THE BOOK--Playing The Percentages In Baseball

Uh, wouldn’t that HAVE TO mean that Wake was a better pitcher than those guys?

***

Well, technically it would mean that Wake has to have a better chance of pitching a great game. I suspect there is a higher variance in the outcome of Wakefield’s starts than there is for most pitchers, but it’s certainly not so great as for him to have better odds of shutting down the Tigers than Beckett or Dice-K.

"The knuckleball messes up your timing so bad it can put you in a slump for three or four games.”

Joe Morgan said that during his playing days, and I assume he believes it, so he is probably still saying it on air.  That is probably the source, but there is also:

“I work for three weeks to get my swing down pat and Phil (Niekro) messes it up in one night.” - Pete Rose

“He simply destroys your timing with that knuckleball. It comes flying in there dipping and hopping like crazy and you just can’t hit it.” - Ernie Banks

“I never worry about it. I just take my three swings and go sit on the bench. I’m afraid if I ever think about hitting it, I’ll mess up my swing for life.” - Dick Allen

Then, according to Banks and Allen, any pitcher with a good knuckleball (I am assuming they are only referring to a good one) would be better than Gibson or Koufax, otherwise their statements could be applied to ANY good pitcher, especially a high K pitcher, which knuckleballers tend NOT to be, I think.  In fact, their statements are an even better example of nonsensical ones that should be obvious if you think about it for more than 2 seconds.

If Rose and Morgan are correct, then we should see a drop-off (worse than average) in production for at least a few days after a team faces a knuckler, which would make a knuckler a lot more valuable than just their own value when they pitch, which would mean that there should be a lot more knuckle ball pitchers in the majors and they should get paid a lot of money.

How about sight unseen I challenge Rose, Morgan, and anyone else who believes that facing a knuckler makes you worse for a few games after that, that we see NO drop-off in offense games following a knuckler?  I’ll even give them 2-1 odds, and they can bet whatever they want.  Heck, Rose should take that bet just for the action! 

We can simply go to the RS logs and look at all games following a knuckle ball pitcher.  Easy to test.

And yes, David, I agree that having high variance might qualify as the requisite for that statement.  I would also bet you that if Wake has unusual variance, it is NOT significantly higher than the average pitcher.  You are welcome to look at his variance in RA per game or something like that and compare it to the average pitcher.

Yes, a knuckle ball is hard to hit, since ANY pitch an MLB pitcher throws HAS to be hard to hit.  But so is a 95 mph heater or an 87 mph slider or a great curve, or a good changeup, etc., etc.

So really Allen’s and Bank’s statements should apply to ANY good major league pitch.

Is it possible to look at how people who relieve Wakefield do, versus their “usual” expectation against those same hitters?

I know these ballplayers are talking about being off for days after a knuckleballer, but intuitively I’d think that pregame BP is sort of a nice reset button… something that’s very standardized from game to game.

On the other hand, if you faced Wakefield 3 times already, and he gets relieved by Manny Delcarmen, does Manny start to look like Zumaya out there because just 30-45 minutes ago, you were waiting on a 68mph pitch?

I like Mike’s suggestion alot.  At the least, it’s easier to program.  Secondly, it’s the exact same context.  Thirdly, the dropoff effect should happen within an hour alot faster than within a day.

IF you were to look at a multi-day effect, I think you’d have to look at the series before and after a team plays the Red Sox.  Wakefield has usually been the team’s #3 or 4 starter, so the guy after him is not going to be very good, but the guy 1-2 games in front of him will often be Pedro, Schilling, or Beckett.

MGL, as to #3, I think a knuckleball could be a different case instead of any good major league pitch.  It messes up your timing.  You are going to see 95 mph fastballs and 87 mph sliders every day.  If not from the starter, then from one of the relievers.

I don’t think the timing is any different facing Ervin Santana’s 97 mph fastball than facing Josh Beckett’s.  Just that Beckett will throw his with better movement and/or location and put you back on the bench more often.

Is this a WOWY-type analysis--e.g. Grady Sizemore v. Schilling on days after Sizemore faced Wakefield versus on days after Sizemore didn’t face Wakefield?

The knuckleball is effective partly for the same reason the wishbone offense works in college football: it’s very rare. If everyone threw the knuckleball, then it wouldn’t be nearly as difficult for major leaguers to hit it. So I don’t think it should be assumed that what applies to conventional hitters necessarily applies to knuckle ballers or any novelty pitching style.

