Why trade Roberts?

[Old#5fan]

So...you never think ahead about or plan anything that you are going to do later in the day, or the week, or the year because of all of the infinite number of variables that could cause the failure of your prediction?

Sure I do and 99% of the time it either changes or doesn't go as planned. So to do so is a waste of time. If it wasn't part of my job I wouldn't do it.

[RShack]

Two points here.

First, what a sabermetrician is, is a baseball fan, with many of the same notions, preconceived ideas, and educated guesses as the casual fan you mention above.

But the thing that separates the sabermetrician from the casual fan, is that the sabermetrician actually scours the data (imperfect as it may be), applies proven statistical theory and methods, and along the way develops new metrics as appropriate, to test those preconceived ideas and educated guesses to see which ones pass muster, and which ones don't.

So the casual fan simply says, "the Cubs will score more runs with Roberts first and Soriano second," because intuitively, that's what he guesses is true. The end.

The sabermetrician may have the same intuititive sense, but his interest is in actually testing that hypothesis to see if a rigorous, objective statistical analysis will confirm or disprove it.

And in this case, at least one sabermetric tool that has been developed to analyze this very question (lineup optimization) does indeed illustrate that the conventional wisdom and "educated guess" of the casual fan is flat out wrong in the immediate case, so long as we're operating under the assumption that Soriano's production falls if he cedes the leadoff spot to Roberts.

The second point is this. Several posters have thrown around the notion of "predicting" the future with "certainty". It ought to be blatantly obvious, but apparently it needs to be said: the point is to illustrate what should be *expected* to happen under a given set of circumstances, not what absolutely *will* happen under those circumstances. The notion that this (or any) sort of analysis has no value unless/until it can deliver a 99% accuracy rate is absolutely ludicrous.

You're blurring completely different issues together in a way that's not right.

*Good* sabremetricians try do what you say they do. However, not everybody who makes reference to SABR stuff is a good sabremetrician. A good sabremetrician would never claim that the lineup analyzer tool can do a good job of actually predicting run production for those 2 or 3 different versions of a given team's lineup in a given season. Just like a *good* psychologist would never say that a "personality analyzer tool" can tell you how Person-X is gonna turn out in his real life. However, you can certainly find crappy psychologists (or 2nd year psychology students, or a researcher who's so into his research that he's completely lost perspective) who would make that claim, just like you seem to be doing with the lineup analyzer tool. There's a cliche about a little bit of knowledge being dangerous, and it's right.

You're trying to equate the value of SABR guys trying to develop knowledge with the practical utility of a very, very early-and-primitive lineup analyzer tool that's never been validated, that is clearly based on highly dubious assumptions, and that has no way to cope with various real aspects of real baseball. That's a bogus thing to do. It's exactly like equating the work of good economists with the utility of some early-AI stock-market predictor tool, or the work of good meteorologists with the utility of current hurricane prediction tools. It's just a dumb argument to make.

[davearm]

You're blurring completely different issues together in a way that's not right.

*Good* sabremetricians try do what you say they do. However, not everybody who makes reference to SABR stuff is a good sabremetrician. A good sabremetrician would never claim that the lineup analyzer tool can do a good job of actually predicting run production for those 2 or 3 different versions of a given team's lineup in a given season. Just like a *good* psychologist would never say that a "personality analyzer tool" can tell you how Person-X is gonna turn out in his real life. However, you can certainly find crappy psychologists (or 2nd year psychology students, or a researcher who's so into his research that he's completely lost perspective) who would make that claim, just like you seem to be doing with the lineup analyzer tool. There's a cliche about a little bit of knowledge being dangerous, and it's right.

You're trying to equate the value of SABR guys trying to develop knowledge with the practical utility of a very, very early-and-primitive lineup analyzer tool that's never been validated, that is clearly based on highly dubious assumptions, and that has no way to cope with various real aspects of real baseball. That's a bogus thing to do. It's exactly like equating the work of good economists with the utility of some early-AI stock-market predictor tool, or the work of good meteorologists with the utility of current hurricane prediction tools. It's just a dumb argument to make.

