THE BOOK--Playing The Percentages In Baseball

MGL, I’ll trust you over TLR on in-game strategy, but:
1) Carpenter was likely safe on that bunt attempt, no matter what
2) Carpenter was pitching incredibly well, and efficiently. Seemed a better bet than Motte, Utley’s scary fly ball notwithstanding.

I certainly don’t care whether Carp would have been safe or not.  It is not relevant to my point.

Do some research on ace pitchers (you can pick 10 of them) and look at their wOBA, ERC, or RA per 9 their 4th time through the order (they would likely be “pitching well” or they wouldn’t be facing the lineup the 4th time - or you can only look at when they have allowed 2 runs or less).

If their composite numbers are better than Motte’s average numbers (say, for the last 3 years), I’ll give you a thousand dollars. If they are worse, you give me $500.  Deal?

You spelt halladay wrong.

Thanks.  Now I can sleep…

i agree tlr is an ass but he should NOT have pinch hit for carp ever in this game...this game was all on carp’s shoulders to win

I agree 100% with the bunts being lame.  I would’ve left Carp in to pitch the 9th though.  But I did think that Utley shot to CF was going out.  Whew!

#5 and #6, and because you think it, that makes it right?  You want to accept that bet also?

I’ll bet you won’t accept that bet.  That is because when people who have little expertise on a matter have an opinion on that matter and those opinions are not supported by evidence, they never take those bets.  I wonder why?

You see, anything that I posture on this blog, I will always stand by it, because it is almost always based on evidence or my experience, knowledge, or expertise which has been gleaned by evidence.  I learned a long time ago that my opinions without a solid base aren’t worth jack…

haha, terrific stuff. many will tell you you’re wrong, mgl. but few (and more likely, no one) will be confident enough in their convictions to bet against you.

if you could book a wager for every unsupported assertion you read on the internet, you’d be a rich man.

Good stuff Sunny!  How ya doing?  I’ll be back in NV at the end of the month. Give me a holler if you are in town.

I’m even willing to lose a bet or two.  But, if there ever were a lock in gambling, it is betting against some of the assertions that sports fans make!

Ah, but would you be as rich as Tony LaRussa who has managed teams to thousands of wins and multiple championships?

I mean if money is the only important criteria…

Quick and dirty, clicking through baseball reference pages—so I only have earned runs and not just runs—but here’s the performance of Carpenter + the top nine pitchers by Fangraphs WAR, 2006-2011, in the ninth inning:

Chris Carpenter in the ninth inning, career: 34.2 innings, 10 ER

Halladay = 72.1 innings, 20 ER
Sabathia = 28 innings, 6 ER
Verlander = 14 innings, 1 ER
Lee = 28.1 innings, 9 ER
F. Hernandez = 16.1 innings, 1 ER
Haren = 17.1 innings, 3 ER
Lincecum = 8.1, 6 ER
Beckett = 6.2, 2 ER
Vazquez = 28.2, 8 ER

And someone might want to check, but if I counted right, that’s 254.2 innings and 66 earned runs between the ten aces. 2.33 ERA in the ninth inning.

Jason Motte, career: 188.0 innings, 61 earned runs, 2.92 ERA.

Again, this is just ERA in the ninth inning being pulled off baseball-reference’s splits pages, and an enormous amount of the info is Doc and not so much for some other guys. So it’s not the best way to do it, but it is the fastest.

But maybe it’s not as cut and dry as it seems?

For one thing, you gotta use out of sample numbers.  By that, I mean this:

If I take all batters who batted .300 in year X and looked at how they batted in the 9th inning in year X, it is going to be .300. 

But when we decide who to pinch hit in the 9th inning of a game in year X and so far batter B has batted .300 in year X, we don’t expect him to bat .300 in that pinch hitting appearance (not counting the pinch hit penalty). We have to use his projection which will be maybe .280 if all we know is that he batted .300 in that year.

So of course if you look at any pitcher who has a great WAR (or ERA or whatever) in year X and then you look at how they did in the 9th inning in that same year, it is going to be great.

What you have to do is identify your great pitchers BEFORE that 9th inning and then look at how they did.

You also have to adjust for the batters, park, etc.  It is likely that a starter pitching in the 9th is facing the bottom of the order of a weak batting team in a pitchers park in cold weather.

I’ll look into some more later today…

Could care less about your stupid bet proposal. lol

You have other issues with using 9th inning runs.  If the other team wins in a walk off and it is not a HR, the pitcher doesn’t get docked for any other runs that might have scored (like a bases loaded triple) or any runners that might be left on base, right?

And if the starter allows a base runner on two and then the closer or another elite reliever comes in, the starter gets the benefit, right?

I say, “right,” because I am not really sure at 2 in the AM…

Yeah, there are a lot of problems using ninth inning runs. You would certainly know better than me, as clicking through baseball-reference pages is basically the limits of my research abilities. I’m curious to see what a better answer looks like.

Aren’t walk offs all kinds of problems for this? Because, as you said, looking at the runs allowed, runs that might have scored otherwise won’t be charged to the pitcher on a walk off. But if you look at something like wOBA against instead, it’s the same problem. If a pitcher allows a hit on ball in the gap that is normally a double or triple, but it’s a walk off with a runner on third, the pitcher is just going to get charged for giving up a single on the play. So even if you just look at wOBA against, it’s going to be lower than it should be. Right?

At the same time, a closer pitching in the ninth is going to have this same advantage.

I don’t understand the second bunt either because it took the bat out of the hands of Pujols. You already had the insurance run in scoring position. But he gave away an out, and ensured that the best hitter in baseball would get walked. Was the addition of an additional insurance run in scoring position worth giving away an out AND taking the bat out Albert’s hands?  I think not.

I agree the Jay bunt was stupid, strategically—it’s just that in your rant you said Carpenter would have been out but for the error.

I won’t take your bet, because I know you’re right. And I read your book! You are playing the percentages. I do wonder if there’s a way, statistically, one could justify LaRussa keeping Carpenter in. Assume Motte has an average performance of x but with a deviation of y. Carpenter has an expected performance of slightly worse than x, but with a smaller deviation. Then some argument could be made about Carpenter being the lesser risk (not to mention saving Motte for extra innings in case Carp gave up a run?).

FWIW, after Carpenter got the hit I thought it might be bad for him expending energy on the bases (maybe a myth… has that ever been tested? Pitcher wOBA against in innings after he gets on base vs. not?).

PJF: b-ref has times through the order, e.g. here: http://www.baseball-reference.com/players/split.cgi?id=carpech01&year=2011&t=p#times

MGL you know yourself that TLR and every other manager wouldn’t remove Carpenter.  It’s Carpenter’s Game To Win. The censure TLR would receive had the closer come in and blown the game -in an elimination game, after Carpenter had thrown 8 scoreless - wouldn’t be worth it for him to take the risk, for any manager to take the risk.  It’s not as if the payoff’s worth it, either; he would still probably get criticized for removing Carpenter.  At best, he’d be “questioned” for it on platforms like Around The Horn. 

I’m not a TLR fan by a longshot.  Until the culture of managing changes, however, and managers are able to make the correct - though not the necessarily popular - decisions without fearing public excoriation, I honestly question whether it’s fair to criticize them for these kinds of decisions.  In football, coaches should be trying to convert on 4th downs way more than they are.  Many times, the decision isn’t even close.  But what can they do?  Bill Belichick “went for it” on 4th down against the Colts YEARS AGO and here, in New England, I STILL hear him criticized for doing so.

Its funny, I’m usually one that is complaining about over managing, especially in the playoffs.  For instance, it seemed very odd to me to pull Edwin Jackson after only 77 pitches in game 4. 

Anyway, I had money on St. Louis in the series, which as you point out is the ultimate test… and I wanted Carpenter in the ninth badly over Motte.  Not saying you are wrong, as you have the evidence to support your assertion, just saying I think there are hundreds of more obvious managerial mistakes throughout the season in the league.

MGL, I have no idea what you are saying in post #12 other than

“Yes, I know I said pick any 9 pitchers + Carpenter, but you need to pick other pitchers, ones that prove MY point, not yours.”

You said pick any 10 pitchers, he picked the top 10 in WAR the last 5 years.  That’s about as fair as you can get.  One of the pitchers in his sample allowed 6 runs in 8 innings.  If he were cherry picking, don’t you think he would have left Lincecum out?  Their combined ERA is superior to Motte’s.  OK, you said use RA, but before you go there, Motte has an RA of 3.3.  The ace group would need 29 unearned runs to surpass that.  You lose the bet, take it like a man.

MGL is saying you can’t pick pitchers that INCLUDE the data that we are trying to isolate on.

What you should do is look at the pitcher’s wOBA the first three times through the order, select the top 10 pitchers out of that, and then look at the out-of-sample wOBA the 4th time through the order.

HOWEVER, given that very short rate of PA the 4th time through the order, you’re going to get the same list of top 10 either way (probably).

And yes, you can’t look at runs allowed in the 9th inning, because of the bottom of the 9th scenario.

In any case, BR.com shows the splits by times through the order, and he shows OBP/SLG as well, so that’s what you should use.

Personally, I wouldn’t have the balls to remove Carpenter in that situation, even if I know I should.

By the way, I’ve got my DB to handle this times-through-the-order like Sean has (through 2010 anyway).  So, later today, I can spit out the wOBA, and then we’ll see the results.

LaRussa may have gone against win expectancy, but I wonder how many managers would have pulled their ace in that particular situation for a reliever?  Maybe CC comes out for Rivera?

LaRussa may have been wrong, but I would bet that most every manager would have done the same.

Most managers probably would have had Jay bunting there too, even though I was yelling at TLR on TV to not.

Now Skip/Jay stuff?  Well…

"MGL is saying you can’t pick pitchers that INCLUDE the data that we are trying to isolate on.”

I’m not following.  You can’t look at those aces because they did well in the 9th inning?  Should we only look at aces who did poorly in the 9th inning?  Should we not look at 9th inning performance at all to determine if they should pitch the 9th?

He said look at aces performance in the 9th innings.  Poster #11 looked at aces, and then looked at their 9th inning performance.  He didn’t pick them because they did well in the 9th (or he wouldn’t have picked Lincecum with 6 runs in 8 IP).  He picked them because they were the top 10 in WAR the last 5 years.  What’s to isolate? 

