THE BOOK--Playing The Percentages In Baseball

Is NERC figured using James’ ERC formula, and then normalized to 4.00 ERA? Or is there something else to it?

I take a pitcher’s components (s,d,t,hr,bb+hp) and simply compute a standard lwts (not BaseRuns) per “game” and then add or subtract that to 4.00 (and add in wp).  As with batters, I don’t include IBB or (an estimate of) sac attempts.

It is park and opponent adjusted.

MGL:  I’ve seen discussions that though he’s taking Buchholz’ spot on Friday, Bos is likely to use him in place of Zink, not Buchholz.  And assuming they aren’t giving up anything worthwhile for him (almost surely the case), it seems like a good idea to me.

Oh.

I really think BaseRuns is the best method for doing component ERAs.

I ran a set of correlations for several available “ERA estimators,” under the condition that they be usable with the data available in the Baseball Databank. (That leaves out Nate Silver’s QuikERA, DIPS 2.0 and 3.0, any of the more advanced BaseRuns formulas and apparently NERC.) I divided BaseRuns (the pitcher version I found on Tango’s wiki) by IP and multiplied by 9 to make it into an RA-style number. Unweighted correlation with same-season RA/ERA, no cutoffs:

RA ERA
RA 1.000 0.973
ERA 0.973 1.000
BsRA 0.909 0.911
ERC 0.875 0.884
FIP-ERA 0.679 0.694
DIPS1 0.494 0.508

And weighted average error:

RA ERA
BsRA 0.433 0.648
RA 0.000 0.444
ERA 0.444 0.000
FIPRA 0.657 0.741
ERC 0.690 0.573
DIPS1 0.719 0.799

BaseRuns takes a hit on weighted average error because I didn’t throw in an unearned runs adjustment. Simply muliply BsRA by .92 and the average error drops down to .427.

Should probably note that I used all pitcher seasons from 1956 through 2007 where a pitcher recorded at least one out. 21,584 pitchers in total.

FWIW, the return from the Red Sox is basically just taking Byrd’s contract off the Indians’ hands.  The Indians save $2 mil.  There technically could be a PTBNL, but Mark Shapiro in the Cleveland papers was quoted as saying that it will be a player of “little consequence.”

Batters have hit .327/.423/.551 off Buchholz in his last 8 starts (38.2 IP). Of course it is a small sample with cherry-picked endpoint, but there it is.

I would like to see a study that looked at pitchers in a comparable situation (young, going through a rough patch, overall projection still good). It’s quite possible that historically, those pitchers bounced back to their projection. But maybe not. I am sure that someone on the Red Sox has done such a study.

Sure, Baseruns is better than lwts for pitchers.  Colin, what formula did you use for BaseRuns and for ERC?  I am surprised that there is such a large difference (in correlations with RA and RMSE).  Unless you used a very basic Jamesian version of ERC.  I have found that a rigorous lwts and BaseRuns work just about as well as one another for most pitchers (other than the extreme ones).

As far as Buchholz pitching poorly lately and Byrd pitching well lately, a projection is a projection.  We construct them from real live data.  They work, more or less, for young pitchers, old pitchers, and everyone in between.  However, we did find that pitchers do have some hot and cold predictive value, so you might have to nick Buchholz a bit for a while and maybe uptick Byrd a bit (although I am not so sure that it works that way - that it is symmetrical like that).

Still, to me, the evidence is strong that Byrd is nothing more than a replacement pitcher at this point in his career.  I wouldn’t have spent a wooden nickel on him let alone 2 mil.

And if had a nickel for all the times that teams (usually stupid ones) pick up these cast-off, bad pitchers (Ponson, Weaver, Lima, etc.), throw them for a while and then realize how bad they are (as if they couldn’t have realized that beforehand with a little sabermetric analysis), well…

I stand by my position that picking up Byrd and using him in lieu of anyone is not a good decision, and that using him in lieu of Buchholz is a really bad decision.  What is wrong with Masterson if you need an interim or emergency starter?  He is another pitcher with a very good projection.

I think that even smart teams can panic when a player with a good projection plays poorly (and his projection is still good) and I think that even a smart team can overvalue an established veteran like Byrd and get enamored by recent small samples of performance.  These things are human nature and are hard to resist no matter how smart and knowledgeable you think you are.

Do you have a rough idea of what replacement level is in terms of NERC, FIP, ERA, or anything else?

I suspect it would have to correspond to an ERA in the mid 5s?

