Now, things are more than a little different. Sabermetrics seems to have lost its way.

Everything was ruined when all those nerds came out of their mother's basements are started looking around at the rest of the world. Everything would've been better if the sports writers had just managed to keep them locked down there.

Teams that hit more triples are generally worse at run scoring on account of being small and speedy instead of power-hitting monsters, but the analysis blamed the triples for that rather than treating it as a symptom

IANAM, but how can an analysis "blame" triples for anything? If teams that hit more triples tend to score fewer runs than teams that hit fewer triples, that's just an observation, not a blame. All else being equal, if such teams hit even fewer triples they would assuredly score even fewer runs.

But, despite the claims to the contrary, sabermetrics is most emphatically not science. Proper science requires, at the very least, controlled experimentation, which is something impossible to manage in baseball analysis.

This peculiar definition of "science" also seems to rule out other observation-based fields like astrophysics, since we can't put stars in the lab and perform controlled experiments on them.

Anyway, in this case, total team runs in a season were tested against home runs, triples, walks, errors, etc. All sounds good and sensible, right? Wrong. When the numbers come back, it turned out that hitting triples cost a team runs.

Obviously, this result is insane. Triples are innately good things, and one would imagine that teams would be rather glad to hit more of them. What happened in the example above is that the wrong effect was picked up. Teams that hit more triples are generally worse at run scoring on account of being small and speedy instead of power-hitting monsters, but the analysis blamed the triples for that rather than treating it as a symptom.

Echoing Bob's point, I highly doubt that whoever first did this study actually claimed that "triples were bad" and that hitting them caused a team to score fewer runs. If he or she did, I'm sure their error was quickly pointed out. More importantly, this example seems to me to illustrate the power of sabermetrics -- an analytic study generates an initially counterintuitive result, we think about it, advance theories to explain it, and learn something in the process. What's wrong with that?

Sabermetrics shouldn't be so incomprehensible so as not to call up the smell of fresh mown grass in midsummer, or the crack of the ball off the bat, the blur of seams as an outfielder whips a throw in towards his cutoff man. Statistics shouldn't be sterile and clean and shiny and soulless. They shouldn't just be about baseball; they should invoke it. Otherwise, they run the risk of losing the language which makes them so special.

What does this even mean? It seems to me that statistic can only "call up the smell of fresh mown grass in midsummer" because of our familiarity with it and the associations it triggers. Batting average would seem bizarre and incomprehensible but for its long history in the game. Conversely, "new-fangled" measures like OPS have now become commonplace enough that they "invoke baseball" pretty readily, at least for me. (Maybe they don't "call up ... the blur of seams as an outfielder whips a throw in towards his cutoff man", but that is asking an awful lot of a hitting stat.) While many new metrics are more opaque and may be inaccessible to people who don't want to take the time to learn about them, nobody is forcing MacAree or anyone else to use them.

What the hell kind of regression did he do where triples cost the team runs? I guess if you just went to a single independent variable regression between triples and runs it might turn out that way, but the who the hell would do that? If you know enough to do a single independent variable regression, then you likely know enough to do one with several independent variables which would fix his objection pretty immediately. Running the numbers for teams from 1955-2010 (when IBB started getting counted), the linear regression comes up with:

1B 0.495
2B 0.578
3B 1.172
HR 1.473
BB-IBB 0.320
IBB 0.233
HBP 0.332
SB 0.161
CS -0.133
SF 0.663
SH 0.036
SO -0.103
AB-H-SO -0.097

So triples are not only positive but worth more than doubles and singles. Even doing it just one to one between triples and runs gives us:

Intercept 690.642
Triples 0.326

So I'm unclear as to how triples came up with a negative value in a regression. Maybe he used a small sample.

The most interesting number from the above is the relatively low negative value of the CS. This is likely a base running proxy; IE the actual caught stealing is more damaging than that but the relationship between good baserunning and higher caught stealing totals lessens it in the regression.

I like the comment where this article is described as similar to Thomas Kuhn's work.

IANAM

In the spirit of IANAL, might I suggest IANUS* instead?