I picked another 11 aces, Santana, Greinke, Oswalt, Shields, Cain, Kershaw, Weaver, Josh Johnson, Buehrle, Lester, and Wainright.  They add another 140 or so innings at an ERA ~2.7, If not those 21, what aces should we be looking at?

@Freddie Mac/26: Yes, you can’t choose your guys because they do well in the 9th inning, and then say “Look, my guys who do well in the 9th do well in the 9th.”

My only significant possible beef with MGL’s analysis is the Punto in the starting lineup thing, because there may be some other reasons (biggest example I can think of is health) that you would want one guy over another in your lineup. Of course, I have no evidence that TLR was basing his decision on this, but I’m always hesitant to second-guess a manager on something like that on a one-game basis, because they almost always know something we don’t.

@Freddie Mac/26: Yes, you can’t choose your guys because they do well in the 9th inning, and then say “Look, my guys who do well in the 9th do well in the 9th.”

But they weren’t chosen because they did well in the 9th inning.  They were chosen because they were aces, top 10 in WAR the last 5 years.  The challenge was to look at ace performance in the 9th inning and compare it to Motte.  The original list had 10, I added another 11.  Most did much better than Motte, some, like Lincecum and Kershaw, much worse, but in the aggregate, these 21 pitchers, the best in baseball, did much better.

You can pick a diffenet 21 I suppose, but if you have a list of top 21 pitchers which doesn’t include Felix, CC, Carpenter, Halladay, Lee, etc, you’re the one with the selection bias.

After all, the issue is not that Tony let Kyle Lohse finish the 9th, it was Chris F Carpenter.

MGL:  “What you have to do is identify your great pitchers BEFORE that 9th inning and then look at how they did”; WW:  “Yes, you can’t choose your guys because they do well in the 9th inning, and then say “Look, my guys who do well in the 9th do well in the 9th.””

Their point is that “being an ace” includes “having pitched well in the 9th”, and when we think we’re simply identifying aces we are actually using this 9th-inning info as part of our id process. 

Intuitively, the objection seems pretty weak.

Take the 21 aces, subtract their 9th inning, does anybody think they won’t still be a group of aces?  No.  Does anybody think there’s a set of pitchers out there, “set AA,” who, if we simply removed 9th inning blots from their RAs, would suddenly emerge as the best 1-8 starters in baseball?  No.  Is the MGL claim that we MUST find and use “set AA”:  a set of guys who were otherwise excellent pitchers but who faltered in the 9th?  No--it can’t be this claim, because “set AA” clearly introduces its own bias.

So the claim is just that “the best 20 or whatever starting pitchers in the league once you remove 9th inning numbers” might be a significantly different data set than “the best 20 or whatever starting pitchers in the league”.

Intuitively, I’m with Freddy:  this seems a very weak claim to me, a claim made by guys who are scrambling around to avoid saying “Boy--I was wrong on that one!”

But I’m not a statistician.

Take a moment here and try to follow what I’m saying.  If you are not following, then ask me to explain it further.  Don’t disagree!  (At least, don’t disagree until you understand completely, at which point, you won’t disagree anyway.)

You have to separate in-sample data from out-of-sample data.

The out-of-sample data is the 4th time through the order (or the 9th inning).

The in-sample data is whatever else EXCEPT the above.

In order to come up with a top 10 or top 20 list, you base it STRICTLY on the in-sample data. 

Once you have the in-sample data, THEN you can look at the out-of-sample data.

You cannot include both in and out of sample data, and then compare it to the out of sample data.

Now, MGL’s point is technical and not really practical, because we know, as baseball followers, that the vast majority of a pitcher’s PA will not occur in the 9th inning.  And so, to select the best pitchers based on in-sample or all data will still result in the same answer.

Got it?

nick/29 cross-posted with me.  He gets it.

By the way, there’s no “guys”.  It’s just MGL that’s putting out that challenge.

OK Tom, I get what you are saying.  You essentially agree that it’s a distinction without a difference (and a petty one at that).  Those 21 “aces” probably have over 20,000 career IP, and fewer than 400 in the 9th inning.  I doubt their 9th inning performance moves their career ERA by more than a point.

thanks Tango--"guys" was meant to refer to mgl and the other guy who I quoted in my first parag, btw…

If we’re adjusting for the spirit of the bet, shouldn’t we adjust for other things as well?

Carp’s performance historically in the 9th, when he was a young man, pre-surgery, in games more likely to be against the Pirates and other lousy teams with poor plate discipline, is going to be different than his performance as an older guy, post-surgery, against a playoff-caliber offense.

Surgery is impossible to adjust for, but age is possible and quality of competition is downright necessary, IMO.

Yes, it’s a distinction not to worry about in this case.  In 6 years, other than Halladay, those guys averaged 4-5 innings per season, compared to their likely 200 innings overall.  2% is not going to move the needle.

I’m not going to get into a discussion of whether it’s “petty”.  Technically, MGL is right.  Practically, you’re going to get back at least 9 of the same 10 guys.

However, definitely you need to use wOBA (or weighted OPS, wOPS) and not ERA, because of the 9th inning abbreviated scenarios.

"weighted OPS, wOPS”

As a proud Italian-American, I suggest you find a new abbreviation, Tango! 

I’m a proud Italian-Canadian, and I’m ok with it!

We can pronounce it whoops.

Tango, do you think there is a such thing as a pitcher just having his stuff that day? I know you’ve done articles about mediocre pitchers allowing 0 runs through a certain point of the game, but do you think it’s possible that these pitchers are pitching in the 9th because they have their best stuff that day and they are performing well in the 9th because of that?

Sure, absolutely.  Most pitchers are human (all of them, if you include Pedro and Rivera).

The question is never: “Is this non-random?”

The question is always: “Will this metric help me find it?”

So, does having a perfect game through 8 portend a better outcome in the 9th, than otherwise?  Does having 12K and 0 BB after 8 mean that he will be better than usual in the 9th?

Or, if those things don’t help, then what metric will?

This is the point.  To see which metric really does illuminate with an answer.

I don’t know the answer (beyond what I’ve alreayd shown in The Book).  That’s why we need research.

They’re really killing MGL:

http://www.baseballthinkfactory.org/files/newsstand/discussion/mgl_worst_managing_ever/

***

I also don’t know how I got dragged into it.  I’ve shared no opinion on the matter, other than on the technicals.

Though near the end there, a couple of people tried to clear that up.

***

As for my comment:
“Personally, I wouldn’t have the balls to remove Carpenter in that situation, even if I know I should. “

I wasn’t saying that I know I should remove him.  I didn’t write that well. 

I meant that even if you could prove to me that the right thing to do is to remove Carpenter, I still wouldn’t do it.

I have no opinion as to whether the right thing to do otherwise was to remove him.  I haven’t researched that yet.

I didn’t realize that my comments of not having an opinion would then be parsed to show that I did have an opinion!

I went over and read the comments. It’s really amazing. Several people there professed to understand about small sample size, then they went on and showed that they didn’t. There were also a few appeals to team chemistry. There was a lot of this sort of reasoning: Yeah, I know such-and-such is true, but in this case…

As for letting Carpenter pitch in the ninth, deciding whether or not it was the right thing to do depends on the criterion one chooses to make that judgment. From the purely analytical point of view, it was obviously a mistake to let him face the lineup for the fourth time. However, LaRussa is also managing human beings, and may have thought that his team felt that Carpenter deserved the right to get the complete game. If I had been LaRussa, I would have brought in Motte to pitch the ninth, but there are other ways of viewing the matter.

"Personally, I wouldn’t have the balls to remove Carpenter in that situation, even if I know I should.”

I probably would not have either.  And I don’t disagree with what #20 said.

As far as, “This was Carpenter’s game to win or lose,” that is fine during the regular season, but not in the post-season, let alone the deciding game in a series.  If a manager does not do everything he can do to increase his team’s chances of winning a game during the post-season, then he is being derelict in his duty. That is my opinion.  If someone wants to say or think that 1% here or there (in WE) is not enough to worry about even in the post season, I would not necessarily quarrel with that either.

“Of course, I have no evidence that TLR was basing his decision on this, but I’m always hesitant to second-guess a manager on something like that on a one-game basis, because they almost always know something we don’t.”

That is absolute nonsense.  That comes from baseball insider apologists.  I have worked in a front office (more or less).  Managers make important decision all the time based on gut, instinct, bad statistics, etc.  Rarely do they “know” things that we don’t know that makes their decisions correct even though on the surface they may seem incorrect.  Sometimes, but rarely. Or just sometimes but not very often, if you will.

Prime example:  TLR was asked why he walked Ruiz to pitch to Francisco in game 3.  Surely he “knew” something that we didn’t know.  No, his answer was, “That was an easy one.  Ruiz has killed us this year.” That is typical of the decision-making processes among most managers.  It really is. The irony was that Ruiz was like 3 for 20 in the series, wasn’t he?

Anyway, if I am wrong about the 9th inning thing (which was just one of around 4 things I mentioned), then great!  I don’t generally learn much when I am right - only when I am wrong.  I long for that…

Wow, nice folks on BTF. I wonder why I left that site years ago.

What is incredible is how many people there use the result of the plays, the game, and the series to refute my claims.

I wasn’t aware that someone took me up on my bet and then they got to choose their methodology, players, etc.

Before someone takes me up on one of my bets, please do the following:

1) Say, “I accept your wager.”

2) Identify yourself and let us know how you intend to pay up if you lose.

3) Figure out a way to choose a third party to verify or conduct any research that is needed.

If I lose I will gladly pay up and learn something in the process. Someone (who has no idea who I am) on BTF actually said something to the effect, “...because the dickhead never pays up...”

I am doing some research now on the 9th inning thing…

Just to be clear, I was not taking up your bet and had no intention of doing so.

I sort of realized that.  The morons on BTF think that you did.  Your research was interesting and appreciated. I was just pointing out some of the things to watch for when doing that kind of research.  As I said, as we speak, I am doing some work to look at elite starters’ and relievers’ performance in the 8th and 9th innings…

Tango - as someone who isolated your “Personally, I wouldn’t have the balls to remove Carpenter in that situation, even if I know I should.” at BTF - I understood what you meant wrt, er, ‘courage’ (see minesweeper’s #20), but misread your original statement to mean that ‘you knew’ it was the right answer - which seemed out of character at the time and like a misreading on my part now (looking back over your original quote). So, sorry about that.

