I, for one, think that a lot of stat heads to treat Sabermetrics as an ideology. Obviously not all, but a lot.

As to whether the correct term is an ideology or a philosophical mindset, well, my mother was a professor of linguistics and of semiotics, and I frankly never got into the arcane discussions she got into, so I'm not going there now.

one has to buy in to the prior assumption that the world can be analyzed in an objective fashion and that outcomes in the world are the result of different distributions of individual, structural, and random causes.

Who would dispute that the world can be analyzed in such a manner?

I think it's very interesting that Bill James doesn't even consider things like WAR to be sabermetrics. He uses the term for acquiring new objective knowledge about baseball, but sees WAR as just reshuffling old knowledge we already had.

In other words, Sabermetrics would say that a player’s WAR adds no new information, once you know all these other things.

Who would dispute that the world can be analyzed in such a manner?

I think a lot of people. Objectivity itself has a long history (as Daston and Galison showed in their book Objectivity), and it is not clear that statistics tell us the truth so much as one version of the truth. You can see it in politics and everywhere: statistics tell the story you want them to tell.

statistics tell the story you want them to tell.

I don't agree with this. I think that is true if the statistics are used incorrectly either by lack of understanding or in a willfully misleading way. If I say Jim Rice hit more home runs than Ty Cobb, that is a fact and there is no way to make that statistic tell a different story. If I say Jim Rice hit more home runs than Ty Cobb and is therefore the superior player well...

Right, and facts are not stories. The stories are about the use and the production of facts.

Why make this harder than you have to?

The ideal "sabermetric mind set" would be one which contained far more questions than answers. Bill James is the perfect embodiment of this.

Unfortunately, the "sabermetric mind set" that all too often is the public face of sabermetrics is one that's more interested in arguing with Murray Chass than it is in asking questions.

Sabermetrics is a language set by which a type of speaker imposes a narrative on the world. As such, in as much as language itself is a tool, it's a tool. How the tool is deployed is fundamentally aligned with an ideology.

Tom Tango doesn't have a secret path to true knowledge of the world.

Who would dispute that the world can be analyzed in such a manner?

I think you could argue that there are so many variables that it is difficult, if not impossible to ferret out true causation in real-life, non-experimental situations. Baseball lends itself nicely to objective analysis because each hitter gets individual at-bats, but there are still variables that come into play that can cloud things like place in the batting order, pitchers, ballparks, weather, injuries, team chemistry - we've tried to account for a lot of this, but who knows how much we're missing.

The ideal "sabermetric mind set" would be one which contained far more questions than answers. Bill James is the perfect embodiment of this.

Unfortunately, the "sabermetric mind set" that all too often is the public face of sabermetrics is one that's more interested in arguing with Murray Chass than it is in asking questions.

The second is not the "sabermetric mind set" in any way, shape or form. It is the "stathead" mind set, which is a related but ultimately very different thing.

I made this distinction years ago on this site in reference to an article that said something like "Alfonso Soriano confounds sabermetrics." No. He may have confounded statheads, but the only way he could confound sabermetrics, in my reading of the word, is if he literally went about hacking our hard drives and corrupting the information.

Why do most statheads agree that "first ballot HOF" is an irrelevant distinction? Why do they agree that the MVP should go to the best player, regardless of the team's record? These come from a type of logical way of looking at baseball that is related to and perhaps derives from sabermetrics ... but it is not sabermetrics.

Sabermetrics doesn't give a #### about Murray Chass.

#10 I think we've got a real good notion of the size of what's missing. We expect to get team wins to within 3-4 2/3 of the time knowing their runs scored and runs allowed (and we know that teams with excellent bullpens tend to out-perform their pythags by around 2 games, so the "true" standard error is probably about 3 wins) . We expect to get team runs scored within 15 runs 2/3 of the time knowing the counter stats of the offensive players. The runs allowed is a little mushier, but all in all I'd say 20 runs on the pitching and defense side. (one thing still not investigated adequately to my knowledge is the whole notion of clutch pitching. From a quick look there seems to be a real -- if not terribly important -- skill. When I regress team runs allowed as opposed to team runs scored I get slightly different results that looking at team runs scored)

Projecting uninjured players? Last time I checked the standard error was just over 10 runs for full time players.

So: We can't do anything about injuries. We only have the most basic knowledge of which young players will in fact improve. And we can model to within spitting distance. Improving the models isn't going to help a great deal with the same inputs. I mean you might chop a run off the team standard error by juggling the offensive stats better. No biggie.

There's room for improvement on the run prevention side (even if you're unable to completely untangle pitching and defense) but not a lot -- again given the same inputs as we have now.

I've been looking at the outliers at all levels for some time. I can't find anything, but that doesn't mean it's not there.

I think you could argue that there are so many variables that it is difficult, if not impossible to ferret out true causation in real-life, non-experimental situations.