However, I can’t imagine a particular hitter’s timing being messed up for more than one or two PA’s after facing a knuckleballer. A three or four game hangover affect sounds like hyperbole.

Right, so at the very least, let’s at least look at how Wafefield’s relievers fare, compared to when they face those very same hitters after some other starter.  A definite WOWY study…

#2 - Do any hitters who weren’t very good think knuckleballers could get them into a slump? If only good hitters are effected, then we would need a different test.

That having been said, I doubt there is a knuckleball hangover effect. But if indeed it is the good hitters who think a knuckleballer can really mess them up, then I think that might be support against their own argument.

It makes intuitive sense that the best hitters would be the ones saying this. Good hitters have fewer proper slumps (as opposed to relative slumps), so a stretch of poor hitting is probably more likely to register as abnormal. If it coincides with facing a knuckleballer, we shouldn’t be surprised for the hitter to ‘blame’ the pitcher because they can’t think of what they themselves are doing wrong. Since they probably are not doing anything wrong and are just in a results slump, rather than a performance slump, they can chalk it up to Hough or whoever. And since they may not be keeping records or properly analyzing their records, they have no reason to assume that knuckleballers’ ability to induce slumps is less evident than they’ve inductively reasoned it to be. Even if we can show that there is no phenomenon of knuckleballers having greater ability or likelihood to induce slumps, it seems reasonable to expect many top hitters to feel this way.

Tom, players (and ex-players and observers) think dozens of things are true when we know they are not, right?

You don’t have to do a WOWY (I say, WWOY, or “with an without you").  Just look at how batters do and adjust for the pool of pitchers faced, as we usually do with these types of studies.

Rally, yes there is a different “reason” why you don’t hit a knuckle ball well, but so what?  You don’t take “3 cuts and then sit down on the bench” any more than you do against any other pitch.

In fact, quite the opposite!  The power of the knuckleball is that your BIP are a little weaker!  So, a more accurate statement is, “Yeah, I don’t try and think against a knuckle ball. I just take my hacks, make contact a little more often than against other pitches, and then ever so slightly more than usual, make out when I do make contact.”

I’ve been piddling around with Wakefield’s game logs from last year when he was a league average pitcher.  This is ridiculously preliminary, but it appears there may be some short term ‘hangover’ effect.  Through the All Star break in 2007, relievers entering the game after Wakefield started had a collective 1.40 ERA.  This does not control for opponent or park.  More to come ...

Well I tried to post but I think using the words “hang” and “over” together made it appear to be spam.  So ...

I looked at 2007 game logs where Tim Wakefield was the starting pitcher.  His relievers collectively allowed 22 earned runs in 82 inninings for a 2.41 ERA.  Here is a list of the relievers and there numbers in games Wake started or did not:

Eliminating Julian Tavarez because his starting innings overwhelm his relief appearances behind a knuckler (and more importantly because I’m not smart enough to figure out a good way to weigh each of these pitchers), here is the collective performance:

That ERA edge appears substantial, but I do not know if it is statistically significant.

I did not track extra base hits, but apparently that is where Wake’s relievers pick up an edge as their hits, walks, and K’s per nine all are worse when pitching after the knuckleballer than otherwise.

Conclusion?  Well this is far too limited a sample to prove anything and does not account for opponents or parks.  But it is a data point that would suggest that there is some type of short term lingering effect from facing a knuckleballer, as much as a full run off the ERA in this limited survey.

edit: I tried several different ways to make the columns line up, but couldn’t figure it out.  Sorry.

#11 - Do any hitters who weren’t very good think knuckleballers could get them into a slump?

No idea, I doubt they get asked the question much.
--------------
So, a more accurate statement is, “Yeah, I don’t try and think against a knuckle ball. I just take my hacks, make contact a little more often than against other pitches, and then ever so slightly more than usual, make out when I do make contact.”

I am not disputing that, but from Allen’s point of view, here are his stats (from B-R) against knuckleballers that I could find with more than 10 PA.

career Dick Allen .292/.378/.534 (21.3% K)
vs.