I Am Not "Uh" Sabremetrician.

But, despite the claims to the contrary, sabermetrics is most emphatically not science. Proper science requires, at the very least, controlled experimentation, which is something impossible to manage in baseball analysis.

This peculiar definition of "science" also seems to rule out other observation-based fields like astrophysics, since we can't put stars in the lab and perform controlled experiments on them.

Yes, it would seem to rule out a whole swath of pseudosciences, like "theoretical" physics, and "mathematics".

Snark aside, mabe it's not controlled, but MLB is one heck of an experiment that is available to sabermetricians. And we can models that are far more "realistic" than many scientists in other fields could dream of (especially compared to actual pseudosciences).


Sabermetrics shouldn't be so incomprehensible so as not to call up the smell of fresh mown grass in midsummer, or the crack of the ball off the bat, the blur of seams as an outfielder whips a throw in towards his cutoff man. Statistics shouldn't be sterile and clean and shiny and soulless. They shouldn't just be about baseball; they should invoke it. Otherwise, they run the risk of losing the language which makes them so special.

What does this even mean?

I don't know, but I want some of whatever it is he's taking.

This is some first class navel gazing. Well written, but still.

Sabermetrics shouldn't be so incomprehensible so as not to call up the smell of fresh mown grass in midsummer,

So what in the #### are we waiting for? Get with the program, you dumbassed seamheads.

#4 The quoted section of the article reminds me of the bad part of RSB. Triples have a low correlation with runs scored -- much lower than doubles. Of course that doesn't mean they're not as important for run scoring, but rather that they're semi-random events.

Triples have a low correlation with runs scored -- much lower than doubles. Of course that doesn't mean they're not as important for run scoring, but rather that they're semi-random events.

Is it that they're semi-random, or isn't just that they're dramtically rarer? (Which, I suppose, might be a different way of saying the same thing.)

Triples were "the" big extra-base hit, the big RBI blow, pretty much until the 1920s. But they've steadily receded in frequency, and have never been rarer than today. Big, strong, power hitters of many decades ago (the Gehrigs, the Musials, hell, even the Rices) often sought to stretch their doubles into triples. That rarely happens today because teams have simply gotten wise to the risk/reward, and also because defenses are just far better on relay throwing plays than they used to be.

Which means that triples have progressively become the province of the speedy low-power slap hitters. Good for them, I love to watch these kinds of players, they're fun and interesting, and there is simply no better play to observe in baseball than the triple. But that's an aesthetic appreciation, not a run-value assessment. Much as we may love our slappy speedsters, we well know that they're far from the most productive run generators on just about any team. And if you are a team that has a high proportion of these guys, then almost by definition you aren't going to be much of a run-scoring team.

How in the world does making observations such as these make watching and thinking about the game any less fun?

we can't put stars in the lab and perform controlled experiments on them.

But what a day that will be!

we can't put stars in the lab and perform controlled experiments on them.

But what a day that will be!

I say we start with Ben Affleck and Will Farrell.

Wouldn't the improvement in equipment have something to do with the decline of triples? In the nineteenth century and through a good part of the twentieth unless a fielder was in front of the ball he wasn't likely to catch it, or having caught or stopped it, follow up with a play resulting in an out or in holding a runner. Why did gloves get so much better so suddenly? In fact, it seems, just as a matter of impression, that there was a quantum leap in design and construction starting around the late-forties, early-fifties. But I could be wrong. Maybe there was no radical jump. We all know shaky how first impressions can be.

Wouldn't the improvement in equipment have something to do with the decline of triples?

I think it's more or less a whole lot of variables, smaller parks, better equipment, less worth it in the short run, the argument that players don't come out of the box anymore thinking triple (Stan Musial isn't a speed burner, yet he has the most triples among all players who's career started after 1930) etc.

I think it's more or less a whole lot of variables, smaller parks, better equipment, less worth it in the short run, the argument that players don't come out of the box anymore thinking triple (Stan Musial isn't a speed burner, yet he has the most triples among all players who's career started after 1930) etc.

