What's the statistical noise?

What's the statistical noise?

Mostly Jazz but a little Reggae mixed in.

It's the mating calls of sabermetricians in love

I'd buy a ticket to go see Strasburg pitch against the Mets at Citi.
Even if Castillo fields all the pop ups in that game...

It's the mating calls of sabermetricians in love

I remember that one:

Statheads in Love
Maury Browne


I can't keep up with what's been going on
I think my brain must just be slowing down
Among the Billy Beanes not selling designer jeans
Am I the only one who hears the schemes
And the strangled cries of statheads in love

God sends his spreadsheets to moms’ basements, the bastion of
Nerds with laptops not watching baseball games, all bastards them
Reading some BP guys, posting on Primer wise
Waiting for Pavement threads while Neyer cries
Out the mating calls of statheads in love

Last night I watched the Nats from Washington, the capitol
Strasburg escaped while we weren't watching him, like Borasans will
Now they’ve still got no fans, not even Kasten, Stan
And I hear Nats Stadium will be open man,
As convention land for statheads in love

Most statistical indicators say nothing about the overall level of play in a league: if both the pitchers and hitters at one level are better than those in another, their performances will cancel out. But a few figures avoid this effect, such as error rate (which measures the quality of the fielders) and the frequency of hit batsmen (which measures pitchers’ control). These numbers, taken together, can provide a quick approximation of a league’s strength. They suggest that Strasburg faced opposition roughly comparable to that of a middling Class A minor league squad.

Wait, what? I've never heard this offered before, and I have grave misgivings as to the truth of the underlying statement and the utility of anything that comes out. Betting men will have Davenport as a 1/9 favorite against me, of course.

According to Clay Davenport of Baseball Prospectus, had Strasburg been pitching for Washington instead of San Diego State this year, he would have compiled a 3.54 E.R.A.

I don't believe this for a second.

I don't believe this for a second.

Why not?

Mark Prior, who wasn't quite as good as Strasburg in college, pitched to the tune of a 3.32 ERA in the majors the same year he signed with the Cubs. That was a higher offensive environment too. Now he had a year off in between, due to the different schedule for signing draft picks back then, but it's not like he was pitching in the minors developing during that time either.

I think it's eminently plausible.

This just reminds us again of how good Prior was. Goddamned Dusty Baker.

All the statistical analysis in the piece is courtesy of Davenport. If you have beef with him or his methodology, please share it, but he is the only guy I know who even attempts to translate college statistics, so he is the person I turned to.

I think it's eminently plausible.

So do I. For another example, Lincecum put up a 4.00 ERA in 146.1 MLB innings in the year after he was drafted (after putting up a 1.01 ERA in 62 innings over his minor league career). Barry Zito was essentially ready as well. Sabathia was already an established starter at Strasburg's current age, and in the process of finishing up a 3.60 ERA season.

Some guys, when they're in college, are already prepared to take on the majors. Strasburg looks like one of those guys.

From Nate Silver at BP:

One thing that distinguishes young hitters from young pitchers is that young hitters can pretty much count on making steady improvements from the time they start playing professional ball until the time they’re 26 or 27. You might have a guy like Cameron Maybin who would be pretty overwhelmed if he tried to play in the major leagues today — but we can be fairly certain that he’ll be able to handle the big leagues in two or three years time. Cameron Maybin is a prospect.

The same is not the case with pitching prospects. Although there are a few categories of pitching prospects — particularly guys with good stuff, high strikeout rates and highish walk rates (think Homer Bailey) — that tend to improve more often than not, in general there is no systematic pattern of improvement after the age of 21 or so. Sometimes guys get better, of course, and sometimes they do so in a hurry — but you can’t take a young pitcher in a vacuum and expect him to improve the same way that you can for a hitting prospect. Mark Rogers (to pick on some low-hanging fruit) will probably never get his command sorted out, Yusmeiro Petit will never add enough ticks to his fastball to become a useful major league starter, Gavin Floyd will never learn how to keep the ball down, and so forth. All of these things are possible — but they’re not very likely.