As for MGL’s: “Of course, I have no evidence that TLR was basing his decision on this, but I’m always hesitant to second-guess a manager on something like that on a one-game basis, because they almost always know something we don’t.”
I freely admit to being an apologist in this regard - intentionally so. For one thing - they certainly _do_ have access to info that we don’t (principally thinking health and biomechanic/matchup stuff) - some of it pertinent, some of it not.  While it obviously doesn’t preclude managers making bad choices based on instinct/whatever - I do believe like we bear the burden on proof in arguing why they’ve made a bad decision - rather than starting from somewhere more neutral.
(FWIW, my take at the time was to pinch hit for Carpenter if the opportunity arose in the 8th, otherwise, bring him out for the 9th - implying that I didn’t see a huge benefit/loss from swapping pitchers or keeping Carpenter out there.)

As for your anecdotes in #42, I’m generally unpersuaded by arguments that can more or less be boiled down to “they’re idiots!” - which is one reason while I miss your presence at BTF, I don’t miss your tone.

Anyway, carry on gents…

MGL, I’m doing research as well!  (I’m not near a TV, so when the rain delay is over, someone pipe up!)

It’ll be interesting to see what each of us turns up.

@MGL/42:
Probably I should have elaborated. I maintain that managers usually know something we don’t. Obviously they’re talking to the players and coaches and trainers a lot more often than we are, and presumably they can glean some kind of medical or psychological reading of the player.
Now what I won’t say is that managers *should* or in fact *do* make decisions based on this kind of stuff very often (I don’t know how often, but I’d guess maybe 5% of such cases). I think much more often many of them base their decisions based on someone ‘owning’ someone else, but any one instnce we can’t be sure of. Of course with TLR, there’s a pattern of this stuff.
Also, I assume that at least some of the managers (quite probably not TLR, at least in this case) are lying in these press conferences. Mostly I’m thinking of some stuff Joe Maddon has said that’s rather stupid and conflicts with his much smarter decisions in general. I presume he’s trying to protect his true decision making process from being known. Also some managers will clearly defend their players - there are examples of guys bunting or stealing, getting out, and the manager being visibly furious, then saying that it was his decision rather than the player’s later in the press conference. Well, this probably isn’t the case very often, but it is possible.

Speaking of bets, if you recall, at the end of last season, there was a thread about Mike Scioscia and we had some bet about whether the Angels would outperform their pythag.
Well guess what?
They did. Just barely, but they did. Again. For the 8th year in a now.
I wish I remembered what the terms of the bet was.

#48, what do you mean by “psychological reading of the player”, and how would it affect who should or shouldn’t play? Are you saying that major-league players have crises of confidence? If so, what kind of evidence do you have for that?

I have my data setup for 1993-2010, so that’s what I’ll be using.

I looked at all starting pitchers, and came up with their wOBA, at night, for the first three times through the order, min 2000 PA.

Here are the top 15:
linct001
martp001
maddg002
johnr005
smolj001
zambc001
santj003
hamec001
peavj001
clemr001
schic002
hardr001
browk001
hallr001
cainm001

All the names should be recognizable.

The totalled 77,530 PA for the first three times through the order at night.

Their average wOBA was .293.

Now, how did they do the 4th and later time through the order?

They had 5,676 PA (which is 7% of all their night time PA).

Their wOBA?  .313.

That’s ridiculously good, especially since the overall league average for all starting pitchers was a bit over .340.  The league average for all relief pitchers was .330.

Motte (through 2010) had a nighttime wOBA of .317 in relief.

Among the 199 relief pitchers with at least 1000 PA at night, their wOBA was also .317.

So, if the choice is between Carpenter or a pretty good reliever, it’s probably a wash. 

Ryan Madson has a .301 career wOBA through 2010, so, his quality level would be the slightly preferred choice.  Motte had a great 2011, and if you add it to his career through 2010, he’s probably going to be close to Madson’s career wOBA.

My call based on this evidence?  A slight preference to bring in Motte over Carpenter.  But, there’s definitely enough uncertainty there that allowing Carpenter to pitch in the 9th is a very reasonable choice.

The tougher choice for Larussa though was letting Carpenter bat in the 8th inning in the first place.  He played with fire there, but he did not get burnt.

***

It’s clear that quoting 9th inning runs allowed was terribly deceiving.  I don’t blame the poster for doing that, because we’ve all been ensnared by the 9th inning partial-inning rules of baseball.

As an example, here is the runs scored per 27 outs of each half-inning, since 1993:

So, from innings 1 through 8, the home team scores 0.44 more runs.  But in the 9th, they score 0.13 LESS.  That’s a 0.57 RA9 swing.  Why is that?  Because of partial innings.  Once the home team wins, the bases get cleared.  Those runners on base have the effect of being considered as “out on base”.

This is why you can’t look at ERA in the 9th inning.  But, you can look at component numbers, like wOBA or OPS.

I created a new thread:

http://www.insidethebook.com/ee/index.php/site/comments/how_do_great_starting_pitchers_pitch_the_4th_time_through_the_order/

***

Thanks for the update!

McCoy, I moved your other post to the other thread.

"It’s clear that quoting 9th inning runs allowed was terribly deceiving.  I don’t blame the poster for doing that, because we’ve all been ensnared by the 9th inning partial-inning rules of baseball. “

Well, that and that was part of the original terms set out by mgl.  From post #2:

“Do some research on ace pitchers (you can pick 10 of them) and look at their wOBA, ERC, or RA per 9 their 4th time through the order (they would likely be “pitching well” or they wouldn’t be facing the lineup the 4th time - or you can only look at when they have allowed 2 runs or less). “

Emphasis mine.

Well, he shouldn’t have included it, for reasons stated.

I was going to look into it, but Tango did the heavy lifting in the other thread about starters the 4th time through the order…

Since coming back from his injuries Carpenter has done this against the 4th through or more:
.205/.231
.220/.156
.284/.299

That is 186 PA and 3 IBB.

At this point I think it looks like Tony knows when to let Chris Carpenter pitch deep into a game.  Now whether or not Chris should have gone up to bat late in this game is another issue.  You pull him in the 8th and you are asking the bullpen to not give up a run to the Phillies over the next two innings.  I could see pulling him for a PH in the 9th but before the 8th seems a little too early for Chris last night.

Jason Motte this season in 268 PA has a .262/.296 line with 2 IBB.

Please don’t use sample sizes of ONE (or two) player(s) on this blog to support a general point (such as, “Is an elite starter better or worse in the 8th or 9th innings than an elite reliever?"). 

Using one pitcher’s stats is like using batter/pitcher matchups to make decisions about whom to bat or pitch against whom.

If you want evidence to try and determine whether Carp should have pitched the 9th or Motte, I know this sounds counter-intuitive, but you can NOT use Carp’s and Motte’s stats, other than to categorize them (such as “elite starter”, “elite releiver”, etc..  Once you can categorize them, then you do back in history and look at large samples of players (and PA’s and TBF’s) with those same characteristics.  That is the ONLY way to do it.  How Carp has done the 4th time through the order or in the 9th inning tells us almost nothing, believe it or not, about whether he should have been used or not, simply because the sample size is infinitely too small to have any certainty about it.  Not only that, but you can easily run into selective sampling issues when you do that.

Here is a good hypothetical example of that last point:  Let’s say that you have 100 starting pitchers and they all have the same 9th inning true talent. One of them is likely to have a performance in the 9th inning that is 2.5 SD’s or so better than the mean of the group and the true talent of himself, just by chance alone.

And let’s say that managers have to choose one of these pitchers to pitch the 9th in tomorrow’s game.  He is likely to choose our one pitcher who got incredibly lucky in the 9th inning all year.  Now, I submit that he should not have been used over a great reliever.  You look at his prior stats and say, “What are you talking about?  He has pitched brilliantly in the 9th all year, and that is unlikely due to chance (p<.01)!  You are an idiot!”

Do you see the problem you can run into there?

I’m not using a sample size of one to argue whether or not an elite pitcher is better than an elite reliever in that situation.  I’m using Carpenter’s number to see whether or not Tony has correctly identified when Carpenter still has “it” late in the game over the last three years. 

How does one answer that question without looking at Carpenter’s numbers?

First you have to establish how much randomness there is in whatever split you are looking at.

mgl said the former, and I said the latter.

mgl and I are two independent people.

As usual, the bullshit thing is that mgl (or tango) posts something thought-provoking, a bunch of people cry “blasphemy”, nobody cites any actual counter-evidence, and in the end mgl (or tango) ends up doing the real research themselves to support the original point. This happens all the time on Gabe’s hockey blog, and it should be the opposite.

In this case it’s unfortunate that the exact words of mgl’s proposed wager make the situation close, because it doesn’t take away from the fact that this is a situation where almost every single manager would leave the starter in no matter how good or bad the starter’s historical numbers are. And usually the manager is wrong. (Probably in this case too. I’m highly skeptical that we can use a list of the 10 best pitchers since 1993 as an accurate prior for 2011 Chris Carpenter.)

As an aside, and totally independent of this particular situation, I do think that before we jump too heavily on the general conclusion that managers, scouts, etc cannot see and know things with their eyes that we don’t, we could test it more thoroughly. I imagine though that selection bias makes it a hard thing to test. I.e., a manager is more likely to play a guy who is “hitting the ball hard” etc. (I doubt mgl or tango would disagree with this latter point though.)

#58 “I know this sounds counter-intuitive, but you can NOT use Carp’s and Motte’s stats, other than to categorize them (such as “elite starter”, “elite releiver”, etc..  Once you can categorize them, then you do back in history and look at large samples of players (and PA’s and TBF’s) with those same characteristics.”

While your argument is very reasonable, it is not as black and white as you make it sound. As you obviously know, this is a standard issue of what is the proper prior to use for your Bayesian analysis. And that’s the problem—you can make all kinds of arguments for why you think one prior is better than another, but there is no single “correct” choice.

And that is the weakness of your entire argument. Almost by definition, TLR’s prior includes much more long-run and short-run specific knowledge about Carpenter and Motte than yours, in which case your over-the-top criticism of the move as “Worst Managing Ever” is completely unwarranted.

