” (such as, as you say, a pitcher has served up good or bad pitches), and the other is when the “bounce of the ball” either is lucky or unlucky, such as a pitcher’s BABIP.  While one is more suggestive of “luck” or noise, they are BOTH potentially noise.

People seem to not understand this phenomenon and ascribe one to luck or noise and the other to skill.  That is not the way it works.  For example, if I am a hitter and I have 100 PA, and of those 100 PA, I hit 10 well-hit screaming bombs for home runs, that is still a (somewhat) “lucky” performance that needs to be regressed toward some mean.  Ditto if I happen to hit bunch of bleeders through the infield for a high BABIP.

Granted, one gets regressed more than than the other and one is “more luck” than the other, but BOTH are performance SAMPLES and hence need to be regressed.

So let’s please not look at a player’s performance in a granular fashion (which is a good thing - don’t get me wrong - it allows us to make better inferences about the ratio of luck to skill in his sample performance) and conclude that one type of performance is luck and another skill.  That ain’t the way it works.

Thanks, Tom. Great points.  Regarding that split, what is “pitching after entering the game with bases empty”?  Does that include all subsequent plate appearances, including those with men on base?  What do we learn from that split?

It’s possible that our proposal would have been more receptive 50-60 years ago:
http://www.baseballprospectus.com/article.php?articleid=383

Great data in there.  And their conclusion:

To describe the differences between the two eras in a sentence: Pitchers in the 1950s came out of the game when their performance dictated it; pitchers today come out of the game when their workload dictates it.

While the teams were probably foolish in the “performance dictates it” thinking, they would certainly be able to buy it if we told them to treat their below-average starters as short-term pitchers, and simply do alot of pinch hitting.

Of course, you don’t see the immediate impact, as you would with Eck.  Such is what we are up against.

Going back to the original issue… I have some questions.

1. How do you manage your bench and how does that affect your strategy later in the game?  For example, if you use your best bat off the bench the first time the pitcher’s spot is up then you can’t use him later in a higher leverage situation.  Similarly, if you use your best backup fielder then you lose the potential to use him as a defensive replacement or as part of a double switch.  If you go to extra innings how much WE do you lose by having to bat pitchers when you run out of bench bats?

2. What do you do with an injury to one of your top 3 starters?  Do you move one of the 3 SP/RP guys into a starter role or do you just bring in the same guy you would have in the 5 SP scenario?  If you’re moving guys back and forth between RP/SP and full time SP that might reduce their effectiveness in both roles.  If you don’t move the sp/rp to sp don’t you risk hurting egos?

3. Does this wear out the rest of your bullpen by making them pitch more days in a row?  You’ve got 2 out of 5 days when you know you will need at least 3 innings out of your bullpen (that’s the pen excluding the 3 sp/rp). Doesn’t that make you more likely to have runs of several consecutive days of wanting to use the same 3 guys?

31/MGL - I fully agree, if that’s not clear. It’s not as if the K/BB numbers are “true” while the balls in play are “false.”

32/Studes - Yeah, it’s all the PA accrued in games where, when he entered, there were no runners on, which is about 4/5 of his PA. Of his 881 PA in those appearances, 539 were with the bases empty (.240 wOBA, 3.64 szERA, .743 DER) and 342 were with runners on (.313 wOBA, 4.00 szERA, .737 DER). Of his 230 PA in appearances where he entered with runners on, 68 were with the bases empty (.247 wOBA, 2.58 szERA, .667 DER) and the other 162 were with runners on (.347 wOBA, 3.55 szERA, .605 DER). The average groundball against Heilman has a .190 wOBA; in the appearances where he entered with runners on, the average groundball with runners on base has had a .329 wOBA.

What do we learn from splitting it up that way? I don’t know, there is obviously a ton of noise in it. But I thought that that was worth looking at since the question was never just whether Heilman pitches better with runners on but whether he should be brought in with runners on. So, it seems worth pointing out that he has thrown more K and fewer BB in those appearances.

1. If the LI is lower than say 0.80, don’t use your better hitters.  You don’t worry about the double-switch too much here.  Extra innings is an issue.  But, we’re buying a whole run per game here, by having better hitters (+.4 runs) and better pitchers (+.6 runs) by going to the bench alot and by turning pitchers into relievers.  How much can it possibly cost to have a depleted sub for the times that extra innings happens?  This is like adding a superstar every game.  Wouldn’t you give up some sub flexibility in that case?

2. Injuries: like usual, you promote one of your quasi-starters into full-time starter mode, and bring up a minor leaguer to replace that guy.  If the timing is off, the minor leaguer will need to make an emergency start to get the timing on track.