The flip side, as Gary also alludes to, is that young pitchers often take less time to become dominant big league performers. Pitching, somewhat contrary to the mad genius reputation of pitchers like Greg Maddux, is more of a purely physical skill and less of a learned behavior than hitting is. Pitchers like Francisco Liriano and Jered Weaver and Cole Hamels — these guys weren’t just holding their own last year, they were among the very best pitchers in baseball. Someone like Hamels — or Tim Lincecum or Philip Hughes — might very well be as effective today as he’s ever going to be, before he’s had a chance for injuries and mileage to accumulate. Keeping those guys down on the farm is not conservative — it’s a downright irresponsible way to run a ballclub.

I think college pitchers have an easier transition to the majors than hitters. The going to wooden bats is tougher on hitters, and probably a welcome relief to pitchers.

The obvious example of a pitcher peaking young is Dwight Gooden, who should have won the Cy Young as a 19 year old (having set the single-season record for K/9, my all-time favorite statistic - that's absurdly impressive) and then of course, ran away with the Cy the next year.

K/9, my all-time favorite statistic

My only problem with that stat is that walks are included. Years ago I saw a k/27 stat (which is k's per 27 plate appearances) but I haven't been able to find it since then.

I think the result is plausible, if at the pretty near edge of it. I have question with what I quoted Dan quoting Clay. How a model of HBP for pitchers and error rate for fielders is turned into something that spits decimal precision numbers necessary for remotely sane translations is highly questionable to me on a number of levels.

Dan, if you've been using Clay's numbers, have you ever checked them systematically at a future date? Sure, not a lot of guys jump straight up, but I would think enough get at least significant cups of coffee within 18 months that you could give a general "ok/not ok" grade.

Alright, I guess my disbelief was misplaced. I didn't realize Prior had come up so quickly. I do think most guys need seasoning in the minors (even Gooden spent some time in the minors), but the Prior example shows some guys can cruise straight out of college. I did not mean to criticize the hard work Clay has done, it just struck me as overly optimistic at first, I withdraw my skepticism.

Wow, CP, for a gonnabe attorney, you gave up on yourself awful quick there. Allow me to defend your initial statement:

1) You guys listed five pitchers.

2) Prior had a year in between, which may or may not be good or bad, but it sure makes him different enough that you can't use him in this. If you're going to anecdotally justify an assertion like this, you can't leave a bunch of leeway on the analogicity (totally a word) of the examples.

3) The other three are Lincecum, who had a 4 that is at least an order of magnitude away from 3.45, and a year after he was drafted, and he had 60 IP in the minors as you note, and which Strasburg could not get and still pitch any notable amount in MLB this year. And Barry Zito, who had *170* MiLB IP, at 3.18, which in no way translates to a 3.45 in MLB. And then CC Sabathia (240 MiLB IP and *400* MLB IP before that 3.60, which was also followed by a 4.12 and a 4.03 and preceded by two 4.30 seasons) and Dwight ####### Gooden. The greatest pitcher of those ages (19-22), with only one competitor who happens to be unquestionably one of the greatest of all time. Dwight Gooden is not a case study for anyone, even if his son Gwight Dooden had the same stuff and was pitching in MLB at 19.

4) All of this ignores the fact that you're using these numbers to say it's plausible that it could be done, when the statement he quoted was that had Strasburg been pitching for the Nats this year and not in NCAA, "he would have compiled a 3.54 E.R.A." That is obviously a translation (as we all know) and therefore I read CP's "unbelievable" as "I find it unbelievable that Strasburg's numbers translate reliably to this number" and not "I find it implausible Strasburg could have posted this number."

His son was put up for adoption and then readopted back. That is why they have different last names.

My only problem with that stat is that walks are included. Years ago I saw a k/27 stat (which is k's per 27 plate appearances) but I haven't been able to find it since then.