Baby/bathwater, right? I think one of the triumphs of sabermetrics is linear weights - assigning run values to events that, over time, prove very accurate - weather and team chemistry do not mar this analysis.

Who would dispute that the world can be analyzed in such a manner?

I've had discussions with people who refuse to believe certain metrics because they forecast outcomes which cannot be proven with absolute certainty. For example, in order to see if player X is worth 2 wins over replacement, you can't run a real life season simultaneously and replace player X with a replacement player and see if there was indeed a two win difference. The same logic could be applied to defensive metrics and runs saved. Not that I agree that this sufficient cause to dismiss WAR or any other metric, but this is the type of resistance that exists.

I've had discussions with people who refuse to believe certain metrics because they forecast outcomes which cannot be proven with absolute certainty. For example, in order to see if player X is worth 2 wins over replacement, you can't run a real life season simultaneously and replace player X with a replacement player and see if there was indeed a two win difference. The same logic could be applied to defensive metrics and runs saved. Not that I agree that this sufficient cause to dismiss WAR or any other metric, but this is the type of resistance that exists.

But by that logic aren't all statistics as a measure of excellence flawed? Perhaps Batista simply sees easier pitches than any other hitter.

But by that logic aren't all statistics as a measure of excellence flawed?

Not counting stats.

this is the type of resistance that exists.

Well, in a sense they're right, in that WAR (as I understand it) seems to be model-based, making certain assumptions about value that are by definition not "real." However, drawing conclusions from well-built models is not controversial in most scientific fields. If they think that modelling itself is invalid, then they probably shouldn't listen to any weather forecasts or enter any man-made structures since their engineering may have required predictive modelling.

However, drawing conclusions from well-built models is not controversial in most scientific fields.

It's not only not controversial, it's practically a built in assumption and done daily. In the field of economics, even a basic concept like opportunity cost involves a degree of estimation that is often not (and sometimes cannot) verified in the manner that is proposed. If that is your bar of accepting baseball metrics, you might as well not accept a lot of research done in most scientific fields.

For example, in order to see if player X is worth 2 wins over replacement, you can't run a real life season simultaneously and replace player X with a replacement player and see if there was indeed a two win difference.

And even if you could, and found that there was a two-run difference when you did, one could make the argument that one trial doesn't prove anything.

drawing conclusions from well-built models is not controversial in most scientific fields.

How do you decide that the model is well-built? I think that's where the problem arises - when we develop a model that produces what appear to be irrational results to an observer (such as Ben Zobrist coming out as *better* than Prince Fielder), is the conclusion actually reasonable or is the issue with the model itself? That was, in essence, the question that Buster Olney was raising the other day, and I think that we have a really difficult time explaining why our model is well-built, in terms of its relationship to the real world of baseball.

-- MWE

I, for one, think that a lot of stat heads to treat Sabermetrics as an ideology. Obviously not all, but a lot.

Next you'll be suggesting that Jeff Francoeur is a better player than Brian Stamps.

I've had discussions with people who refuse to believe certain metrics because they forecast outcomes which cannot be proven with absolute certainty. For example, in order to see if player X is worth 2 wins over replacement, you can't run a real life season simultaneously and replace player X with a replacement player and see if there was indeed a two win difference. The same logic could be applied to defensive metrics and runs saved. Not that I agree that this sufficient cause to dismiss WAR or any other metric, but this is the type of resistance that exists.

That sort of resistance applies to statistics in general. Many a time I've heard someone pooh-pooh some political poll or other by saying "Do you know how many people they interviewed for that thing? Only 5,000!" in the sort of voice that indicates they'd be equally outraged if it was 50,000.

And even if you could, and found that there was a two-run difference when you did, one could make the argument that one trial doesn't prove anything.

Depends if you held everything else constant for every trial run (which is probably what is fairest to the spirit of original objection). If so, you'll get 2 wins every time!

That sort of resistance applies to statistics in general. Many a time I've heard someone pooh-pooh some political poll or other by saying "Do you know how many people they interviewed for that thing? Only 5,000!" in the sort of voice that indicates they'd be equally outraged if it was 50,000.

How do the statistically minded (I'll hold off on calling you ideologues just yet) here cope with the fact that statistical analysis doesn't really tell anyone much of anything useful with regard to day to day management, much less at bat to at bat strategy for hitters or pitchers?

#22 I'd be pleased to attempt to answer the questions, but I honestly don't understand what you're asking.

If I understand the last question properly, I'm extremely doubtful that the vast majority of hitters are capable of hitting strategically. Best I can tell, it's react in fractions of a second using the training of a lifetime.

Pitching? I'm totally open to the notion that strategic decision are of importance. As I've pointed out before the problem with simple adjustments to defensive stats based on the handedness of the staff can be summed up in the sentence, Al Leiter and Tom Glavine are both (successful) left-handed pitchers.