Ego: I didn’t realize that MLB were such crybabies that they need to be coddled.  NHL players routinely are asked to forego their scoring assignments and backcheck instead.  You don’t hear them crying about it (except Brett Hull, and he won a Cup as a result).  I think this crybabying egotrip is madeup by fans and media.

3. Wear out bullpen? I guess you haven’t read The Book, where I show that relievers can handle 30% more pitching without losing effectiveness.

In any case, we have our top 3 starters average 6.2IP each x 34 starts, the 3 swinging-starters who average 2 IP per game each x 60 starts.  That gives us 992 innings from our starters and swinging-starters (6 pitchers in all).  That leaves us 466 innings spread out to 5 relievers, or 93 IP each.  Of, if the closer get 75IP, then 98 IP for the other 4 relievers.

11 pitchers, means 14 nonpitchers, of which 8 are starters and another is your backup catcher.  That leaves us with 5 bench guys, of which 4 will come in to hit.  And the last guy is Willie Bloomquist, a guy who finally finds a system where he has value.

Thanks, Tango.  I guess I need to go buy the Book and read that chapter.

2. Injuries.  I was thinking of a case where you have one sp go down for a month, then he gets better.  Then maybe a few weeks later another one goes out for a few weeks.  So the quasi-starter is being moved back and forth between being a starter and being a quasi-starter.  It is my understanding that this sort of behavior can result in worse performance in both roles.

I’m not sure where your understanding comes from.  It could be.  But, I’d like to see the research on it.

However, we don’t need to keep using the #4 guy as the guy to bounce back and forth.  The #4, 5, and 6 guys could each do that role in turn, if we’re looking for some stability.

What is really needed is studies showing the fastball speeds of pitchers, if their expectation is to pitch under 20 pitches (94mph?), under 40 pitches (92mph?), between 40 and 80 pitches (91mph?), and over 80 pitches (90mph).

We also need to know how much bouncing around they can do at various threshholds.

Then, we can translate the change in fastball speeds into runs.  What did we say just a little while ago?  That each mph was 0.25 runs per game?

This can be a huge competitive advantage here.  Heck, this one is better than every other game strategy advantage combined out there.

Teams need to target some low-level players (say 23yr olds pitching in A, or 26 yr olds in AA/AAA), and use them as guinea pigs.

For some reason, corporate America hates free stuff and advice, but will pay $10-$50K for the same service.  Weird right?  It’s all in the “you get what you pay for” mindset.  Ok, MLB, pay me $10K… Read the above… I just tore up your check… You’re welcome.

Yes, this is such a powerful strategy that even though many of the counter-arguments probably have merit, they could not possibly make much of a dent in the number of runs you would gain.

And it is really not such a radical idea at all.  You simply tell your pitchers what you are going to do beforehand.  That way, if some pitcher who is throwing a shutout or even a
no-hitter complains when he gets taken out for a pinch hitter, you can simply say to him, “I told you exactly what we were going to do and you shook your head OK.  Why are you complaining now?  What part of ‘now matter how well you are pitching’ don’t you understand?”

And as I have said before, a manager can leave a pitcher in if he comes to bat in a low leverage situation or even a bunt situation (where the pitcher bunting is not much worse than a pinch hitter hitting away).

Plus, we are talking about 4th, 5th, and 6th starters. My standard answer to the “crybabies” would be, “Well, if you were good enough to be a #1-#3 starter, we wouldn’t be taking you out!”

We’re not talking about the Webb’s, Peavy’s, or Santana’s here.  We are talking about a lot of guys who get shuffled in and out of the pen or even the minor leagues anyway.

"a manager can leave a pitcher in if he comes to bat in a low leverage situation “

But not for more than an inning… if Joba is expected to throw 30-35 pitches, and is throwing to that capacity, I don’t see how I can then keep him in for 2 innings.

I’m wondering if we can learn from extra innings… do relievers pitch all out in tie games in extra innings, or do they pace themselves as if they may have to stay in there longer than 2 innings, because they know they can’t deplete the bullpen?  A good study can be done there…

We have 2 sources of gain from using this strategy.  One, the pitchers never hitting, and two, the pitcher having a better performance when he only throws (and knows it) 30-35 pitches and faces a lineup once, or twice at the most.

For the first source, you can leave a pitcher in to hit in a low leverage situation.  For example, say that it is the top of the first inning and your team has just scored 7 runs and is winning 7-0 and you come to bat (as a pitcher).  I assume it is a low leverage situation.  You can bat.  But lets say that it is a tie game in the second inning, you have pitches an inning already, albeit to only 3 or 4 batters, and it is bases loaded and 2 outs and you come to bat.  Good idea to take the pitcher out, even though he probably won’t like it.