On BTF, someone said my post Tango/30 was condescending.  While it can certainly be interpreted as that, and I don’t blame the non-regulars in interpreting that way, I would hope there was enough ambiguity in the tone of the post that the regulars here can cut me some slack on it.

When in doubt, either ignore what I say, or try to look at it in as a positive light as you can.  In return, I’ll give you my best effort, even if it means going to bed at 2 in the morning.

If after all that you really need to swim in snark-infested waters, have at it!

Mike Silva’s take on this thread:

http://www.nybaseballdigest.com/?p=40547

I think it’s too bad that so much attention has gotten focused on the question of whether Carpenter should have been pulled (which is perhaps a 50-50 proposition), when MGL’s underlying point—that managers often leave starters in too long, especially if it means allowing them to hit—is correct and very important.  In game 3 vs. Philly, LaRussa was clearly wrong to allow Garcia to hit in a high-leverage AB in the 5th, just so he could blow the lead in the 6th (as MGL noted in an earlier thread).  And I think LaRussa was wrong as well to allow Garcia to face Braun for the third time with two on yesterday, even though it was only the fifth inning.  (And lest I be accused of Monday-morning-QBing, I was screaming at my TV in real time in both cases).  This isn’t a flaw of LaRussa’s alone, of course—few managers pull starters fast enough in these situations.  Over the past two seasons, hitters’ OPS is highest in the 6th inning (.756, compared to .724 overall), even higher than in the 1st. 

I think fully recognizing the diminished capacity of starters in the middle innnings, compared to a decent reliever, is the area where sabermetric insights are most likely to impact the game over the next few years.  This knowledge is becoming widespread, and eventually it will overwhelm the resistance produced by inertia, tradition, and ego.  We will see:
* Non-elite starters with a lead will be yanked more often in the 6th, and eventually even in the fifth.
* In the postseason, some team will have the guts to use a committee to pitch game 4—letting the #4 starter and decent relievers each pitch 2-3 innings.  (In fact, that’s what LaRussa should have done with Westbrook in game #161—Westbrook should never have been allowed to face an Astros hitter for a second time in a must-win game.)
* Even in the regular season, fifth starters will be allowed to pitch only 3-4 innings, especially if the next day is an off day.  Indeed, the “fifth starter” for some teams will really be a long reliever.

The key hurdle will be getting teams to recognize that you can’t let your non-elite pitcher keep going “until he gets into trouble.” Trouble simply comes too fast in baseball.  When Braun came to the plate yesterday, StL still had a 3-run lead; a couple minutes later, Prince Fielder is fist-bumping his teammates and they’re down 6-5 (and there simply wasn’t time for a reliever to warm up).  If you have a lead and your starter has gone through the lineup twice, you need to be thinking about how and when you can get him out of the game, not looking for ways to keep him in as long as possible. 

GMs will likely figure this out first, and will have to force it on managers.  But I do think it will happen....

It seems kind of weird that on one hand the statheads are bemoaning the giant pitching staffs and then on the other hand advocating looking to pull starters after 18 PA.

One issue I have with looking at pulling pitchers after the second time through is that the top of the order naturally comes first and if you can survive that you then quite obviously have the back end of the lineup.  So if you go into an inning with the 9-1-2 hitters up and you get through it with the #3 or #4 hitter ending the inning you now have the bottom of the order coming up and it is quite possible that you can get almost 2 full innings of pitching more out of your starter by letting him pitch against the bottom of the lineup. 

Do you really want to burn through that many relievers that often?

"It seems kind of weird that on one hand the statheads are bemoaning the giant pitching staffs and then on the other hand advocating looking to pull starters after 18 PA.”

It seems kind of weird to me that someone would expect all “statheads” to agree.

And in any case, I was simply describing what I think 1) is the most effective way to win ballgames, and 2) therefore, the direction I think the game will eventually take.  I made no comment on whether I thought the result would (or would not) produce esthetically pleasing baseball.

I’m making no comment on the aesthetics of the game.

I was commenting that going your route would lead to bigger pitching staffs and less and less innings for each individual pitcher.  You’re talking about a 13 to 14 man pitching staffs and basically two non-C bench guys or three if they go to some sort of hybrid backup catcher/utility man.

No, I’m talking about 12-13 pitchers—the high end of current practice.  Just a question of giving more innings to the pen in some situations. 

McCoy, you note above that “I think it looks like Tony knows when to let Chris Carpenter pitch deep into a game.” Any thoughts on why he doesn’t seem to know when to let Garcia keep pitching?

If MLB allowed 30 players, there’s no question I’d have 16 or 17 pitchers.  Pitchers pitch MUCH better when they know they only need to go 1 or 2 innings.  I’d let maybe 1 or 2 of my guys go 6, but not more than that.

"I think it’s too bad that so much attention has gotten focused on the question of whether Carpenter should have been pulled (which is perhaps a 50-50 proposition), when MGL’s underlying point—that managers often leave starters in too long, especially if it means allowing them to hit—is correct and very important.”

I freely admit I’m a newbie here, and maybe I don’t understand the nuances here, but when I read “Larussa letting carpenter pitch the 9th is the worst managing decision ever” I did not interpret that to mean merely “managers often leave starters in too long”.

If that’s the way you guys roll here, that’s cool, it’s not my sandbox.  But when I and others were focusing on the specific point of should Carpenter pitch the 9th in this specific game, given his specific performance, I didn’t feel like we were derailing any larger point.

I didn’t say anyone “derailed” the discussion. It’s just too bad the valid point is being missed by many people.

"Any thoughts on why he doesn’t seem to know when to let Garcia keep pitching? “

Aren’t we dealing with probabilities here? 

Tony also couldn’t correctly identify a decent reliever to relieve Garcia.  Or it could be that Tony knew there wasn’t a good replacement for Garcia.  Or it could be that Tony simply thought a starting pitcher should go at least 5 innings.  There is a whole host of variables in play here.  Some Tony could be unwisely placing too much importance on while placing too little importance on others.

"I didn’t say anyone “derailed” the discussion. It’s just too bad the valid point is being missed by many people. “

That “valid” point wasn’t missed and in fact was heavily disputed and it turns out that MGL’s blanket statement as you interpreted it was incorrect.  One could argue, based on the data, that it is possible that starters in some games are being pulled too early.

I for one think that is an interesting question.  How many pitchers get through say the 7th or 8th without getting to the lineup a 4th time and then get pulled?  Is it possible that these pitchers could have gone another inning and put up a very good performance in that inning?  And if they get pulled for a pinch hitter how much better does the pinch hitter have to be to offset the possible decrease in pitching ability of going to the reliever (if in fact there is a decrease)?

Tango and MGL,

Since it looks like you have the software all setup to go for this what kind of performance in the 7th and 8th inning lead to a starting pitcher pitching in the 9th?  What are the flags or conditions that we can identify that would lead one to believe going with the starter in the 9th is an okay decision?

For instance if the 8th is a 1-2-3 inning how much more often does that lead to the starter pitching the 9th than say a SP facing 5 batters in the 8th?  How likely are they to come out for the 9th based on pitch count?  Score?  When they bat?  So on and so on.

What I find especially outrageous, almost scandalous, is that someone could actually write, as in #72, that I said, “TLR is a terrible manager because he let Carpenter pitch the 9th, “ when the title of the post was hyperbolic, as titles or headlines often are, and that the last example of several gave of Tony’s mistakes was leaving Carp in to pitch the 9th. I suppose a loose characterization of my post is, “TLR is a terrible manager because (insert anything I happened to mention in the post),” but in my opinion that is classic spin, mischaracterization, taking words out of context, etc in order to launch an ad hominem attack and deflect attention from the issues at hand and is otherwise totally uncalled for.

Here here for Guy #66.  I’ve been trumpeting this for a while now but never said it as well as Guy did!

"That “valid” point wasn’t missed and in fact was heavily disputed and it turns out that MGL’s blanket statement as you interpreted it was incorrect.”

It has not “turned out” to be incorrect.  There is now SOME data suggesting that elite starters MAY be equal to a good closer in the 9th.  My own guess is that even this will not turn out to be true, once we account for HFA, the fact that these starters often have large leads, and perhaps other factors.

But even if there are 10 or 15 starters who are that good when managers think they are “on,” that hardly undermines the larger point that I (and MGL) was making:  that starting pitchers generally decline significantly as they see hitters a 3rd and 4th time, to the extent that they are inferior to several relievers who could enter the game at that point, and that MLB managers tend to let starters stay in too long as a result.  Your comment, ironically, serves as pretty good evidence that I was correct in saying this point is being missed by some people.

"It has not “turned out” to be incorrect.”

Well, it hasn’t been proven correct yet either. 

It is weird how MGL’s point has morphed into this that never got mentioned after MGL specifically stated what is he was saying in his posts.  He quite specifically said to compare Jason Motte to a bunch of elite starters facing the lineup a fourth time through.  Now you wish to move the goal posts to a point that has absolutely nothing to do with the original post.  The point you think I miss is a an almost meaningless point.  The point you wish every one to talk about is that sometimes managers leave pitchers in there too long?  That is the big thing we all need to discuss?  That’s life?  It happens and that point is meaningless in the discussion about the game in question.  That is what you want to hang your hat on?  Sometimes managers leave pitchers in too long therefore we should ignore all other evidence and rail against Tony for leaving Chris in there? 

By the way it isn’t just elite pitchers that pitch really well in the 9th inning.  My own limited sampling is showing that starters are putting up very good numbers in the 9th inning.

“My own guess is that even this will not turn out to be true, once we account for HFA, the fact that these starters often have large leads, and perhaps other factors. “

Which is all a big “so what?”.  Even if it turns out that starting pitcher performs very well in the 9th because they have a large lead, HFA, and martians come down from space and steal the bats it doesn’t somehow mean that starting pitchers shouldn’t pitch the 9th.

"It is weird how MGL’s point has morphed into this that never got mentioned”

It’s not weird at all.  Carpenter is an example.  The underlying point is that managers often fail to pinchhit for starters late in the game when they should, and/or just leave starters in too long.  MGL made the same point about Garcia in game 3 vs. Philly, and in approximately 112 other posts on this site if you’d like to search for them.  MGL may or may not have been wrong in his claim that even great starters aren’t as good as Motte in the 9th, but he is undeniably right about the larger point. 

“The point you wish every one to talk about is that sometimes managers leave pitchers in there too long?”