I think Baseball Analysts tracks this statistic. Anyway, I was referring specifically to Gooden's mark, not the stat as a whole. :)

Am I missing something? Wouldn't K/27, if the 27 is PA, "include walks" just like K/9? Wouldn't you need K/27ABagainst for what cfb wants?

Am I missing something? Wouldn't K/27, if the 27 is PA, "include walks" just like K/9? Wouldn't you need K/27ABagainst for what cfb wants?

The issue is that a guy with 9 K/9 and 2 BB/9 is a better strikeout pitcher than a guy with 9 K/9 and 6 BB/9, since he's striking out a higher % of the batters he faces. FanGraphs has K%, which is measuring the same thing as K/27, just reported differently.

Edit: I am wrong, FanGraphs doesn't have K%, I thought they did. You can just do K/TBF though.

Am I missing something? Wouldn't K/27, if the 27 is PA, "include walks" just like K/9? Wouldn't you need K/27ABagainst for what cfb wants?

k/27 plate appearances would include walks in the figuring of the stat in a bad way. there is a drastic difference in k/27 for a guy who pitches 1 inning with 3 strikeouts and 3 walks (no hits) vs 3 strikeouts and no walks. K/9 would give you the same numbers no matter what. (note:I don't know the actual formula for k/27 just that it takes into account walks and hits while k/9 doesn't care. I guess I said it wrong the first time.

I think Baseball Analysts tracks this statistic. Anyway, I was referring specifically to Gooden's mark, not the stat as a whole. :)

Thanks, need to check that out then.

The issue is that a guy with 9 K/9 and 2 BB/9 is a better strikeout pitcher than a guy with 9 K/9 and 6 BB/9, since he's striking out a higher % of the batters he faces.

Ah, good point.

I admit I always thought K/27, when I saw it, was per 27 outs.

Why not use strikeout %, as helpfully supplied by BB-Ref?

Why not use strikeout %, as helpfully supplied by BB-Ref?

Ah, they do have it. I think that's fairly new (with the redesign maybe?). Cool.

Why not use strikeout %, as helpfully supplied by BB-Ref?

Umm, I'm stupid and missed it, even after looking for it when they did the redesign, never thought of it as strikeout % (which of course makes perfect sense)(heck I just noticed the platoon advantage split for the first time earlier today)

Dwight Gooden, who should have won the Cy Young as a 19 year old (having set the single-season record for K/9, my all-time favorite statistic - that's absurdly impressive)

I think it may be equally impressive that the first person to break it was Nolan Ryan in 1987 -- at the age of 40.

When PA is the denominator, you would always see it as K/PA and BB/PA. Whoever is trying to start the trend that K/27 means K per 27 PA, please ask him to stop.

I don't know that it's a trend and not just my misunderstanding of what I was seeing when I saw K/27 and cfb not explaining himself 100% correctly, Tango.

They suggest that Strasburg faced opposition roughly comparable to that of a middling Class A minor league squad.

Does this quote alone not strike everyone is absurd? I consider rookie ball to basically be a college all-star league less college draft picks from the first 3 or 4 rounds...with a few elite HS prospects sprinkled in. Are Mountain West Conference teams really better than that? or "middling Class A minor league squad[s]"?

That said, its behind a pay wall but I'm sure any analysis converting MWC stats to MLB stats would have to have error bars approaching ~6K/9 and ~3BB/9 so its hard to quibble either way.

His son was put up for adoption and then readopted back. That is why they have different last names.

Thank you for explaining this.

Wait, what? I've never heard this offered before, and I have grave misgivings as to the truth of the underlying statement and the utility of anything that comes out. Betting men will have Davenport as a 1/9 favorite against me, of course.

Bill James presented a list of statistical indicators of league quality. Like Rosenheck says, most stats represent both offense and defense, and don't directly represent quality, but some are "pure" measures of performance. James doesn't suggest that you can derive translation factors just from these stats, and the article doesn't really contain much detail about what Davenport is doing.