You say these things without any understanding of the model itself. And in doing so you look like a fool.

If you actually took the time to look at the links I gave you, you would realize that the heart of this lineup analyzer tool is a pretty standard regression model.

Regression analysis is not in any way shape or form primitive. It's a widely accepted and broadly applied statistical method.

There really aren't any assumptions involved, dubious or otherwise.

All we're seeking to do is quantify how certain offensive statistics in each position of the order correlate with a team's propensity to score runs.

You can try and paint the situation as being imminently more complicated than that, with "real aspects of real baseball" that must be explicitly modeled, but really it's not.

All we really need to know to have a useful and reliable model is how much run-scoring goes up and down with changes in a properly-selected subset of independent variables. Which is precisely what this lineup optimizer gives us.

[thundercleetz]

Good thing we didn't trade Roberts. The season is still young, but Roberts is doing exactly what we knew he would: getting on base, scoring runs, stealing bases, and playing solid defense. None of the rumored Cubs are looking too hot. Even Pie is looking very bad. The sad thing about Pie is he is probably going to continue to struggle if the Cubs keep playing him only once every few days. Like Adam Jones, he is going to need a large amount of at-bats to blossom into a solid major leaguer.

[Since'54]

You say these things without any understanding of the model itself. And in doing so you look like a fool.

If you actually took the time to look at the links I gave you, you would realize that the heart of this lineup analyzer tool is a pretty standard regression model.

Regression analysis is not in any way shape or form primitive. It's a widely accepted and broadly applied statistical method.

There really aren't any assumptions involved, dubious or otherwise.

All we're seeking to do is quantify how certain offensive statistics in each position of the order correlate with a team's propensity to score runs.

You can try and paint the situation as being imminently more complicated than that, with "real aspects of real baseball" that must be explicitly modeled, but really it's not.

All we really need to know to have a useful and reliable model is how much run-scoring goes up and down with changes in a properly-selected subset of independent variables. Which is precisely what this lineup optimizer gives us.

There's a lot of us who don't agree with you who are pretty sure we aren't fools. And we know for certain that you don't successfully reinforce your argument by insulting us.

[DrungoHazewood]

That doesn't mean they can't learn something. There's 2 issues here: whether their efforts can pay off with additions to the knowledge base vs. whether their lineup analyzer tool can be trusted. Those are completely different questions. (Same thing with hurricane prediction.) Why are you blurring them together? Don't you realize the difference?

You haven't offered any constructive criticism of the lineup analyzer besides throwing up your hands and saying it's too complex a problem to be analyzed. Ok, you did say this:

* That a baseball game is simply a series of discrete events (in this case, AB's) in which prior performance is a precise determinant of future performance, and

* That you can adequately model different lineups by simply taking each player's stats from whatever past-situations, and simply daisy-chain them together in different sequences and get a valid result, as if there are no interdependencies involved.

But you didn't bother to provide anything to back up your assumptions or generalizations. Do you have any real data that says the analyzer tool is inaccurate? Since the tool is based on a regression analysis of real results from real major league games, I'd believe it before I believed someone who just said "I don't think it works."

I tend to believe George Box:

"All models are wrong, but some are useful" - George E. P. Box

You think all models are wrong, and that makes them useless.

This all comes back to you needing to distrust and discredit analysis because you don't believe, or don't want to believe, the results. You've set a farcical, ridiculous standard that sabermetrics can't possibly approach and then declared victory when it can't meet that standard.

In reality, imperfect models can and do offer invaluable insight into problems without being able to act as a crystal ball.

[Owatcher]

Most of us Cubs' fans would rather Soriano hit in the middle of the order, but for whatever reason he feels more comfortable and produces better at leadoff. One of the reasons that we want Roberts is to bat leadoff and give Soriano a better chance to feel comfortable in the middle of the order. Since we don't have a "traditional" leadoff man, Lou might as well let Soriano bat first.