Joe Niekro is a special case.
Pre-knuckleball:

w/ knuckleball:

He hit Ken Johnson extremely well (.400/.455/.600 in 55 PA), but he only threw 1/3 knuckleballs according to http://www.oddball-mall.com/knuckleball/list.htm

I am certainly not disputing anyone’s claim that they have done poorly against a certain pitcher or type of pitcher.  Just the general comments like this and many others by announcers (and players, managers, etc.) that no one questions but are clearly nonsense once any reasonably intelligent person takes more than a second to think about them.

I was referring more to at the “pitchers stopping their delivery” comments that you hear every time it happens, than to the knuckleball comments.  Then again, if you hear any baseball insider talk about the knuckle ball, you would think that any knuckle ball pitcher must have an ERA of around 1.00, since it seems to be “nearly impossible to hit,” according to them (obviously it isn’t any harder to hit than any other major league pitch).

Speaking of knuckle ball pitchers, I’ve seen batters take 3-0 on Wakefield and he throws them a 72 mph fastball.  I don’t know how often he throws a fastball on 3-0, but it seems like if you were ever going to hit on 3-0 it would be against a knuckler who throws a fastball a lot on 3-0, especially when the walk is not that valuable, such as with a base open or 2 outs and the bases not loaded.  Even if you know or suspect that you are going to get a fastball near the middle of the plate, the difference between a 92 mph one and a 72 mph one is enormous.

Back to the original thought, it bothers me when announcers over-praise a player from the past.  To listen to baseball announcers, any player who is brought up who once played was terrific, outstanding, excellent, etc. 

The most-recent example I heard was during the Brewers’ game yesterday when Jason Kendall came up to bat and one of the announcers mentioned that his father was Fred Kendall, who “was an outstanding catcher.” If they’re setting the bar for “outstanding” at Fred Kendall, what superlative is left for one of his contemporaries, like, say, Johnny Bench?

BR, no doubt.  Announcers should all just remind us once, that which we already know, which is that ALL MLB players are fantastic baseball players, as compared to the general population.  Now they can go ahead and tell is which players suck, as compared to all MLB players, and which players do not.  As opposed to everyone is dangerous, has a little pop, is a professional hitter, knows how to play the game right, or is as tough as nails.

Here is some interesting data:

I looked at batter performance versus relievers in games that Wakefield started, from 1998-2007.

The hypothesis of course is that the batters’ timing is messed up from facing him for several PA such that they perform worse than expected against “regular” (no knuckle balls) relievers for the remainder of the game.

I looked at all game in which Wakefield started, regardless of how long he lasted, and I looked at all batters’ performance against his relievers regardless of whether they batted against him or not (perhaps they were a late game pinch hitter or defensive replacement).

First, here is the actual batting line of all batters when they batted against these relievers, all per 500 PA, IBB and sac bunts excluded from PA:

s 69.5
d 24.1
t 1.80
hr 10.8
bb+hp 46.3
so 92.9

PA=2,784

If you are not familiar with batting lines per 500 PA, this is NOT a good batting line.

The wOBA is .309 (1 SD=9 points, if wOBA were a binomial).  League average wOBA for 98-07 in the AL is .343.

Now of course, the batters could have been bad because the Boston relievers are good or for some other reason.

I looked at the batting line for all opponents against Boston relievers when someone else started, besides Wakefield.  Sort of a WOWY.  The pool of batters and pitchers should be approximately the same.  Here it is:

s 76.3
d 24.0
t 2.40
hr 11.9
bb+hp 43.3
so 88.9

The wOBA is .322, still quite a bit less than league average, but a lot more than against Wakefield’s relievers.

When facing relievers after facing Wakefield, batters hit a lot worse, walked more, and struck out a lot more, than when they faced relievers after facing another Boston starter, other than Wakefield.

Let’s check the pool of batters and pitchers in both situations, after facing Wakefield and after facing another starter on Boston:

When initially facing Wakefield, here is the overall (3-year) batting line of all batters:

81.8 25.0 2.51 15.9 46.9 78.6

That is a wOBA of .356 (all wOBA’s include ROE’s).

So, for some reason, they are very good overall batters, maybe simply because all batters who face relievers are good because they include pinch hitters and exclude poor hitters who get removed for pinch hitters.  I am not sure.

Remember that the above batting line is NOT how they hit against those relievers after facing Wakefield, but their 3-year overall batting line, against all pitchers.  IOW, it is their “true talent” batting line.