At one time Leiter was the most frequently pulled pitcher and Glavine the least and there's an aspect of freedom of choice in this. I would assume (hope) that teams are very heavily into this (and working on the implications of defensive positioning within the context of location.

There hasn't been a great deal of work done on this because the people likely to be interested in this kind of stuff (I'm really not, just thinking out loud) haven't had enough to work with. They probably do now and I'd anticipate interesting stuff coming out in the next few years. To date it's been mostly one off stories rather than systematic work, but it's only a question of time.

If I understand the last question properly, I'm extremely doubtful that the vast majority of hitters are capable of hitting strategically. Best I can tell, it's react in fractions of a second using the training of a lifetime.

This goes against everything anyone has ever said about hitting. If hitting isn't strategic, if it's nothing but read and react instinct, why does Chipper Jones and every other successful hitter, pretty much ever, talk about the importance of "going up there with a plan?"

To clarify my question, there is no statistical model which tells me which players to bat on a given night. Sure, you can make presumptions about probability based on long-season/sample aggregate data, but that fails to account for things like "who's seeing the ball well" and "who is completely lost at the plate mechanically."

To bring it down to cases, I'm currently embroiled at another site in a conversation/debate/argument/flame war about the validity of starting Jose Constanza ahead of Jason Heyward. I am firmly in the "Heyward's hopeless at the plate right now, you start the hot hand with Constanza" camp. (Some might say I *am* that camp entirely.) The arguments against Constanza always come back to "Heyward has better MLEs, and he had a historic year last year, while Constanza's a 27 year old no one had ever heard of on a hot streak." My position is that all of those things are true, but they are all utterly irrelevant to the question of which of those guys should start. Heyward's previous success literally has nothing to add to the conversation, because he is not currently that hitter.

To clarify my question, there is no statistical model which tells me which players to bat on a given night. Sure, you can make presumptions about probability based on long-season/sample aggregate data, but that fails to account for things like "who's seeing the ball well" and "who is completely lost at the plate mechanically."

You are looking for a statistical precision that does not exist. In simplest terms \"#### happens."

At some point you trust all the information at your disposal. That information includes both historical and recent performance. How do you balance the two? If I knew that I'd be selling the info to the highest bidder among 30 MLB teams. A single game is simply impossible to predict with any level of certainty but it's all about the odds, if Player A is better than Player B, Player A should be expected to have a better game tonight. That doesn't mean it's impossible that Player B will be better tonight and maybe due to injury or illness or off field distraction Player A is a lesser version of himself. OF COURSE that should be factored in, it would be stupid not to, but it should be factored in some ratio to the player's true talent, not as the only factor.

#24 Listen to the specifics when they talk about a plan. It really comes down to pitch recognition. Don't get fooled. Don't get out in front. Best I can tell hitters don't actually execute differently. (Oh they develop mechanical flaws and the like. They sometimes correct them. But they don't actually do anything different with a given pitch once it's been identified)

And #25 well there is. Or at least Earl Weaver or Casey Stengel sure believed there was. Stuff, handedness, park and a lot of other things enter into the equation (although in today's 17 man pitching staffs teams don't have a great deal of flexibility in the matter)

And you are aware that what we see as hot and cold streaks will happen in stratomatic or APBA or any sim. It's far more common for us to see a pattern that doesn't have any basis in ability level change than for something to have actually happened that's resulted in a change of ability level.

In general I adopt the null hypothesis on streaks (and there's enough research on this topic to make me confident that this is generally correct)

So to take your specific example of Costanza and Heyward, I think it's nuts to change my evaluation of Costanza's ability level. Voros' law is very much in play. But that doesn't mean he's a stiff. I think he ought to be expected to hit for a decent average, a few walks and zero power.

So I wouldn't play him over Heyward unless there's something specific that can be pointed to. It's tempting to suggest a platoon (I have zero doubt that even if Heyward's current stats represent his ability level he's the better offensive player against right-handed pitching) -- and I would in a tabletop league -- but that's something I think should be done on a case by case basis. Some young players who were platooned never really master hitting lefties (think Ryan Klesko or Andy Van Slyke -- both very talented hitters with unusually large platoon splits). Some step out of the platoon role with no particular problem. The Braves have to think strategically. What's in the long and short term interest of the team (yes, it can be argued that having him fail against lefties is also not in the long term interest of the team. I'm completely confident that he's not a .165/.260/.306 hitter versus lefties but I have no idea what he'll hit against them for the rest of the year)

when we develop a model that produces what appear to be irrational results to an observer (such as Ben Zobrist coming out as *better* than Prince Fielder), is the conclusion actually reasonable or is the issue with the model itself?

To me, that is not a surprising conclusion in the least. First of all, Ben Zobrist is having a heck of a year with the bat himself, 145 OPS+. He's not as good a hitter as Fielder, but close enough that his multiple advanages - far better baserunner, far better fielder, playing tougher positions - easily makes up for it, at least this season.