The Cubs aren't getting Roberts so I don't see how discussing it any longer can't be considered anything less than "disturbingly obsessive"!. The Cubs aren't getting Roberts. They're not interested in giving up what AM is willing to accept to complete the trade. If they were, Roberts would have been a Cub on opening day and hitting where ever Lou wants. 1st, 2nd, 3rd, 4th, 5th, 6th, 7th, 8th or 9th. Who cares really? That's a topic for yet another Roberts thread I'm sure.

The O's should trade Roberts for whatever package AM decides is adequate. There is an obvious (and understandable) bias underlying the discussion here and it's getting rather old. The Cubs obviously need to offer more than what has been rumored on this site. Until that happens, no trade with the Cubs is going to happen so further discussion is insanely futile.

We all know why the O's should trade Roberts and what impact it will have on the Oriole's future. That's whay we have to maximize it! Now get a life everyone. Or atleast get Hendry to trade Pie, Gallagher, and ...

[RShack]

You think all models are wrong, and that makes them useless.

This all comes back to you needing to distrust and discredit analysis because you don't believe, or don't want to believe, the results. You've set a farcical, ridiculous standard that sabermetrics can't possibly approach and then declared victory when it can't meet that standard.

In reality, imperfect models can and do offer invaluable insight into problems without being able to act as a crystal ball.

Where do you get this stuff? No, I do not believe that all models are wrong. I know for a fact that some models are right, and that many imperfect models can be useful. In fact, I've done a lot of work with models of complex human phenomena. I know how hard it is to get good things from imperfect models, but I also know it's very possible to do if one is careful and works at it. Nor is it true that I "need to discredit analysis because I don't believe or don't want to believe the results". I want to believe whatever results have been demonstrated to be trustworthy. And, for the record, I've been a huge Bill James fan for 20+ years, so don't gimme this crap where I'm somehow against it when I'm not. It's just that I know the dangers of getting fooled by numbers, and I try to maintain some semblance of critical reasoning about them. In short, I am happy to believe those things that have been demonstrated to be valid and trustworthy.

Sadly, you don't seem to have that same criteria. You are on record as saying things that are absolutely absurd, like your claim that SABR stuff is like physics and that poor Drungo is like a physics prof who has to keep explaining the laws of physics. You also routinely claim that various predictions are reliable when there is nothing to back up such claims... except that they take the form of numbers. Frankly, I think you just don't understand very much about what science is and isn't, nor do you understand the huge difference between pioneering efforts in the *discovery phase* of a scientific effort and the actual laws that mature science eventually can identifiy. As a result, you routinely mistake speculative and tentative preliminary findings for being "laws" and then you get indignant when somebody points out the huge difference. I say this because time after time you make it clear that you think whatever the current state-of-the-art happens to be is worth trusting... even when there's no evidence that it's trustworthy. That's exactly the same kind of thinking that had medical people performing lobotomies on patients, sticking ice picks into their brains through their eye sockets to sever parts of their brains, back before they knew what the hell they were really doing.

I've tried reasoning with you about this about 100 times, and it's always the same thing: You start painting black-and-white pictures that just aren't true, and you put phony words in my mouth that have nothing to do with what I believe, ignore half of what I've said, and directly contradict statements I have made. And, when all else fails, you resort to making personal attacks, just to change the subject when you've got no actual point to make, except that you're mad. All I can say is that it's a free country and you can believe anything you like, no matter how groundless it might be. But don't get all mad at me simply because I'm not blindly swallowing it all hook, line and sinker like you do. And, just to be clear, the thing I'm not buying has nothing to do with the good efforts of good SABR people, it's the *misuse* of their work by people who keep trying to act like baseball stats are a cookbook for success when they're just not. What the SABR people are trying to do is just a whole lot harder than you think, and I betcha most of their cutting edge guys would tell you the same dang thing.

[Old#5fan]

I think stats obsessive people are like mad scientists. Brilliant but not grounded in reality and basically one step away from being off the deep end. The more I read the interchange between a handful of posters here I know my take isn't wrong on this. However, I find it humorous that somewhere along the way they lack the basic skills needed to watch a game of baseball and know what the heck they actually are seeing!

[RShack]

You say these things without any understanding of the model itself. And in doing so you look like a fool.