For some reason, women continue to pick guys who they know are scum, but think that they are the ones who can “change” him.  When all you do is focus on the half that gives you hope, and half that time your hope is indeed fulfilled, you feel like you made the right choice.

MGL, I wondered the same thing as Jon—I thought replacement level ERA was around 5.75.  Byrd’s around 5.00 (xFIPs have been extremely consistent the past three years.)

Colin, would you necessarily want a strong in-season correlation between DIPS/FIP/NERC and actual ERA?  It seems you want that correlation to be LOW and then have a high correlation between year X DIPS/FIP/NERC and year X+1 actual ERA.

Three points:

1. Byrd is between replacement and average on a reasonable accounting.
2. Masterson’s projection is no better, and arguably worse than Byrd’s actual performance.
3. Byrd fairly consistently gives you 6-7 innings a start.  That’s not great, but with Boston’s middle innings difficulties, it does have value.

As for Buchholz, I wonder about something more insidious than a fingernail injury.  He has been awfully ineffective in 6 starts since his return (he pitched very well in April).

Buchholz was sort of forced by the Sox to undergo a mechanical change when he was down in the minors this year (lowering the arm slot I believe).  I’d bet there’s some mental/personal issues there, due to him conforming to their mechanical suggestions and then getting shelled back in the majors.  That’s got to be tough to take as a young kid who’s already thrown a no-hitter “his way”.

Byrd seems to have had an abnormally high HR rate this year.  Certainly that will fall.

I tend to operate under the assumption that the Red Sox are pretty savvy.  Rather than lambaste them over their decision, I’m much more interested in hearing theories as to why they’d do this.  (Besides, “they’re stupid").  My guess: Byrd gives up a lot of balls in play, and the Sox think they have the defense to handle that (especially the fly balls) a little better than other teams.

Is it really that dumb to acquire a replacement (or arguably a bit better) pitcher for virtually nothing?  Anything can happen the last few months of a season.  No team in the playoff race wants to start running out starters who have never pitched in the majors and may very well be sub-replacement.

Replacement level is a fine construct, but at the end of the day, you need to actually have such players available.  Sure, if they pitch Byrd instead of young guy projected to do much better, that doesn’t make a whole lot of sense.  But it doesn’t seem like their other options are significantly better and there’s value in having as much depth as possible.

MGL - For BaseRuns I used:

A = H + W - HR
B = (1.4*TBe - .6*H - 3*HR + .1*W)*1.1
C = 3*IP
D = HR

For ERC I used:

ERC = (H + W + HB)*PTB*9/(BFP*IP) - .56

where PTB = .89*(1.255*(H - HR) + 4*HR)+.56*(W+ HB - IW)

If the result of (H + W + HB)*PTB*9/(BFP*IP) is less than 2.24, then instead of subtracting .56, multiply by .75.

Sky - It depends on what you’re looking to do. DIPS/FIP aren’t just component ERAs - they’re also regression tools. If you’ve already regressed your components seperately (which is what I believe MGL does) then you’re better off using something that doesn’t include its own regression component.

I don’t doubt that DIPS or FIP would work better with Y+1 ERA than BaseRuns, although it’s something I’m planning to look at. But if you’re using, say, the Marcel projected components, I think then you’re probably better off using BaseRuns than FIP or DIPS to figure out ERA.

Ugh on the ERC one.

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

The Sox fans are suggesting that perhaps there is a mechanical (or worse) issue that Clay simply can’t get over.

Clearly, when discussing single players, scouting information should be given great weight, if the information they provide point to some drastic changes.

HOWEVER, at no time can that observation be influenced by his K/BB ratio or his BABIP or the number of balls thrown outside the strike zone.

Anything that is already quantified should be part of a forecasting (i.e., talent evaluation) system.

Now, one can argue that MGL’s talent evaluation system is deficient because it doesn’t include pitch-by-pitch analysis.  And you can argue that it lacks scouting observation.  Those are valid complaints.

But, you simply cannot say stuff like “he’s getting whacked around out there” or “he’s giving up more walks than strikeouts” or things to that effect.  Those do nothing to help in the evaluation.

These evaluation systems that MGL and others have simply *reflect* the results of past data with past performers.  That’s all they are.  What is missing is more data, as I noted above (pitch by pitch and scouting), to get better comparables.

Is there a better ERC formula than the one I’m using? I got it off the wiki.

Yes, that is the James one, so that’s correct.

It’s the “-.56” and other constant business that I don’t like.