No, not “sometimes.” We never talk about “sometimes” here.  The point is that managers SYSTEMATICALLY leave starters in too long.  Does that mean managers always do that?  Of course not.  Is it true that managers “sometimes” take starters out too soon (as you suggest above)?  Of course it is.  But “sometimes” is always a trivial observation, and I’m not sure why you’d want to discuss it.  What’s interesting is that managers could improve their performance ON AVERAGE by using the hook more aggressively.

“Even if it turns out that starting pitcher performs very well in the 9th because they have a large lead, HFA, and martians come down from space and steal the bats it doesn’t somehow mean that starting pitchers shouldn’t pitch the 9th.”

It means that the comparisons being done are inaccurate.  If great starters do well in the 9th in part because they often have 5-run leads, then the comparison to closers is biased.  If they do well in part because these games are in pitchers’ parks (are there fewer CG in Arlington and CO?), the comparison is biased.  If they do well because some of the opposing hitters are injured, or just “cold”, that biases the comparison.  What we want to know is how starters and closers would do under identical conditions: home/away, same lead, same offensive performance by hitters in innings 1-8.  It’s not that complicated an idea, though hard to parse out in the data.

You’re comment about “moving the goalposts” is instructive.  You seem to consider this a battle, or game.  Perhaps you are invested in the idea that you “beat” the big bad MGL, and don’t want anyone taking that away?  In any case, none of that interests me.  I’m just interested in baseball strategy.

#80, the authors of this site, these “statheads,” have done their homework.  In theory, this discussion is easy to have.  MGL has a chapter in the Book on this very topic.  The data is there, up for debate.  If you disagree with him, do your own research and show that he’s wrong.  Asking a ton of contradictory questions is no substitute.

Tony pulled Jackson after just 4.1 IP last night. Good move. Maybe he’s reading this thread?

Yes, I may have been wrong about the difference between an elite starter in the 9th who has been pitching extremely well and a closer.  Can we put that to rest and move on?

Regardless of how assured I make myself out to be on various issues, I am sometimes wrong, thankfully.  Let’s all have a party, hand out awards to everyone who doubted me, and move on…

It’s not weird at all.  Carpenter is an example.  The underlying point is that managers often fail to pinchhit for starters late in the game when they should, and/or just leave starters in too long.  MGL made the same point about Garcia in game 3 vs. Philly, and in approximately 112 other posts on this site if you’d like to search for them.

Guy, I think you are being a bit unfair to McCoy with this statement.  You know that MGL was using Carpenter as an example for a larger point, and I know that as well because we spend a lot of time reading this blog.  But there is no way that McCoy could devine that from reading MGL’s “Worst managing ever?” post, and its a little much to expect him to have done background research on MGL’s previous writings before he responds.  There is nothing in the original post or MGL’s bet response in post #2 that makes it seem that MGL is using Carpenter to discuss a larger issue.  And it now seems from MGL’s admission above that if someone had actually taken up MGL on his bet that MGL would have lost.  McCoy’s and PJF’s and Freddie Mac’s responses may not have been perfectly researched and reasoned, but they did help lead to new insights that would not have occurred if no one had questioned the original premise that Carpenter was a worse choice than Motte in the ninth.  And let’s not forget that that premise was not based on any research that MGL had done on the specific situation, but was solely MGL’s opinion based on his “years of baseball research and experience”.  If MGL really wanted to talk about the larger issue of leaving starting pitchers in the game too long as you seem to be claiming in posts #66 and 81, then he chose a particularly bad example to introduce the subject.

Peter, I don’t see how I could possibly have been unfair to McCoy in this regard, since I have never said anything about McCoy’s (or PJF’s or FM’s) responses to MGL’s original post.  And I’ve said not one word about what McCoy should or should not have known at that time about MGL’s intentions or prior commentary.  Re-read my post #66.  McCoy then decided to argue with me, at which point I had already clearly explained what I took to be MGL’s underlying analytical point. 

“And it now seems from MGL’s admission above that if someone had actually taken up MGL on his bet that MGL would have lost.”

I doubt that’s the case. For one thing, MGL specified 4th-time-thru, not “9th inning.” But more importantly, I think there’s good reason to doubt that the 9th inning starter performances being reported here truly indicate that great starters have some special ability to overperform in the 9th.  I’m prepared to believe that 8 innings of superior performance has some predictive value for inning 9—although that’s certainly not been proven—but the magnitude of improvement being reported almost certainly reflects something special about losing teams’ 9th inning hitting strategy, a small sample size fluke, or both.  If we compared starters and relievers under identical conditions—home/away, identical score—and adjust for quality of hitters, I suspect we’d find closers still have an edge.

I doubt that’s the case. For one thing, MGL specified 4th-time-thru, not “9th inning.”

Isn’t that exactly the question that Tango researched here?  Aren’t the parameters exactly those that MGL described for the bet?

http://www.insidethebook.com/ee/index.php/site/comments/how_do_great_starting_pitchers_pitch_the_4th_time_through_the_order/

If we compared starters and relievers under identical conditions—home/away, identical score—and adjust for quality of hitters, I suspect we’d find closers still have an edge.

I understand that the above is your opinion, but you have not provided any data to support that opinion.  Does this not apply to you as it is being applied to others that post on this blog?

I learned a long time ago that my opinions without a solid base aren’t worth jack…

If we focus on 4th time through the order, then it’s just a simple math problem.  Each time through the order has an effect of say +.010 wOBA points.

So, a guy who is a true .290 wOBA through 27 batters would be
.280
.290
.300

each time through the order.  And so, 4th time (notwithstanding any other variable in play) would imply a true talent of .310.

Madson’s career is .300, so we can probably guess his true talent at .310.  Motte has a much shorter career, but for the sake of argument, let’s say he’s also a true .310.

Therefore, a wash.  And that’s without even thinking about the possible 9th inning / hothand effect.

So, it’s possible that MGL simply didn’t think of the math in those terms.  He probably had the elite starter at higher than .290, and he probably had his closer at lower than .310.

Maybe he thought of Carp specifically, etc.

If the gap is 20 wOBA points between your starter and your reliever, then it’s a wash when considering 4th time up.

I should also note that the 4th time through the order effect, of what I’ve compiled, does not show a 10 point jump, but more like 3 or 4 points at most.  And even less under other considerations.  We’ve speculated that this could be do to hothand.  But we’ve never researched this, until the other readers of this site chipped in. 

So, if the hothand does exist, if let’s say it’s worth say -5 or or -10 wOBA points, then it’s even more support for leaving the starter in.

The problem of course is in being able to identify if he has the hothand.

Only 30% of the pitchers who pitch through 8 were actually brought back for the 9th inning. This could be for reasons that the pitcher was not really hot, or it could be that the manager is more worried about overworking the pitcher, or that his relievers can do the job almost as good for a 162-game season.

This is where I want to see run expectancy figures for games of this nature, very low scoring, elite pitchers on the mound, etc.

I don’t want to take league average data and apply it to a situation that is far from league average.

For example, show me elite pitcher’s performance against 1st, 2nd, 3rd, 4th time through the lineup when they are taking a shutout into the last inning. The average data for all pitchers in all situations doesn’t inherently tell me what I wnat to know in regards to this very specific situation does it?

FWIW, I would have left Carpenter in the game, especially when my options are Jason Motte and Arthur Rhodes. We can talk about talent, fatigue, and all that stuff that we estimate, average results. What we know is that Carpetner isn’t going to shit his pants in the 9th inning of an elimination game, combined with the fact that he is absolutely dominating the Phillies. For him to “lose it” to where he becomes very hittable in the span of one half inning or 15 pitches would be more stunning than the bullpen giving up a run.

We continually point out how dumb managers are when they do things that work out even though they are against the “average result”. What does that serve? I don;t think it causes people to realize that managers may be lucky. Isn’t it a amanager’s jhob to decide when this is one of those situations that needs a ‘non-average’ result?

How many elite pitchers have lost shutouts in the 9th inning of game causing their team to tie and/or lose? How many have finished the game with the shutout? How many average or below average relievers in 1-0 games have come in and blown the save?

@#11

I would be primarily concerned with the number of occurences that the leite pitcher gave up a run (or more) in the 9th inning.

If my guy goes 9 times without allowing a run, but then gives up a grand slam in the 10th outing ... I’m not going to look at it as 0.4 runs per 9th inning outing.

I’m going to look at it as “9 out of 10 times, he finishes the game without giving up a run ... and that’s a better rate than my closer.” (or something along those lines).

To me, it’s the same kind of thing as with defense and IBB. The “average” runs scored is meaningless because in many IBB situations if you don;t give up zero runs, then it really doens;t matter if you give up 2 or 6. The point is that you have to get out of that situation/inning without allowing a run so that you can have a realistic chance at coming back. BFD, if your pitcher gives up a bases clearing triple in the 8th while you were losing 5-2. They were probably going to score 2 runs that inning anyway. It’s an all or nothing situation.

In a closer vs. ace in the 9th during a 1-0 game, it’s probably much closer to 50/50 than all or nothing. Considering who the Cards relievers are it might be in Carp’s favor.

My guess is that there are not a lot of occurrences where an elite pitcher is “on” for 8 innings and then blows it in the 9th.

I wonder if Philly wanted Carp to stay in for the 9th or bring in a reliever? My guess is they weren;t thrilled to see him come out in the 9th.

Only 30% of the pitchers who pitch through 8 were actually brought back for the 9th inning. This could be for reasons that the pitcher was not really hot, or it could be that the manager is more worried about overworking the pitcher, or that his relievers can do the job almost as good for a 162-game season.

Or it may be the qualities of the 30% that are left in that are the determining factor.  One possibility that I am beginning to research is whether the starting pitcher is being left in because he has the platoon advantage on the first batter that he is to face in the ninth.

Nevermind.

The THT article regarding “Great Day Carpenter” vs. Motte is exactly what i would prefer to see more often.

The “average result” usage for all situations is the exact opposite of what I prefer, unless you’re dealing with situations that are well within the “average” setting.

Game 5 of the NLCS was not close to an average event (in any way).