It's all of that, and also it's just the case that fielders throw better than they used to, and that in combination with better gloves makes the relay throw play more effective than it once was. Truer bounces on better-groomed fields is also part of it.

I think this is a pretty good article with a pretty lousy headline. We have certainly seen many instances where groups of statisticians or mathematicians have attempted to perform analysis of baseball concepts without having the requisite baseball knowledge. with hilarious results. I fully agree with what he's saying about "top-down" rather than "bottom-up" and Tango in particular (I think) has written some excellent articles in the past about regressions gone amok.

I think the author has a good point but don't really see that this is a "problem" that encompasses the "field" of sabermetrics. I think it's a small subset.

Further:

But, despite the claims to the contrary, sabermetrics is most emphatically not science. Proper science requires, at the very least, controlled experimentation, which is something impossible to manage in baseball analysis.

I see it was already highlighted above, but this certainly caught my eye. Is it not possible to have good science through careful observations and falsifiable tests? Isn't this what cosmology?

It's all of that, and also it's just the case that fielders throw better than they used to,

What's your basis for that? Not saying you're wrong, but what's the evidence?

My guess would be park size/shape (almost every park in the 1910-40's had a 420'+ CF b/c they fit in city blocks), no more turf, and the decline of busting it out of the box, inthat order.

double post

What's your basis for that? Not saying you're wrong, but what's the evidence?

A lot of it is simply and directly personal observation of baseball at many levels over the past (oh my god) nearly half-century. Admittedly that's entirely subjective, but it's there nonetheless and I can't ignore it: I can clearly see that players just throw better than they used to. They're taught better and more consistent technique from the time of their boyhood than they used to be, and they're bigger, stronger, vastly better-conditioned, and just better athletes.

This is consistent with all the quantifiable demographic data we have on the pool of talent. And it's obvious that with vastly more money in the game, players are highly motivated to perform at the role as the serious profession it is. I don't think there is any good reason to doubt that the quality of play today is better than it's ever been, and throwing and catching fundamentals aren't far from the heart of that.

Will Jose Reyes break Musial's modern career triples mark? He has 99 in his first 1,000 games, is hitting them as often as ever, and has not bulked up or slowed down at all.

Carl Crawford is on a similar pace, but he's dropped off in recent years and is now stuck in an awful park for triples.

#17 Tango's written an article called "Don't use Regressions". His thesis being that as soon as you have full PBP data you can get better results by looking at the change in game state than you can via regressions. I've seen at least two follow on articles which served to double-check his results.

Don't know if I buy it. Regressions are easier than the game state work and from what I can see, the difference in the results is in the noise.

Will Jose Reyes break Musial's modern career triples mark? He has 99 in his first 1,000 games, is hitting them as often as ever, and has not bulked up or slowed down at all.

I selectively endpointed my list of triples to get Musial to the top of the list. If you go by 1920(which is when many people argue is the real start of 'modern' basebally) Musial is second behind Paul Waner. (Clemente is fourth though...the top four -Waner, Musial, Goslin and Clemente- triples hitters since the 1920's have more triples(707) than they do stolen bases(441). You probably won't see that again.

the blur of seams as an outfielder whips a throw in towards his cutoff man

I'm a lifelong Cub fan ... these are not my summer memories. Give me a stat that reminds me of a throw sailing over the cutoff man, 15 feet off target, hitting a brick wall on two bounces and rolling into the LF bullpen!

The Problem With Sabermetrics

Woo-hoo!! We've got it down to one!

Anyway, the article is OK enough but doesn't provide any sort of solution to THE problem. And, like Voros, I want to see this regression with a negative coefficient for triples. I'd bet dollars for doughnuts (how much do doughnuts cost these days anyway?) that he's misinterpreting the coefficient. He already shows signs of being as ignorant as he claims others are -- regressions haven't been "hard" for at least 30 years (with free publicly available software since at least EPI-INFO); his explanation for the negative coefficient on triples doesn't seem to make sense if singles, doubles and HRs are in the equation ... a team's (or player's) HR production is controlled in the regression so that coefficient (as described which is surely incorrect) is saying that _given the same number of HR, doubles, etc_, the team with more triples scores fewer runs. That's got nothing to do with low-power speedsters not hitting HR because you've controlled for HR. Unless he's saying there should be an interaction between triples and HRs or some such.