How about the overall “true talent” (3-year) batting line of all batters who faced relievers when the starter was another Boston pitcher, not Wakefield?

81.5 24.8 2.42 16.0 47.6 78.5

wOBA = .356

So these batters, as expected, and as hoped, were identically talented, and in fact, the individual components are almost identical too.

How about the pool of relievers after Wake and the pool of relievers after another Boston starter?  Let’s look at their “true talent,” overall (3-year) pitching lines:

Wake relievers:

78.3 21.5 2.17 12.4 44.9 95.1

wOBA=.323

As expected, these are very good pitchers, 20 points better than the average AL pitcher.

Non-Wake Boston relievers:

77.2 22.5 2.12 13.2 44.7 91.0

wOBA=.328

These pitchers are a little worse than the Wake relievers, but they are very good nonetheless as compared to the average AL pitcher.

So even if we adjusted the batting line against Wake relievers to account for the fact the relievers are a little better, we get these two lines, adjusted for pitchers faced:

Batters facing Wake’s relievers, actual batting line, scaled to the relievers who followed some other starter:

68.5 25.2 1.76 11.5 46.1 88.9

wOBA=.313

Batters facing another Boston starter’s relievers, actual batting line:

76.3 24.0 2.40 11.9 43.3 88.9

wOBA=.322

So, they do hit 9 points worse overall after adjusting for a slightly different pool of pitchers.  Basically they walk more and hit less after facing Wakefield.  They strike out about the same.  The initial difference in K rate was apparently due to the difference in K rate between the relievers who follow Wake and those who follow another starter. I guess they are a little reluctant to swing, after they get done facing Wake, or something like that.

In the interest of full disclosure, 9 points in wOBA difference is NOT statistically significant.  Assuming that wOBA is a binomial (thus using the binomial SD formula), which it is not, one SD of the difference between 2,784 PA (Wake relievers) and 17,356 (non-Wake relievers) is around 10 points in wOBA.

Nonetheless, it is an interesting result, and one that requires further research. 

Can someone figure out how much in WE 9 points in wOBA for the remainder of the game is worth?  I think it is .87 runs per point of wOBA.  If we multiply that by 13 PA, which is about how many batters relievers face per game, I think, we get around .10 runs saved.  That makes Wakefield, if that 9 points in depressed ERA is accurate (we are certainly not confident in that number, given that one sigma equals 10 points), worth around an extra .14 runs in ERA, I think, which is around another 1/3 of a win per season.

Next is to look at other knucklers and to look at games following the ones Wake starts to see if there is any carryover or hangover effect.

Great stuff.

Since Papelbon is such an outlier, can you break up the relievers as Papelbon and not?

Post 14 was in queue for moderation, and has now been posted.

He’s got the Boston relievers, as I asked for in post 20.

#14 - You have to use the word code inside square brackets [] to get things to line up. It puts them in a monospace type.  You close each section with /code inside of square brackets [].  They will look like my post above.

Tom,
Is there any way to activate the [noparse] command on this blog?  Then we could actually show what commands look like.

I fixed post 14, as per 22.

SirK/22: I don’t know.  I’m fairly familiar with the documentation, but have never come across that.  If you wish, you can check yourself:
http://expressionengine.com/docs/

Cool data MGL.  Even if we can’t say it’s statistically significant, looks like there might be something to a knuckleballer hangover effect.

I didn’t say anything about the “injury waiting to happen” but can’t see any reason to argue that.  I don’t remember a pitcher ever hurting himself by stopping a throw, and considering all the things pitchers do that can cause injuries, it seems pretty far down on the list of worries. About the same level of being worried about falling satellite debris landing on the pitcher’s mound.

replying to mgl’s #16

According to Kalk, he threw the fastball 53% of the time last year in 3-0 counts and that agrees with my data, but he only went 3-0 15 times for games with pitch f/x data. According to my data, he threw 57% fastballs in 3-1 counts (of 22 times) and no curves. But in 3-0 he threw his slow curve the most of any count (perhaps to get weak contact?). Of all counts, he throws his fastball 12% of the time.

For all pitchers (well, a large enough dataset of all pitchers), 79% of the time in 3-0 counts the fastball is thrown and including cut-fastballs (some of which I’m sure were intended to be 4-seam fastballs) it’s 92%. In all counts, the fastball is thrown about 60% of the time.