If you actually took the time to look at the links I gave you, you would realize that the heart of this lineup analyzer tool is a pretty standard regression model.

Regression analysis is not in any way shape or form primitive. It's a widely accepted and broadly applied statistical method.

I *did* read the links. Look, a regression analysis is *just* a statistical technique. That's all it is. It is not some Magic All-Purpose Truth Machine, it's just a means of processing data. It does nothing to address the problems I mentioned. It does nothing to compensate for the underlying assumptions. And it does nothing to compensate for the data they don't have. You're back to talking about the trees, not about the forest. They can do whatever stat techniques are advisable, and they can do them all 100% correctly and appropriately, and that does not mean that they are finding any truth that goes beyond the limits of their data and their assumptions.

And I am *not* criticizing the guys who did this. It's just that there are various things that they have no straightforward way to investigate, because they cannot do experiments. It's not their fault, it's just a limitation of their situation that's caused by the complexity of the phenomena they're trying to study. This is very, very similar to the problem that hurricane guys face: they're trying to build models in the absence of knowing what all the factors and variables are. So, not only do they not have the data they need, they're also not even sure exactly what data they need that they don't have. In both cases, they know a lot more now than they used to know, and they don't know nearly as much as they need to know, and they'd tell you the exact same thing. In both cases, they are in the *early stages* of figuring this stuff out. What about this is so hard to understand? Do you think hurricane predictions are trustworthy just because they're the best anybody knows how to do right now? It's the same basic thing.

[davearm]

I *did* read the links. Look, a regression analysis is *just* a statistical technique. That's all it is. It is not some Magic All-Purpose Truth Machine, it's just a means of processing data. It does nothing to address the problems I mentioned. It does nothing to compensate for the underlying assumptions. And it does nothing to compensate for the data they don't have. You're back to talking about the trees, not about the forest. They can do whatever stat techniques are advisable, and they can do them all 100% correctly and appropriately, and that does not mean that they are finding any truth that goes beyond the limits of their data and their assumptions.

And I am *not* criticizing the guys who did this. It's just that there are various things that they have no straightforward way to investigate, because they cannot do experiments. It's not their fault, it's just a limitation of their situation that's caused by the complexity of the phenomena they're trying to study. This is very, very similar to the problem that hurricane guys face: they're trying to build models in the absence of knowing what all the factors and variables are. So, not only do they not have the data they need, they're also not even sure exactly what data they need that they don't have. In both cases, they know a lot more now than they used to know, and they don't know nearly as much as they need to know, and they'd tell you the exact same thing. In both cases, they are in the *early stages* of figuring this stuff out. What about this is so hard to understand? Do you think hurricane predictions are trustworthy just because they're the best anybody knows how to do right now? It's the same basic thing.

Regression analysis is *just* a statistical technique, sure -- one that is widely accepted and broadly applied to answer exactly the sort of question being posed here.

It's a fundamental tool in statistical analysis expressly because it's powerful, it's proven, and it overcomes the sort of nitpicky minutae you keep getting bogged down in.

Does care need to be taken to specify the form of the regression appropriately? Of course. That's why many alternate forms of the regression equation are tested and assessed, and then the one that delivers the best fit is chosen.

Is the regression model dependent upon the quality of the data being fed into it? Naturally. This is not a problem here, since there are thousands of team-years available (every MLB team's season from 1958 to 2004 is used in the latest analysis), and any metric of importance is readily obtained.

Are there traps to be avoided, such as multicollinearity and heteroskedasticity? Sure. Fortunately those are easily tested for, and effective controls exist should they be a problem.

I found it awfully telling that you chose to delete the last half of the post you responded to:

All we're seeking to do is quantify how certain offensive statistics in each position of the order correlate with a team's propensity to score runs.

You can try and paint the situation as being imminently more complicated than that, with "real aspects of real baseball" that must be explicitly modeled, but really it's not.

All we really need to know to have a useful and reliable model is how much run-scoring goes up and down with changes in a properly-selected subset of independent variables. Which is precisely what this lineup optimizer gives us.

All of your handwringing about the complexity of the problem, and babbling about hurricane modeling doesn't change these fundamental truths.