Ahh, I think I found it. It appears to be a regression of FA salary, not runs, on various characteristics by David Gassko (www.insidethebook.com/ee/images/uploads/dsg_freeagent_Thesis.pdf). Gassko surmises:

The negative coefficient on triples is a mystery, but not a particularly
important one. As shown in Figure (5), the average hitter in the sample averaged fewer
than two triples per season, with a relatively tight spread at that. Moreover, since triples
likely decline strongly with aging due to their great dependence on a player’s speed, it is
likely that both the mean and spread decline further past the first season of a player’s
contract.

I'm not bothering to read the whole article nor pay much attention to his regression to see if I agree with Gassko's approach but I'm not appalled by the idea that triples have a negative impact on FA salaries ... although Carl Crawford would probably now throw that equation off. :-) (Actually I'm not even sure if this is triples before or after the contract was signed.)

Worth noting that you'd probably see players in the range of Goslin, Waner and Clemente if you had parks like Forbes Field or Griffith Stadium. Musial's triple totals weren't due to his park though. Sportsman's Park was a good place to hit but it didn't create triples. Waner hit ~2/3 of his triples at home. Goslin isn't near as dramatic (though I suspect the splits are pretty large in his Wagington days). Clemente has a large home/road triple split too -- for the same basic reason as Waner.

#24 Just thinking. What happens if you're silly enough to put hits and triples in? I think you'd still need some other form of GI to get the kind of GO he's reporting but it'd be a start.

One of the interesting things I've noticed when I ran regressions is that caught stealing appears to function in some way as a proxy for speed (caught stealing consistently comes back as less negative than you would expect). But if you don't include caught stealing I'd expect triples to kind of pick up the slack a tad and show up as more positive. Don't know. though. I was getting sensible results from regressions as soon as I got a copy of the Lahman database. I'm not sure how to get silly ones.

I'm really not a mathematician; I just reckoned that TFA, despite referring to "regression," was talking about some kind of simpler correlation between many triples and few runs. If he confused a study of FA dollar value for one of run-scoring, that's even weirder :)

Waner hit ~2/3 of his triples at home... Clemente has a large home/road triple split too -- for the same basic reason as Waner.

Willie Stargell had 55 career triples, 33 of them at Forbes Field - and Forbes was the Pirates' home stadium for less than half of his career.

So I'm unclear as to how triples came up with a negative value in a regression.

Voros - it could just be a different sample. Your regression covers 1955-2010; if you ran a single-variable regression on triples and runs that goes back to the Deadball Era, you might get a negative correlation just because there were far more triples back then. (You'd probably have to exclude the high-scoring, triple-happy 1890s, though.)

Willie Stargell had 55 career triples, 33 of them at Forbes Field - and Forbes was the Pirates' home stadium for less than half of his career.

And 13 of the remaining 22 at 3 Rivers. 46 of 55 career triples at home. The only other stadium in which he hit more than 1 was Jarry park. He had 2522 ABs in 10 different parks and hit 0 triples.

When I first saw this article the other day, I was left scratching my head, wondering what his specific objections are. I can agree with some general statements in the article (for instance, who endorses "number crunching for the sake of number crunching?"), but I have a hard time connecting those statements to his conclusion that sabermetrics has "lost its way." You can go to practically any field and find people who misapply the tools or analyses; it's not isolated to sabermetrics. I can understand the author's desire not to point the finger at any particular person, but it would have been helpful if he had given more specific examples that caused him to be upset. Does he want sabermetrics to address more interesting topics? Does he object to sabermetrics becoming more mainstream for ordinary fans? I'm not sure.

#29 Anything is possible if you're running single variable correlations over a short enough time frame.

#29 Anything is possible if you're running single variable correlations over a short enough time frame.