#25, I would also guess that the 43% of the time he does not throw the fastball is when he does not mind walking the batter and he was pitching around him in the first place.  In other words, the batter may know that a fastball is coming or not coming more than the 57/43 split we see at the 3-0 count.

#14, I think standard deviation in runs scored is around 3.3 runs per 9 innings.  So for the difference between 65 and 300 innings, it is around 1.30 in ERA.  So your difference of 1.1 runs in ERA is less than 1 SD, as was my result for wOBA.

As I say many times, just because something is not statistically significant does not mean that the effect is not present or that the hypotheses should be rejected.

The results are what they are.  The “significance” merely tells us how confident that there is a true effect and if so, what that true effect might actually be, withing a certain level of confidence.

Plus, as I also like to say, these are Bayesian problems.  If we think, logically, that there might be a hangover effect, AND players think there is, then that is evidence that there might be, and it effects the significance of the empirical results.

I had a long post about other knucklers, which exploded in cyberspace.  I’ll post the condensed version.

I looked at other knucklers, namely, D. Springer, Candiotti, and S.W. Sparks, from 93-04.

I did the same thing, looking at batters hitting against relievers in games in which these pitchers started and batters hitting in games for the same teams, in which some other pitcher started.

After adjusting for the pool of pitchers and batters in the two groups, batters facing relievers after a knuckler batted 6 points worse in wOBA.  So it looks like a similar effect for other knucklers.

We have over 8000 PA against relievers when the knucklers started, and of course 20,000 or 30,000 against relievers with other starters.

Even with that many PA, one SD for the difference in wOBA between the 2 groups is only 6 points in wOBA, assuming a binomial.  So 7 points or so of difference (all the knucklers combined) is not particularly statistically significant.

But nonetheless it is a suggestive and interesting result.

I would like to look at games following a knuckler.  I am skeptical that there will be a carryover effect.

If wOBA = .330, and PA = 8000, then, as per the Appendix in The Book:

1 SD = SQRT(.330 * (1.1-.330) / 8000) = .0056

With 32,000 PA, 1 SD = exactly half the above.

Square root of adding the squares = .0063

OK, so it is only a smidgen more than if it were as true binomial.

I suspect there is a higher variance in the outcome of Wakefield’s starts than there is for most pitchers…

In all of Wake’s starts from 95-07, he allowed an average of 3.39 runs per start with a SD of 2.07.

For all other AL pitchers in that time span, they allowed 3.25 runs per start with a SD of 2.00.

Given that he pitched 6.17 innings per start and the average other pitcher pitched 5.75, I think that is exactly the same variance, no (6.17/5.75 = 2.07^2/2.00^2)?

One thing I would be curious about is if the teams that have a knuckleballer themselves belive that it has a significant enough hangover effect for them to influence how they set up and manipulate their rotation.

If (for example) wakefield shows an “unnatural” ratio of starts in the first game of a series vs last game of a series that would sugest that his team at least THINKs there is a hangover effect.

Fantastic thought, and easy enough to code for.

So, how do Redsox (and Pirates) pitchers pitch when Wakefield preceded them at some point in the series, as opposed to not.

Love it…

I’m not sure if that last comment was a reply to me but if it was that was not at all what i was curious about. (But I aggree that it is also an interesting thought.)

My reasoning was more something in line with this scenario.

If Wakefield is the Redsox 4th starter and they find themselves in a situation where they “need” a start from their 4th and 5th starters in the last game of a series against one team and the first game of the next series against another team.

Assuming that the normal rotation is to have Wakefield (4th) pitch the last game of the first series and also assuming that they could do so without anyone having to pitch on short rest is there any evidence that they switch it around and let the 5th start the last game of the first series and let Wakefield start the first game of the second series?

Now such opportunities are probably rare, but when a reasonable opportunity occurs a “thinking” ballclub like the redsox would probably want this hangover effect to benefit its own pitchers, and not those of some other team.

Right, the two would be related.  Yours was a question of frequency, to see how often he starts and ends a series, to see how random this is.

In my case, it was a question of performance to see that if they were thinking this, then how effective were they.

As I said, the next step is to see if the batters in the next game still had a hangover effect, regardless of who they are playing.  That should be easy enough to check in the same way that I checked the “in-game” hangover effect (look at “expected” batting, based on 3-year stats, versus actual batting, adjusted for the pitchers faced).