While you’re doing all your work to quantify values of elite starters vs closers in 9th innings of close games, be sure to factor in the effects of having said starters run the bases in the previous half-inning. Seeing Carpenter slide into second on that bunt/failed fielder’s choice, my first thought was that he’d be scratched not for a closer, but for a pinch-runner ... and for an uninjured starter in the NLCS (if you catch my point). Was La Russa’s managing the “worst” ever? Perhaps not—“worst” carries heavy connotations in the extreme—but he certainly went very strongly against sabermetric type in the ninth inning of that fifth NLDS game in Philadelphia. Because that game ended up as only the third 1-0 playoff series-ending game in MLB history (an “instant classic,” as ESPN Classic would have it were ESPN the rightsholder on Friday) we now have much `a posteriori reflection that La Russa somehow was playing the probabilities correctly, when in fact he went strongly against them.

Tango/88:  Good point.  The talent gap between the elite starters and the average closer is actually large enough that we should expect them to be roughly equal in the 9th, even leaving aside a possible hot hand effect.  So the decision will often be a wash, except when the pitcher is scheduled to bat, in which case removing the starter may still be the clearly correct move.

The interesting question (to me) is how much weight, if any, should we give to a pitchers’ performance in the first 7 or 8 innings?  If a starter is great in the first 8 innings, should that change our projection of his 9th inning true talent?

"This is where I want to see run expectancy figures for games of this nature, very low scoring, elite pitchers on the mound, etc. “

The Book has the RE matrix for runs per game of 3.20 and of 5.00.

I agree, we need to be able to establish what kind of hothand effect do we want for the 9th inning.  Right now, let’s just say it’s 10 wOBA points.

So, if all pitching things considered told us it was a wash between leaving him in the 9th, or putting in a reliever, a 10 wOBA point hothand effect would be a good tiebreaker.  And with around 4 batters faced in an inning, it’s not inconsequential.

HOWEVER, the gap between a pitcher batting and a guy off the bench is 0.130 difference in wOBA!  That’s 130 point difference, which is enormously huge.

To put it in similar terms: let’s say you think that Carpenter is a 0.45 runs per game improvement over Motte or someone else in the 9th inning.  That’s a gain of 0.05 runs for the inning.

On the other hand, the gap between a pinch hitter and a pitcher batting is 0.10 runs for one plate appearance.  That’s a ridiculously huge gap.

No matter how you slice, Motte for Carp on the pitching side is not going to approach anything like the gap between a PH and a pitcher batting.

Tom, I agree completely on the difference between a P as batter and a PH. I was surprised to see him bat honestly, but not stunned ... was even more surprised when he singled up the middle ... was even more surprised when he executed a take out slide at 2nd to break up the play ... and was even more surprised that he was not injured on the play.

--- Q: By bunting the second time, did that actually help Philly by allowing them to IBB Pujols and thus putting them in a position where 1 ground ball could get them out of a jam with no runs scored. Conversely, if Pujols hits into a DP, then they have runner on 3rd, 2-outs. ---

My thought at the time was that TLR pretty much decided after the 7th that Carp was going to finish the game if the pattern held.

Carp batted in the top of the 8th, so now it’s a “great game Carpenter” being replaced with 2 IP of average or below average relievers, not just 1 IP of Motte.

I think the sac bunts are fine. Not my preference, but not totally against my beliefs in that scenario either. The Berkman K hurt, but I thought initially that Holliday hit a Grand Slam (F-7)

I think if you let Carp bat in the 8th, it’s because you feel he is an overall better option than a PH, RP #1, and RP #2.

I think when the math is done, it’ll come out as being around 50/50, in which case you decide whether or not you want carpenter or Motte/Rhodes deciding the game.

I would not be surprised if the odds actually played in StL’s favor if the 2 choices are:

[1] Carpenter finishing the game as a 1-0 shutout.
[2] StL scoring a run off Halladay using a PH and then the bullpen finishing the shutout.

McCoy made this unfortunate post over at BTF:

“To rant a little bit. . . I’ve seen it a few times now on the Book blog where someone will basically disregard you or think they can ignore your point because you don’t post the numbers. You gotta post the numbers. That would be an excellent guiding principle unfortunately it isn’t applied to everyone. I mean this whole thing started because MGL didn’t run the numbers and then people defended him and they did so without running the numbers. Then the numbers got run and instead of these people opening their mind and thinking about the problem they simply say stuff like the once all the numbers are run he’ll still be right. Yet if you try to show why you think he won’t still be right you get the whole GOTTA SHOW THE NUMBERS spiel.”

It’s hard to imagine a less accurate description of the response here.  Both Tango and MLG have actually done new research in response to those disagreeing with MGL’s original post, and have acknowledged that a starter’s performance in innings 1-8 may have more predictive value for this 9th inning than they previously believed.  The question isn’t settled, and more research should be done (I think Max made a good start), but their thinking on the issue has already changed, by their own admission.  The claim that no one here “opened their mind” is without foundation.

As for “show me the numbers,” well, it depends.  It’s fine to say “I suspect X may be true, and this would be a good topic for someone to research.” But if you say “I believe X is true, although I have no data to support this claim,” then you should expect to be challenged.  It also matters whether your claim is consistent with prior data (and it’s true that the regulars here assume knowledge that a new reader may not have—just the nature of this kind of blog).  MGL didn’t cite it, but he has a mountain of data that shows, in general, pinch hitting for pitchers late in close games is the right move.  In fact, it was likely true even in this Carpenter case.  But he likely went too far when he speculated that Motte would be superior to even the most elite starters in the 9th inning.  So fine:  he engaged in speculation, he was challenged, he turned out to be wrong (probably), and he’s now acknowledged as much.  What’s the problem?  Why all the whining?  I don’t get it.....

Perhaps McCoy’s threshold for being disregarded is very high, and so we disappoint him.

It could also be a sample size issue.  Things move very fast and furious around here, and it’s possible that someone’s post is “disregarded”, when really it’s skipped over unintentionally. 

I mean, I have a full-time job, and I do my best to address just about every post, so things slip, or I just don’t have a good answer, so I’m not going to make up for it with a snark.

All I can say to McCoy is to give this place more of a chance, and get into some sort of paradigm shift.  This place simply operates at a different wavelength.  Not necessarily good or bad, or higher or lower… just different, not better.

And, if you still have a problem, then post here!  (Or email me.  There’s an email link at the top right of every page… “You ask”.)

Making a summary conclusion with selectively-sampled biased evidence is.... well, the regulars around here know the answer to that.

After 2011 NLCS Game 5, TBS showed a graphic saying that STL starters had pitched 22 innings so far in the NLCS, and STL relievers had pitched 21 innings so far (fractions were also involved, sorry I missed that). One of the talking heads even noted that it was as if the Cardinals were using their bullpen in an even more specialized way, relieving their starters as early as the 5th inning, if not the sixth. It is as if La Russa read this thread. Much different from the STL-PHI NLDS.

Tom/97, How does leverage figure in? Leading off Top 8 with a one run lead, the LI is 1. Leading off the bottom of the 8th with a 1 run lead, the LI is 2.5.

You probably read the LI chart wrong. 

If it’s this, this is based on the home team:
http://www.insidethebook.com/li.shtml

If the pitching team is up by 1 in the top of the 8th, LI is 2.2.

This is how I read it;

When Carpenter came to bat in the top of the 8th, the home team was down by 1. I look at top 8, 0 outs, -1, and the LI is 1.0.

Then when Philly comes to bat in the bottom of the 8th, they are still down by 1. I look at bottom 8, 0 outs, -1, and the LI is 2.5.

Is that incorrect?

You read that correctly.  When the Cards come to bat in the top of the 8th, they are up by 1 (pitching team down by 1).  Cards are in the driver’s seat, and have a great chance of winning.  This is why you don’t see closers normally come in when the defense is down by 1 in the 8th.  The LI = 1.0 captures that perfectly.

When the Phillies come to bat in the bottom of the 8th, they need to score one run, and now it’s a whole new game.  Going from down by 1 to tied has more impact than going from up by 1 to up by 2.  So, with the pitching team up by 1 in the bottom of the 8th, they need to shut down the opposing offense.  This is why having a closer in this situation is really valuable.  LI of 2.5 captures that perfectly.

Now that you, me, and everyone else is caught up, what is the question exactly?

In post 97 you come up with a scenario where Carpenter is .05 runs better than Motte and a pinch hitter is .10 runs better than Carpenter. Do we factor in the LI of 2.5 to Carpenter’s runs saved, or had you already accounted for that?

That is a good point.  Tango, isn’t the impact of any difference in pitching in the bottom of the 8th (if that impact is measured in runs) 2.5 times that of the impact of a pinch hitter as compared to Carp (again, in runs) in the top of the inning?

Anyway, here is a summary of an estimate of the difference between Carp and Motte in the 9th.

I don’t think Carp is an “elite” (as in #1 starter) pitcher anymore, as most of the projections will attest.  I ran the numbers using around my 40 best starters for each year, projection-wise, going into that year (IOW, the projections are pre-season projections). That produces starting pitchers who perform at a truly elite level, as you will see.

I do think Motte is an elite reliever, where I define elite simply as a legitimate closer.  I generated the data using my 20 best relievers, by my pre-season projections, just as with the starters.

In order to compare apples to apples, I only used 8th and 9th innings where the pitching team was up by 3 runs, 2 runs, 1 run, or tied. 

Unfortunately, we’re not talking about a lot of PA here, so I doubt that we are going to find a lot of statistically significant differences, and any conclusions we draw about the relative performances of closers and elite starters in the 8th or 9th innings are not going have a high level of certainty because of these small sample sizes.

Starters at home in 8th inning

791 PA.  242 games.  3.27 PA per that 8th inning.

Their WOBA against is .286.

Batters in this inning are 5 points worse than average.

These are .317 pitchers (their overall performance for the year).  The average pitcher is .340.

Starters on the road in 8th inning

649 PA.  205 games.  3.17 PA per that 8th inning.

Their WOBA against is .302.

Batters in this inning are 3 points worse than average.

These are .319 pitchers (their overall performance for the year).  The average pitcher is .340.

Starters at home in 9th inning

226 PA.  68 games.  3.32 PA per that 9th inning.

Their WOBA against is .272.

Batters in this inning are 3 points worse than average.

These are .313 pitchers (their overall performance for the year).  The average pitcher is .340.

Starters on the road in 9th inning

192 PA.  59 games.  3.25 PA per that 9th inning.

Their WOBA against is .284.

Batters in this inning are 1 point better than average.