In any event, I don’t see any reason to even bother with the way James is combining the events when David Smyth is combining the same events in a better way.  It’s like trying to do OPS, when you are already doing RC or LWTS.  Why bother?

Aside from injury issues, there is always with young pitchers the confidence question.  Buchholz has truly struggled since immediately before his DL stop (although he did well in Pawtucket).  How many young pitchers have been left in to get an everyday start while struggling as Buchholz has in the middle of a pennant race? It is very unusual, I think.  In recent years, Pelfrey in 07 and Reyes in 06 did not get left in, while Garza in 06 was.  Garza did not thrive.

Buchholz has particularly struggled with runners on base.  Whether that is due to a mechanical problem throwing from the stretch or pure bad luck or something else, I have no idea.

Tango--

Doesn’t FIP have a fielding constant on the end of it? 3.2 or something IIRC.

Buchholz’s pitch selection has dramatically changed since early in the season. Since he has come back, he has been changeup shy. This is interesting only because one of the reports when the Red Sox sent him back to the minors was that they wanted him to work on the fastball and “pitch forwards” (e.g., fastball first) rather than “pitch backwards” (breaking stuff first) as he had been doing up to that point.

He had minors stints where he threw many more fastballs than breaking pitches and that has obviously translated back into the Majors. But it’s very different from his normal M.O..

In the very dominating performance @ Tampa earlier this season, he threw 33 through 74 pitches (45% changeups)… and in his very ugly performance last week, he threw 13 through 74 (17% changeups). [74 is not a random number of pitches here, he was pulled after 74 pitches in Chicago].

His changeup is still dominant, but he’s throwing it a lot less. This might be just pure speculation, but I’m not sure that there’s been a drastic mechanical change so much as a “strategic” change that isn’t working. I think they’ve tried to make him a more standard fastball first type of pitcher and it doesn’t appear to be working.

Pitch FX data here:
http://www.brooksbaseball.net/index.php?post=2008/08/one-great-pitch-one-terrible-pitch_10

The point about Buchholz is that with a DIPS NERC of 3.74, which he has as of a week ago or so, he has NOT pitched badly. If you want to say that he has pitched badly by virtue of any of his garbage stats, then I’ll stay out of the discussion and leave it to the mainstream.

Replacement NERC is 5.00 generously, but realistically more like 4.80.

Mike, if 5.95, which is the number of innings that Byrd has averaged per start this year, is “6-7”, then your argument might have some credibility.  Actually, the last thing in the world you want is for a pitcher like Byrd to pitch deep into a game, regardless of how bad your relief core is.  There is no middle relief core in baseball now, or ever, that is 1 run above average in true talent.  There are plenty with a sample performance in that neighborhood, I assume, but none with that kind of true talent.

Anyone who thinks that Byrd is decent pitcher is welcome to throw me an over-under on his ERA and I’ll probably take the over.  As usual, if you have a conviction, put your money where your mouth is and we’ll see if you really do have a conviction or you are just yapping.

Colin, one of the problems with that version of ERC is, obviously, that it approximates a lot of things and therefore those correlations and RMSE will not be very robust.

If you want to compare BaseRuns with ERC for pitcher, I think you have to start with the whole enchilada, which is a pitcher’s s,d,t,hr,bb,hp, and IBB (in order to subtract those from his total BB).  If you are just using the traditional pitcher stats line, hits, HR, BB, and SO, you are in trouble from the start.

The correct way to do an ERC is using lwts.  Actually it is BaseRuns, but absent that, I think it is traditional lwts.

Like Tango, I am also not crazy, to say the least, about using those constants in an ERC formula, and changing them depending on the value of the first term.  No wonder you are getting low correlations and high RMSE for ERC.

Dan/21: I’m not opposed to constants per se.  I’m opposed to it in the ERC construction, especially with that extra fudge at the end. 

James is subracting .56 to turns Runs Allowed into Earned Runs Allowed, using historical data.  But, he’d be better off using a multiplicative factor.

Dan/22: good charts.  If you are going to talk about fastball command, it’d be good to see the location in the strike zone of the fastball, and how much movement they had.  And comparing to the “good” Clay would be nice.

MGL/23 - Well, based on Bill James’ published formula, that’s what ERC is.

It’s like saying, “A better way to figure Runs Created would be (2 x (TB + BB + HP) + H + SB - (.605 x (AB + CS + GIDP - H))) x .16.” No, that’s something completely different (and for certain uses, better) than Runs Created. Maybe your way of figuring NERC is better than James’ ERC formula, but he first laid claim to the name ERC, and so it is what it is.