I was thinking this one would actually be a long timeframe - 1901 to present or so. The Deadball Era being the lowest-scoring and highest-triple period out of that sample would have a chance of producing a negative correlation.

But as a general rule, yes, this is very much correct.

#31 I agree that it would be helpful if the author cited a specific article. And to be more specific, one that was widely seen and didn't attract criticism. One of the regulars here (blanking on who it was) has complained about a lack of peer review in Sabremetric publication. Thing is that while you don't need peer review to be published, anything of significance gets taken apart.

Show your work was RSB's mantra and that general spirit is carried forward in places like this -- or the Book blog or ... well lots of places.

Should have read Walt's #24 more closely. I'll personally bet that there is no significance between FA salary and triples and that Dave should have dropped triples from his model.

Certainly that consistent with what Walt quotes from Dave's article.

hello boys

newsbreak - the astros uck-say

now if only i could find one of those great writers like BITGOD to explain to me what is the problem. it must be that carlos lee is hitting triples and not stealing many bases because you sure can see THAT being the problem. and the guys are playing great small ball - even sac-bunting with your #7 hitter (who was hitting like .280 at the time) in the second inning to move the runners along. or was it walking the .105 hitter hitting 8th to pitch to the pitcher?

anyways, i am not getting how a bunch of guys and a few grrrls arguing over regression and coeficcion of whatever has anything to do with what the media says. seeing as how the media doesn't go around quoting THT, Bpro, the book blog, fangraphs stuff like FIP+ or SIERA etc.

what does that have to do with either enjoying going to the ballpark, watching it on TV, listening to the modern radio announcers who talk about all kinds of shtt don't have a THING to do with the ball game and never shut up for a SECOND or give you an idea of what is happening on the field. or watching your kid play?

this is like if i was grousing about all the total jerk chefs on tv arguing about 1/4 tsp of some herb grown somewhers i never heard of or how terrible it is if you cook without el-expensivo pan - and all this is done for money/prizes, we all know - and then i say - oh i can't enjoy cooking no mo because some buttmunch on the TV

oh i can't enjoy cooking no mo because some buttmunch on the TV

Well that's your problem right there, a television makes for a terrible cooking implement. I always saute my buttmunch in a pan. Or better yet stuff peppers with buttmunch, rice and tomatoes, and bake them.

But, despite the claims to the contrary, sabermetrics is most emphatically not science. Proper science requires, at the very least, controlled experimentation, which is something impossible to manage in baseball analysis.

Funny stuff. Once I saw this, there was no need to go any further with the article.

What would make an interesting article would be a look at the as yet underexplored countries of sabermetrics. A survey of the field, if you will. What's known, what remains to be discovered, and what tools or data might best help us do that. Things such as:

1. Starting pitcher use
2. Bullpen use
3. Injury prediction
4. Falling off the cliff prediction
5. (perhaps a subset of 3 and 4) Finding a player's most similar players
6. Precise fielding numbers
7. Chemistry, it's effect
8. Catcher defense

And so on. What did I leave out?

One of the comments was interesting. The writer of it probably has a lot of company:

It feels like is continually trying to expand itself simply for the sake of expanding, rather than expanding in an effort to help people better understand what goes on during a baseball season. And, once that direction is lost, newly discovered sabermetrics methods no longer serves a purpose for most of us.

This is true of any field. It's probably simply a fact of life for any field, rather than an indictment of researchers in the field. Generalists get left behind. The things that are still unknown often aren't as interesting as they used to be in the sense that they become relatively arcane, or become impossible for a bright layman to discover or even conjecture usefully about.

If Bill James had been born in, say, 1990, he wouldn't currently be studying baseball statistics in a useful way, assuming he'd had the same education as Bill James v1 had through the age of 21.

I think he may just be misremembering the triples regression. IIRC (a big if) there was some well known regression ran on pre-1920s baseball, perhaps in "Hidden Game of Baseball" or some other similar work, where the triples coefficient came out GREATER than the HR coefficient.

There's a good reason for triples to reflect poorly on salary - salary looks forward and triples have a horrible aging curve. It's not even a curve, it's essentially a line going down from the start of a player's career.