I thought that James listed these factors in the New Historical Abstract, but I couldn't find it in there now.

About minor league equivalency stats in general: As an outsider to the realm of advanced statistical analysis, here's a basic question that's always puzzled me:

In any given year, you can go to the leaderboards of any AAA league, and once you throw out the retreads and just stick to the prospects, you'll invariably find a mix of future stars, future journeymen, and career minor leaguers among them. As you go down the ladder to the lower minors you'll see the same thing, only the first two categories will be smaller and the last category bigger. Get a series of randomly selected baseball guides over the past 100 years and try this little experiment yourself, since that's where all the minor league stats are located.

And the question is, what's the real predictive value of these equivalency stats, when such a relatively small percentage of the players even get to the Majors in the first place? IOW if you take the top 25 statistically "equivalenced" AAA prospects in any given year, and it turns out that ten years later nobody outside a group of hardcore statistical junkies could pick them out of a police lineup, what does that say about the methodology?

To me it might say little more than there are factors that the methodology hasn't figured out yet, but it also might say that there's less of a "continuum" between the minors and the Majors than the whole idea of a "continuum" seems to assume, and that the only statistics of any real predictive value are those which are compiled in the Major Leagues itself, against 100% Major League competition. Not that we should simply throw out minor league stats (duh), but that their value is not necessarily as great as advertised by the concept of "equivalancies."

NOTE: I'm only offering this as a way of trying to draw out an answer, not as a final answer of my own (duh again). I'm simply trying to follow the Jamesian advice of beginning with a question.

But Andy, given that hardly no one misses the minors completely and yet we still have major league stars and regulars and the like, that means that for every guy who doesn't make the majors because he wasn't "as good as his equivalencies", another guy made it instead. The arbitrary roster limit is what limits the opportunities, not performance. If teams were allowed to have a roster as large as they wanted but every guy had to pass some basic threshold of production to be eligible, then I think you'd have an argument. But Stubby Clapp hitting .320 in AA at age 29 (totally making this up) doesn't mean he's got a spot on the Cardinals because .320 translates to .280 in MLB and they could use that guy. Meanwhile Joe Future Star hits .320 and does get a spot on the Cardinals. You have to, if you accept the notion of translations at all, weight those two .320s the same in the end, and the fact that Clapp spends 3 more years at AA and then manages a Kinko's night #### while Joe goes on to hit .300 every year for 20 years doesn't make Clapp's equivalency any less valid.

I'm not sure I've got Andy's point here, but I'll take a stab at answering the question.

Bill James observed that minor-league performance was, broadly speaking, predictive of major-league performance. The equivalencies attempt to make that more precise. So, if you like, the real predictive value of the equivalent stats is in slicing away some of the noise that comes from minor-league parks and leagues being more idiosyncratic than the majors. That is, instead of having to make a mental adjustment when looking at PCL stats, a computer can churn out some equivalencies and everybody is normalized.

In order to provide something of Andy's police lineup, I went through the BPro Top 40 for 2002 and compared the list against the the top 40 of the BA Top 50s of Lingo, Simpson, Callis and Boyd. These are names on the BPro list not on any of the BA Four:

Jack Cust
Josh Phelps
Orlando Hudson
John Stephens
Morgan Ensberg
Alex Escobar
Kenny Baugh
Jason Lane
Ken Harvey
Bobby Hill
Esteban German

Some of those names do appear in 41-50 of the BA Four, but none at 40. Cust is the highest on the BPro list at 20. (I think there's also some differences in eligibility between the two lists, which may account for some names.)

Assuming that the BPro list was compiled with the guidance of MLEs, while the BA Four only used traditional unequivalenced stats and chats with coaches/scouts, the basic answer to Andy's question is that the real predictive value is small, but perhaps worth the effort since you can get a player with some good value seasons in the likes of Jack Cust or Josh Phelps. But they are not going to turn up the 'next Albert Pujols', unless the scouts have already noticed anyway.