To me, ERC is essentially Earned Runs Created. ERC is a Runs Created formula, with all the baggage it entails, and then a factor to convert it into earned runs per nine innings. You can argue that this is a less-than-optimal approach (in fact I thought I did).

As far as the components go, it was an apples-to-apples comparison - the BaseRuns formula used the same data as ERC. (TBe is figured as 1.12*H + 4*HR.) They both used hits, walks (plus hit by pitch) and home runs. (Runs Created uses batters faced, actually, so it uses more data than BsR in this case.)

I could do another study with the Retrosheet data to look at all the components, using one of the Tech versions of RC and Tango’s BsR implementation, but I don’t think that would prove anything that hasn’t already been gone over before.

Colin, rereading this thread, my bad on what you were trying to do with ERC.  I thought you were taking an advocate-type position, when really you were just being a good researcher.

***

MGL: I’m not sure I follow your position here, compared to the Pierre position.  Pierre, by your calculation, costs 20 runs per 162 G, or say 5 runs for the 40 or so remaining games.  Even if Byrd is 1 run worse than Bucky per 9 IP, how many innings are we expecting here?  45?  That’s 5 runs.

On the one hand, it seems like people make a big deal by playing the inferior player (Pierre) and on the other hand it seems like people are silly for playing the inferior player (Byrd).

I’m lost!

Come on Tango, you know my position.  Of course it is no big deal if they play Byrd or not.  Same with Pierre.  But, a stupid move is a stupid move.  Of course some stupid moves are more stupid than other stupid ones.

I truly think that Boston is being “mainstream” with this decision.  IOW, that it is based on some common, but erroneous notions and/or perceptions.  I am not particularly confident with my opinion though.

I read the comments on BTF when they occasionally link to threads on this blog and sometimes I want to scream, “This is a freakin’ blog - not an online journal. Sometimes, in fact more often than not, I read something at 2:00 in the morning, and then think out loud on the blog.”

That is the extent of my position on this matter.  I think that Byrd is a replacement pitcher more or less.  I could be wrong, just like any of us could be wrong about our assessment on any players.  Given that, it is a little odd that a smart team would pick him up, regardless of his “veteran status.”

I also think that Buchholz is a pretty good pitcher.  If he is indeed going to be replaced by Byrd, which is what I read (if that is not true then I was misinformed or read something incorrectly), then it is doubly odd.

Here is another thing I don’t like:

We (I, whatever), say that short term performance, good or bad, has very little predictive value, a little more for pitchers than for batters, but all in all, very little for anyone.  We look at 10 years of data or so, which backs up that assertion.  No matter how good or bad a player plays in any spate of time (at least for the spates we looked at - a week, two weeks), he performs around at his recent career level in the next day, week, or 2 weeks, and presumably for any random point in the future.  OK, no problem there.

Now, a player like Buchholz performs poorly (and, again, I am not even sure that that is the case, but I’ll assume that it is) for a spate.  I say, “Well according to our research, that means very little for the future.  We still expect him to perform at his career level, with regression toward the mean of course, which, in this case, happens to be very good.” No problem there.

Now, everyone comes along and says, “Wait. I don’;t don’t your research, but in this case you are wrong.  I don’t think that Buchholz is a very good pitcher.  At least not now. I don’t expect him to pitch well at all in his next start or next 5 starts.  I have no problem taking him out of the rotation.”

And I ask, “Why is that?  Why is he an exception”

And they respond, “Because he has pitched so bad lately!”

No one else sees any problem with that logic? 

I hear that ALL the time.  That is the basic argument against Buchholz.  No one likes him.  And it is NOT like Boston has told us or anyone for that matter, “Yes, we know that recent performance, even from a pitcher, has very little predictive value, but we have watched him extensively (of course) and we know things about him that you guys don’t.  And that is why we are pretty sure that he does not fall into that category (of recent performance having little predictive value).”

No, we have not heard that at all.  So I would like everyone to shut up about the fact that Buchholz does NOT project well regardless of his numbers (and the fact that he is a TOP prospect, which makes the mean we regress to better than just any old rookie pitcher) because he has not pitched well lately!