These are .316 pitchers (their overall performance for the year).  The average pitcher is .340.

Relievers at home in 8th inning

1167 PA.  387 games.  3.02 PA per that 8th inning.

Their WOBA against is .299.

Batters in this inning are 1 points worse than average.

These are .289 pitchers (their overall performance for the year).  The average pitcher is .340.

Relievers on the road in 8th inning

1259 PA.  420 games.  3.00 PA per that 8th inning.

Their WOBA against is .309.

Batters in this inning are 1 points worse than average.

These are .289 pitchers (their overall performance for the year).  The average pitcher is .340.

Relievers at home in 9th inning

4581 PA. 1224 games.  3.74 PA per that 9th inning.

Their WOBA against is .282.

Batters in this inning are 1 points worse than average.

These are .281 pitchers (their overall performance for the year).  The average pitcher is .340.

Relievers on the road in 9th inning

3625 PA.  953 games.  3.25 PA per that 9th inning.

Their WOBA against is .290.

Batters in this inning are 3 points worse than average.

These are .282 pitchers (their overall performance for the year).  The average pitcher is .340.

If we combine home and road, and adjust for batters, we get this:

8th inning starters

They pitch 20 points better than their seasonal overall stats, or around .298.

9th inning starters

They pitch 35.5 points better than their seasonal overall stats or around .279.

8th inning relievers

They pitch 16 points worse than their seasonal overall stats, or around .305.

9th inning relievers

They pitch 7 points worse than their seasonal overall stats or around .288.

So, it appears that 9th inning starters pitched 9 points better than 9th inning relievers when the pitching team was tied or leading by no more than 3 runs.  (In the 8th, the starters pitched 7 points better.)

As I said, at the outset, because of the small sample sizes, principally only the 418 PA for these starters in the 9th, one standard error of the difference is around 25 points in wOBA, so a difference of 9 points, while indicative of something (absent any a priori probabilities), it doesn’t really allow us any reasonable level of certainty with respect to who is better in the 9th.

In fact, if we use a 95% confidence interval, we might conclude that there is a 95% chance that the true difference is anywhere from 41 points in favor of the relievers to 59 points in favor of the starters.

I am not trying to waffle on my original declaration. I already stated that I may be wrong.  I am just presenting the data and statistical inferences relating to that data.

rwperu: excellent point!  I don’t know how I didn’t consider that.

So, the PH thing could be a wash, depending on how much better you think Carp is over Motte.

rwperu is pointing out, based on Leverage Index, that there is nothing to leverage with a PH if you are the leading team, but there IS something to leverage with a pitcher pitching as the leading team.

If the PH provided +.10 runs per PA of value, then it is +.01 wins of value.

On the other hand, if Carp is +.36 runs per game (+.04 runs per inning) better than Motte, then the unleveraged value is +.004 wins per random inning, or +.010 wins in the inning with an LI of 2.5.

That becomes the breakeven point: if you think Carp is +.36 RA9 better than Motte, the PH thing is not an issue.  If he’s not that good, then the PH thing starts to become a variable.

if you think Carp is +.36 RA9 better than Motte, the PH thing is not an issue.

On that night, in that place, against those hitters ... Carpenter might have been that much better than Scrabble, Dotel, and Motte.

It’s a lot closer when viewing just Carp v. Motte for the 9th. Considering Motte’s limited pitch selection, and facing two lefties with power, including the latter that can’t hit breaking stuff, Carp could have been the better choice (all things considered).

---------------------------------

Aside from the data reseach, did anyone watching the game really feel as though either pitcher was going to give up a run after it got past the middle innings? I’m not saying that as a smart-ass, but rather a pitching fanatic that saw two guys essentially toying with very good major league hitters.

I can’t stand smart-guys like Circle say things like this:

Aside from the data reseach, did anyone watching the game really feel as though either pitcher was going to give up a run after it got past the middle innings? I’m not saying that as a smart-ass, but rather a pitching fanatic that saw two guys essentially toying with very good major league hitters.

I would love, Circle, for you to document every time a starting pitcher enters the 8th inning, and based on your scouting, you will assert that this pitcher is going to have better than expected results.

You will be better than a pitching coach, because the pitching coach that allows his starting pitcher to enter the 8th ends up with results EXACTLY AS EXPECTED without any special knowledge.  How can you ask a question “aside from the research”?  The research is our prior!  It’s the basis of our feelings.

I know someone like this, who makes predictions about EVERYTHING.  And, then a couple of them occur, and it was like “see?  I told you”.  Yeah, I’m told about all the times the person wins, but what about all the time the person loses?

How about a MLB hitter that says “I saw the ball as big as a beachball” on a day they struck out 4 times.  That NEVER gets said.  Only if they go 4-4 do they say it.

All of this is just after-the-fact bullsh!tting.  There’s a million places around the web for that.  Please, don’t make this place a million and one.

You’re better than this Circle.

Right, the research suggests that managers and coaches who are at the game and have watched every pitch intently and their team’s success and ultimately thir jobs hang on every decision they make and they know these guys better than anyone other than perhaps their mother and wife, can’t tell if they are “on” that particular day, but circle, you can tell?

We are all in agreement that managers generally elect to use their starters in the late innings because they are pitching great so far (and the data show that this is absolutely true) and presumably their manager thinks they have not just been lucky and presumably they don’t often have a high pitch count?

Yet the data also shows that the manager is wrong assuming that the manager thinks that the exceptional pitching will continue.  Those pitchers pitch no better in the later innings than we would expect based on their normal everyday stats and not withstanding the fact that they have thus far pitched great.  That’s what the data shows!

That’s not to say that there aren’t some pitchers or class of pitchers whose early inning success does not have some predictive value, but according to the data, managers and coaches do not appear to have much if any ability to identify them or at least to leverage such knowledge. 

So, Circle, are you kidding us?

If the PH provided +.10 runs per PA of value, then it is +.01 wins of value. On the other hand, if Carp is +.36 runs per game (+.04 runs per inning) better than Motte, then the unleveraged value is +.004 wins per random inning, or +.010 wins in the inning with an LI of 2.5.

Except....if your pinch hitter succeeds, and you pad your lead by (say) one run, then you also reduce the leverage you face when your reliever enters.  The two LIs are not independent.

You basically have to use the average expected LI in the 9th inning in the analysis.

Again, my criticism of not using the pinch hitter and even pitching Carp rather than Motte in the 9th were but one of 4 or 5 moved I did not like.  Each one, by itself, is not that big a deal, and the Carp/Motte thing was literally an afterthought.  I did not consider (or even know) the leverage difference between the 8th and 9th innings, which was a mistake on my part.  But you know, if you make a mistake, as we all do, and admit it, and correct yourself, it means nothing to the many idiots on BBTF and the few of them on this blog.  Ironically, you get called all-knowing, arrogant, insufferable, etc., no matter how often you admit to a mistake (which I do all the time).  But that is par for the course for those delusional nitwits…

Right, the research suggests that managers and coaches who are at the game and have watched every pitch intently and their team’s success and ultimately their jobs hang on every decision they make and they know these guys better than anyone other than perhaps their mother and wife, can’t tell if they are “on” that particular day, but circle, you can tell?

Managers can NEVER tell? Or they can’t ALWAYS tell? Or they can;t tell most of the time?

What? You’ve never been ever to tell? Never? Ever?

Does being able to tell in one instance mean that you can tell in every instance or even most instances?

To put it simply and straight-forward ... I thought it was obvious. Location & Command and MLB batters looking foolish, and continuing to look foolish the 3rd time through the lineup.

In the 8th, Carp had to basically set down the bottom of the lineup, and then I figured Carp had to get Utley out and then would own Pence, and Howard ... well, he didn’t stand a chance, not in his current state. 

You seriously could not tell that Carp was hitting his spots with good movement a larger than average % of the time? Or that good hitters were looking foolish and/or making poor contact consistently? ... and it wasn’t getting any easier for the batters as the innings pass by? To me, it was obvious. It was more than just “Garsh, that Carpenter fella seems to be throwing pretty well tonight ain’t he?”

I’d bet that Philly was praying to see Rhodes or Motte. I bet they didn’t say “awesome! starting pitchers don’t do as well the 4th time through the lineup.” when they saw Carpenter come to the plate in the 8th.

That’s what the data shows!

I guess this is where I somewhat divert from the sabermetric norm on occasion. When I see “the data”, I see an average. That means there will be games above and below the average. The data, or really that means “the average” is just that, the average.

I’m saying it seemed obvious to me that this was one of the “well above the average” games for Carpenter, perhaps his best location/command game ever. Given his pitch efficiency, the only “warning light” was that of the average data. One just had to decide if this was one of those “average” situations or if this was special.

I felt it was special, and rather obviously so.

Likewise, you could tell in the first inning of the next start against MIL that it was a “below average” game for Carpenter. Everything was up and away, and breaking pitches were in the dirt as he was flying open and falling toward the 1st base line more than usual. He was “off”. It happens, and it’s usually not something you can “fix” in the middle of a game. I would not have expected him to put up the “norm” or “average” Chris carpenter start that day.

I would not consider myself a pitching genius. IMHO, and not intentionally trying to be a smart-ass, but if one could not tell that Carp was not going to give up any more runs after the 7th, then that says more about that person than it does me.

When good hitters look like idiots at the plate late in the game, you know the pitcher’s performance is not slipping. 3rd time through the lineup and they’re still looking bad. To me, it was an indication of the pitcher and not just good luck on BIP.

I’m comfortable enough in my own skin say this while being presented with the word “data” or even be mocked into making it seem like I am saying that I can ALWAYS tell, just because I could tell one time. Again, my tone is polite, as if we were discussing over a steak and a beer. I don’t want anyone thinking that I am bashing them or talking as if I’m IT in regards to pitching performances.

Observing good, bad, great, terrible, etc., pitching is not the issue.  Whether that has any predictive value is THE issue.

The thing about “averages” is this, and this is an important concept so please read it carefully:

If managers were able to tell whether a starter is “on” and thus will continue to be on, even some of the time, then it WILL show up in the averages.  Same thing for hot and cold players.  Same thing for clutch.

If we find NO differences in the averages, that necessarily means that there are FEW if ANY individuals where the effect can be found.