EDIT: It should be borne in mind that not even Major-League stats are great at predicting either. Isn't the number really '70% deadly accurate'?

Bill James observed that minor-league performance was, broadly speaking, predictive of major-league performance.

Yes, but it should be noted that he made this observation relating to Double A and Triple A performance.

He made no such claim about the lower minors.

Kinko's night ####

That's a first for me, as that is totally an inadvertent typo.

Two good answers by Jeff and fra paolo, especially the part about how equivalencies can "slic[e] away the noise that comes from minor-league parks and leagues being more idiosyncratic than the majors." And as long as "predictive value" is taken with an acknowledgment of the inevitable unknown factors that make up a player's DNA, then I can certainly see their practical use.

But another question: How far back do equivalency stats go in normalizing the idiosyncratic differences between the various minor leagues and parks? Obviously I ask this because the farther back they go, the better we could establish their true "predictive value," since we'd have a much bigger pool of players whose careers are over. And it would also be interesting to see how the effects of modern statistical breakdowns (which wouldn't always be retroactively available) would affect this predictive value. For example, I'd love to see equivalancies for the year 1949, which was when the minors hit their all time peak of 64 leagues. It's also interesting in that among the three AAA leagues that year, there wasn't ONE future Hall of Famer in the traditional leaderboard categories of either the position players or the pitchers. In fact, there was scarcely a recognizable name at all to anyone not either a hardcore baseball junkie, or of a certain age. So it'd be nice to see what a reconfigured leaderboard of those three leagues might have showed us.

Sorry to keep throwing out questions, but you've got to begin somewhere.

Back to proper Internet access, I can reply to Andy.

How far back do equivalency stats go in normalizing the idiosyncratic differences between the various minor leagues and parks?

I don't have a good answer to that. I'm only aware of the data Szymborski has published here in a form amateur sabermetricians like myself could access with a simple cut and paste. BPro has published data in all the editions I have, which go back to 2000 I think. (I only have a couple with me at the moment.) One could easily compile it with the appropriate data (as with major-league park factors) if it was in a Baseball Guide, but it's been years since I've seen one.

BPro did publish equivalencies, and I think Equivalent Average offers a normalized measure, so one could look at a few early editions to make comparisons with active players as to the predictive value. My handy 2002 BPro gives Mauer's 2001 Appalachian League an EQA of .230, which perhaps understated things a little! Really, though, you'd want to compare Mauer with other App Lg players, and maybe all players at that level of the minors, rather than with major leaguers, to get a real sense of potential.

One has to remain cognizant of a few things when using translations and minor league stats in general.

- Translations are not projections. Mauer's .230 EQA in the Appy League isn't saying anything about his future, simply that a player in that league is very unlikely to be a qualified major league hitter (I know you know this, Paolo, I'm just making a general point).

- Park effects tend to be more extreme in the minors than the majors. You combine this with the variability in the league offense in general and eyeballing stats becomes much more dangerous. Coors Field in its "prime" was an extreme park, but it was far more extreme than typical for a major league park and wouldn't even be much of an outlier in the minors. For example, take HR factor, which ranges at a high (3-year weighted) of a ridonkulous 172 for Joseph L. Bruno Stadium in Troy to a miserly 72 at Nat Bailey Stadium in Vancouver.

Statheads in Love
Maury Browne

Why wasn't this one mentioned in other threads? That's the best thing that I read here all week.

EDIT: It should be borne in mind that not even Major-League stats are great at predicting either.

This is where I think many mainstream fans and those in the MSM media make their main mistake- it is not that they mis evaluate how accurate MLEs are as a predictive tool- it is that they greatly overestimate how predictive past MLB stats are.

Grizzled veterans tend to be far less reliable going forward than the MSM thinks they will be.

You get things all the time like so and so has hit .300 for 3 straight seasons "there is no reason to think that now at age 33 the next 3 will be any different"

well yes there is a reason, multiple reasons in fact.