In fact, we can create another poll and everyone can choose Buchholz’ and Byrd’s ERA going forward, if they start at least a couple of games this year.  I’ll bet that it will be just like the Vidro poll.  When people are forced to actually take a written, certified stance on something, where they can actually be proven “wrong” on paper, all of sudden opinions change.  I’m willing to bet that the mean ERA for Byrd will be, like, 5. something and for Buchholz, a little less! If not, I’ll gladly take the appropriate over or under to conform to my projections!

But, as I said, I agree that whether they acquire Byrd and/or use him for 4 or 5 starts in lieu of Buchholz, Zink, or any other alternative is NOT that big a deal. As I said, I was just thinking out loud in the middle of the night. I was not doing any scientific research, to say the least.

If anyone thinks that that makes me as “dumb” or dumber than I am making the Red Sox out to be, so be it.

Buchholz is an exception, in my eyes, because he underwent a fundamental change in the way he pitches between his recent stint in the majors, and the rest of his career.  And, according to Dan/22, he has also changed the way in which he goes about his pitch selection.

No problem, Tom.

Just spitballing here… you can take the B from the BaseRuns above:

(1.4*(1.12*H + 4*HR) - .6*H - 3*HR + .1*W)*1.1

and do a little algebra and simplify out to:

(1.06 * H + 2.86 * HR + .11 * BB)

Ok, now from a FIP point of view, H is simply:

(.290 * (BPF - K - BB - HR))

So put that in the B factor above in place of H, simplify out, and you get:

(3.07p - 3.07 k - 1.97 w - .21 r)

Then your A factor is:

((.290 * (p - k - w - r)) + w - r)

Simplified to:

(2.9*PBF - 2.9*K - 1.9*BB - 3.9*HR)

And you get:

BsR=(2.9*PBF - 2.9*K - 1.9*BB - 3.9*HR)
*((3.07*PBF - 3.07*K - 1.97*BB - .21*HR)/((3.07*PBF - 3.07*K - 1.97*BB - .21*HR)+(3*IP))+HR

And… voila! FIP BaseRuns.

Buchholz is an exception, in my eyes, because he underwent a fundamental change in the way he pitches between his recent stint in the majors, and the rest of his career.  And, according to Dan/22, he has also changed the way in which he goes about his pitch selection.

Maybe so.  To be honest, I have not seen him much at all.  And despite what most people think of analysts of my ilk, I am very much an observer of pitchers.

Typo. Formula should be:

BsR=(2.9*PBF - 2.9*K - 1.9*BB - 3.9*HR)
*((3.07*PBF - 3.07*K - 1.97*BB - .21*HR)/((3.07*PBF - 3.07*K - 1.97*BB - .21*HR)*(3*IP))+HR

The Sox could easily be taking the view that the run difference between Byrd and Buchholz isn’t a big deal, but that continuing to put pennant-race-grade pressure on a kid whose future they own is undesirable in the circumstances. We don’t know how Buchholz has been holding up mentally in the 10 weeks since he last got anybody out.

Fangraphs has Buchholz’s pitch selection as close to identical as last year (though small samples):

http://www.fangraphs.com/statss.aspx?playerid=3543&position=P

mgl - let’s say that you have a young player who’s performing substantially better or worse than his pre-season 2008 projection.  Obviously his 2009 projection will be affected by his full-season 2008 stats. 

But when does his current season performance start to matter in terms of predicting the rest of the 2008 season?

Traced, I wouldn’t call that close to identical.  His Curve has gone from 15% to 21% selection, while his fastball has dropped by the same amount (53% to 47%).

His fastball speed has also gone way up, which may also mean that his fastball movement was sacrificed.

Or, there may be some misclassification of course.  As an aside, I’d probably call the 25% of all a pitcher’s pitches the fastest as “fastballs”, and then any pitches that is within a certain speed of the average of his top 25 (say 2mph) as also his fastball.  Anything that you can’t classify as a breaking pitch or changeup, but you would otherwise have called a “fastball”, I’d call it something else (an oops-fastball).

When you look at Clay’s fastball numbers, and you see both the frequency go down, and the speed goes up, it is possible that the 48% through 53% of his fastest pitches, which in 2007 were called fastballs (and thereby lowering his average speed) are now being called something else in 2008 (like say an 86mph curveballs, bumping up their average as well).

In fact, if you take ALL of Bucky’s pitches, the average speed is 85.0mph in 2007 and 84.8mph in 2008.

Isn’t it therefore too coincidental to think that his fastball speed went up, his curveball speed went up, while his overall speed stayed the same?

Hawerchuk/35: forecasts should be revised after every single play, just like a stock price is adjusted after every single trade.