So if we find that all starters on the whole pitch around the same in the 9th inning as any other times, despite having pitched brilliantly in 1-8 AND despite being allowed to pitch the 9th, then:

That means that it is a rare thing for a pitcher who pitches brilliantly in 1-8 to have that brilliance continue (true talent wise) AND it is a rare thing for a manager to be able to identify it.  Because if a manager could identify it 10% of the time (and the other 90% of the time, he knows that the pitcher is not really on, despite pitching brilliantly in 1-8), then why is he letting those other 90% pitch the 9th when he likely has lots of relievers who are better?

I don’t think I am explaining this real well other than to say (and repeat) that mathematically when the averages indicate no “splits” that means that there are virtually no people in our sample who have any true splits and/or that there are some or even many people in our sample who do have those splits but the split is very small.

So if you say, Circle, that there are some pitchers who we can identify that truly have great stuff that day, and that will continue into the 9th, then our entire sample will show some splits!  If it doesn’t, then those “some pitchers” are very small or if they are large, the effect is very small.

That is why we say, “We don’t know that there are NO clutch hitters or that there has never been a hitter that has truly been hot or cold (with predictive value).” We can only say, “That it is virtually impossible to identify these players and if we can, they are going to be VERY few and far between.”

If you truly think that there are instances like Carp in this game, where are those pitchers/instances in our sample?  And why are managers letting so many starters pitch the 9th, even in close games, when they are not “on?”

Bottom line is that you can opine all you want about how you, a manager, or anyone else can identify such “onness” and more importantly, that you somehow know that it will continue (why can’t a pitcher have great command, etc., just by sheer luck for 8 innings?), but without ANY data whatsoever, and without you being an expert in such things (despite knowing a lot about pitching, which you clearly do), I see know reason to take those opinions seriously, no offense of course.

We say all the time on this blog, and it is a good concept to teach your children, “Opinions without evidence and support using the scientific method is B.S.”

You can ask 100 intelligent people about just about anything and you will get about 37 opinions.  That is even true with so-called experts.  They can’t all be true, can they?  How do we decide what is true?  Evidence and the scientific method, man!

I don’t expect anyone to believe anything I say, without evidence!  Why should I expect anything different from you?

To put it more tersely:

100 times, a pitcher enters the 8th inning, and of those times:
20 times Circle can spot an “on” pitcher
75 times Circle has no opinion
5 times Circle is confused that a “tired” pitcher is allowed to enter the 8th

Therefore, the AVERAGE must show something better than expected.

Right, in order for the average to be equal to expected, as it is, you would have to be wrong just as often as you are right, or simply have no idea at all…

Whether that has any predictive value is THE issue.

That’s a great point.

I don’t think the data presently avbailable is all that high quality nor should it be expected to be highly predictive. I do think that given large enough samples of data that the “important inputs” (so to speak; the causes) will show up eventually in the outcome data (the effect). So, in that regard it’s possible that we do have the data we need, just that I’m not dsmart enough to recognize the fact.

If we were to have access to “hit the mitt” data in the form of a pitching chart showing the strike zone within 20 squares and an additional 15 squares outside the strikezone, and we could compare the HTT%, like we’re doing with the other linescore items ... we might find certain items that are more pitcher-specific factors are highly predictive. If we could combine that with pitch movement, and other aspects, we might find some data is highly predictable.

Now, if you showed me data that demonstrated that Carp wasn’t really hitting his spots more often than usual, and his pitches weren’t really more spectacular than his usual stuff and it was just my erroneous observations that thought that, then I’d giggle like a school girl, blush ... and then change my opinion.

If the data showed me that essentially all pitchers that had shutouts through the 7th were in the same small range of “hitting the target %” (or whatever combination we’re using) and yet half of them lost the shutout while the other half kept it, I’d change my opinion.

If you could demonstrate to me that the outcomes of these situations should reasonably be expected to include the data that I am discussing, then I would have to change my opinion.

I don’t expect anyone to believe anything I say, without evidence!  Why should I expect anything different from you?

You shouldn’t. One could just tell me “You’re fulla shit”, and the best I could say is “You might be right.”

why can’t a pitcher have great command, etc., just by sheer luck for 8 innings?)

Based on my experience, I cannot imagine that happening ... for the same reason I don’t expect me to hit 20 consecutive 3-pointers tonight while I’m playing basketball with my son, or that Vince Young could throw 30 passes right on the money.

Note: I am not requiring you to “change my opinion”. Please don’t take it that way. I was simply using those examples of how one would need to change their opinion when it’s incongruent with the data.

Really, I was just saying that the data you are using in your samples is not the same as the data that I was referring to in terms of what was being measured.

We don’t have “target” information, and certainly, we need to have several years of data anyway.

So, why not give us something more reasonable as a proxy.  What if you say something like every batter he faced the third time through the order, he got out? 

Or, if we assign a number to each of those 9 batters he faced something like this:

We’ll just make up a fun stat called Hammer. Give the pitcher one point for every double, triple, walk, hit batter, wild pitch, or passed ball. Give him two points for a home run, and half a point for a single. Take off one point for a strikeout. Add all that up. That’s a pitcher’s Hammer points. The higher the number, the more he’s been hammered, of course.

And anyone with negative two Hammer points or less, we’ll call this pitcher “on”.

As it so happens, I ran this very study in The Book, and there IS a “on” effect!

In this sample, we’ve got 789 pitchers. After that point, they faced 2086 batters, with a wOBA of .321. Now, it starts to get interesting. Based on the talent levels of the pitchers and batters involved, and the batter advantage later in the game, our expectation is a wOBA of .350, and our actual result is 29 points better than that! At 2086 PA, one standard deviation is 11 points. So, at 2.6 standard deviations from the mean, we are getting significant results.

The Book Says:
If a pitcher is still getting outs late in the game, keep him in there; he may have a bit more left in the tank.

What if you say something like every batter he faced the third time through the order, he got out?

Initially I was thinking “No” because not all hits and outs are an equal indication of pitch quality, but over a large enough sample it should reflect the pitch quality.

As it so happens, I ran this very study in The Book, and there IS a “on” effect!

Thanks. I keep The Book and Between The Numbers handy (as well as james’s big green book), and look at them regularly. I’ll check it.

@Tom #111

You will be better than a pitching coach, because the pitching coach that allows his starting pitcher to enter the 8th ends up with results EXACTLY AS EXPECTED without any special knowledge.  How can you ask a question “aside from the research”?  The research is our prior!  It’s the basis of our feelings.

I must have missed this post earlier.

This quote resonates, and I get it.

The data indicates that managers as a group have no special ability to predict whether or not a pitcher, who is pitching very well, will continue to do so in the late innings.  There may not be enough data to see if certain pitchers, or even certain types of pitchers are significantly more predictable.  But I think there is enough data to see if certain managers/pitching coaches have this ability more consistently than other managers/pitching coaches.  After all TLR/Duncan have been together for decades.

"But I think there is enough data to see if certain managers/pitching coaches have this ability more consistently than other managers/pitching coaches.”

KDS, first, if there were some managers/coaches who could do this, then we would likely see some effect when looking at all the numbers.

Two, once you start splitting the already small samples into even smaller subsets, you will not be able to distinguish signal from noise.  There will inevitably be lots of random variation among the teams/managers/coaches.

Basically when we do this kind of research, we do this:  First we see if anything shows up in all the data. If it does, we try and see how much is “real” (skill) and how much is luck.  If we can do that, then we apply the appropriate regressions to any one or group of players (or managers/coaches perhaps in this case) to estimate their skill.

Unfortunately if we find little or nothing in the aggregate, the inquiry is done. It doesn’t do any good to try and find individual players or groups that might have some skill, for the reasons stated above (can’t separate the signal from the noise).

Basically, as Tango and I said in the other thread, if you find nothing in the aggregate, that means that no skill exists or that it is very small (and practically unidentifiable) or there are a very few players that have significant skill and it is probably impossible to identify them.

http://www.washingtonpost.com/national/on-leadership/how-cardinals-manager-tony-la-russa-rewrote-the-book/2011/10/18/gIQAjwf8uL_story.html

The author of the linked piece asserts that La Russa’s pitcher management is “rewriting” the book vis-`a-vis the 2011 NLCS. Ironic, since what he did in the 2011 NLCS seems to be reinforced, generally speaking, by THE BOOK.

I don’t know that Larussa/Eck started this.  I remember Tom Henke being lights out and 1-inning only.  I haven’t really checked, but I’d say it started a year or three before Eck.  Reardon, Gossage, Sutter all had their patterns changed before Eck.  I’m sure our Reliever historians can chime in here with actual data.

And yes, playoff manager manage differently with starters/relief and the quick hook.

This is not specific to Larussa, or this year.  I seem to remember looking one year that Mariano Rivera’s innings total was half those of Andy Pettitte in the post-season (when both were with the Yanks).  Normally, it should be closer to one-third. 

Given that Pettitte in the playoffs would probably start 25% of the games (rather than 20%), there’d be even more bias to drop that one-third down to under 30%.  But just as off-days helps Pettitte get more starts, then off-days gets Rivera more games.

Again, just going off memory, and whatever else is in my head.  Someone could write a good research piece about Pettitte and Rivera, seeing they overlap so completely.  (And I think we talked about this a few years ago too.)

Larussa I think was a more stronger push for the setup guy being more well-defined.  I’d bet you’ll see the pairing of relievers, and specific to 8th/9th, more with Larussa than someone else.

This whole lauding of managers when their team wins is sickening to me (nothing would be said of LaRussa’s bullpen use had they lost either of those series, OR, he would be criticized).  There is almost zero correlation between a manager’s strategy, good or bad, and the winning of a series or even the w/l record of an entire season.  And even if there were a decent correlation for an entire season (like .05 or something like that), it would be masked to the human eye, by the random flucs and by the talent of the team having nothing to do with the manager…

Back 40 or so years ago, (sigh), my father got a new calculator that; A) generated psuedo-random numbers, and B) given a set of (x,y)points, would generate a least squares regression line.  So of course I “randomly” generated points and checked out the function it generated.  Didn’t know then that I was intraining to be a sports commentator.  Where did I go wrong?  (Too much math and stats?)

"This whole lauding of managers when their team wins is sickening to me (nothing would be said of LaRussa’s bullpen use had they lost either of those series, OR, he would be criticized).” As we are now seeing now that Game 2 is in the books.