Since it seems it may not be clear, I was only speculating that it was the Gassko salary article that the author was remembering. It seemed the appropriate thing that came up with a google of "triple negative coefficient baseball" or some such. And we all mis-remember stuff like that.

But, yeah, the possibility raised in #29 could be it as well. Triples might well peak in low-scoring eras (no HRs) and you might get a negative bivariate correlation (or even partial correlation depending on what you controlled for).

Anyway, without an actual cite, we probably should wait for him to clarify.

#40 Does salary in fact look forward? I know payroll doesn't and I'd be fairly surprised to find that the components of payroll turn out to look forward.

#39 the big difference I found between pre and post 1920 baseball is that before 1920 batter strikeouts were in fact both negative and significant while after that you really need not concern yourself with batter strikeouts in run scoring models. (that is in distinguishing among various types of outs. If you break out DPs separately then Ks matter very slightly. Not clear how it works it you adjust DPs for opportunities but I doubt it matters)

It's tricky to model deadball offense. So many errors. So many baserunning outs. But including Ks in any model helps.

My memory, which I haven't time to check out, is that Bill James, in one of the early ABSTRACTS, ran a correlation of different offensive stats with runs scored, and triples correlated negatively. If I remember the study right, that could be the one the author was thinking of. Also, Paul Waner hit lots of triples because he played in an old-style asymmetrical ballpark where left field was so big that you essentially had to have two center fielders (you can make a case tht Paul was the second best center fiedler in the NL of his day, behind only his brother). If you hit a ball hard down into that death valley, and you could run, you were going to hit a lot of triples. Musial is different, Cardsfanboy has it wrong (he's often right, just not here). Musial came up with the nickname "Donora Greyhound." That means that the people who saw him developing into a MLB star were more impressed with his speed than anything else. He would likely have been a career CF except that the Cards already had Terry Moore out there. His ballpark had an odd right field. It was small, but had a screen in front of the right field stands, like the Green Monster, but in right field, and a screen. Because it was a screen, there was no bounceback. The ball just fell down to the bottom of the fence with a thud. Stan got triples by running those out to third when a RF with a weak arm was playing for the opponents. That and he hit the wall/screen an awfully large amount of the time.

As to the main point, I'm not sure because the author is less than clear. But he does sound like he sees some of what I see. What I see in sabermetrics is 1) everyone can do the math (which is why the uberstats are almost all done by teams, instead of one person). 2) The differences between methods almost always comes down to either a) lack of a robust enough data set to make any real starting assumptions possible (see 19th century ball), or that there is a disagreement between the math teams about standards. This may be what the author thinks of as "applied philosophy." His term is silly, because the correct field is, very obviously, applied mathematics.

The only arena in baseball where double-blind placebo-controlled studies are needed and VERY lacking is in trying to actually figure out what steroids did and did not do. Good luck getting 30 matched pairs of MLB players and telling them that one guy will get the steroids and another will get the placebo. They've got games to play, stats to pile up, and pennants to win. They're not going to take the risk that steroids do work and they got the placebo. Other than that, baseball is just like analyzing any other historical collection of data, which you cannot change or arrange into controlled study groups. That's what applied math is for, not applied philosophy. - Brock Hanke

Carl Crawford is on a similar pace, but he's dropped off in recent years and is now stuck in an awful park for triples.

One might think Fenway would be a good triples park for a speedy LHB with good line drive power. Lots of room in right center and tricky angles near Pesky's Pole to bounce would-be doubles past the RF for an extra base. (Not that this proves anything, but Ted Williams' 71 career triples [35 at home] would place 9th among active players.)

My baseball interest began when Musial was in his mid 30s, so the speedy version (however speedy that was) had passed by. However, he didn't get much triples help from his home field, getting 90 there and 87 on the road. Oddly, considering his precisely even H/A split for hits, he was better at home for all kinds of EBH:
2B: 394/333
HR: 252/223
I think The Man's ability to hit loads of screaming LD to all fields has as much to do with his triples as his speed.

More or less a dup. Old thread.