This lineup tool is neither a crystal ball, nor a silver bullet, nor a "Magic All-Purpose Truth Machine." Nobody has suggested it is, or even is intended to be. That would be preposterous, of course, just like your suggestion that it's useless since it's not all of these things.

[allstar1579]

Not that I don't appreciate statistical analysis in baseball, because I do, but when you show me a statistical tool that predicted last week that the O's would be 5-1 right now, then I will believe that one can accurately predict player performance year to year.

[RShack]

Regression analysis is *just* a statistical technique, sure -- one that is widely accepted and broadly applied to answer exactly the sort of question being posed here.

It's a fundamental tool in statistical analysis expressly because it's powerful, it's proven, and it overcomes the sort of nitpicky minutae you keep getting bogged down in.

Does care need to be taken to specify the form of the regression appropriately? Of course. That's why many alternate forms of the regression equation are tested and assessed, and then the one that delivers the best fit is chosen.

Is the regression model dependent upon the quality of the data being fed into it? Naturally. This is not a problem here, since there are thousands of team-years available (every MLB team's season from 1958 to 2004 is used in the latest analysis), and any metric of importance is readily obtained.

Are there traps to be avoided, such as multicollinearity and heteroskedasticity? Sure. Fortunately those are easily tested for, and effective controls exist should they be a problem.

I feel like I'm talking to the damn wall.

Look, you're talking about apples and I'm talking about oranges. I am not doubting that appropriate care was taken with their statistical techniques. That is not the issue. The issue is that statistical techniques do not manipulate Actual Ballplayers. All they do is manipulate statistics that represent a limited subset of ballplayer phenomena at limited resolution.

Everything you're talking about concerns what happens after you enter Stat World. It would be perfectly reasonable *if* we were talking about a computer game where ballplayers were automata whose behaviors were governed by statistical models. But ballplayers are not automata that are deterministically driven by stat calculation. Reality is the reverse of that : ballplayers create stats, stats don't create ballplayers. Your entire story is based 100% on the notion that stats adequately model what ballplayers do. If you could demonstrate that is actually true, then you'd have a case.

Which is 100% of the reason that I brought up the issue of predictive accuracy. Until you can demonstrate that the statistical models can predict what actually happens in a baseball season, your huge assumption is not supported. All you've got is a *theory* without empirical confirmation. Until you have empirical confirmation, it's an unsubstantiated claim. You don't know how many runs different lineups of BRob and Soriano and the rest of the bunch will score if you rearrange the lineup. Nobody does. The most you can say is that, given the data we have, and given the assumptions we are making, and assuming that there's not other stuff going on that our stats don't see, then our best guess is that they'll produce X-runs this way and Y-runs that way. But it's just a guess. Nobody knows if the assumptions are correct, and nobody knows what parts of reality matter that the stats don't see.

The way to demonstrate that the model is adequate, the way to demonstrate that stats see everything that matters, is to use the model to make predictions that hold up. Absent that, you're just making a claim that your Abstract Stat World adequately models the Actual World of Real Ballplayers Playing Real Baseball. That claim is predicated on the bogus assumption that computers in Our World act like they do on Star Trek. They don't. Look, we can't even get computers to tell UPS how to load their delivery trucks and route them in the most efficient way. We can't get computers to do that, no matter how fast they work. Even if they work at the speed of light, we still can't do it. And that's a circumstance when we *know* what all the factors are. When it comes to lineups, we don't even know what all the factors are, all we know is the small number of kinds-of-stats we have.

What the simulation guys are doing is coming up with the best thing they can using the stats they've got. They never picked the data points that baseball stats track, they're pretty much stuck with them. They're doing the best they can with what they've got, which is an admirable thing to do, and they're managing to figure out some things. But that's a completely different thing than saying the best they can do adequately models Reality. It's apples and oranges *until* you can prove it. Until you can prove it, it's just an unsubstantiated claim. What's crazy is to say, "This is our claim, and we can't prove it's right, but we're just gonna say it is, and we're gonna say that it's somehow up to other people to prove it's wrong." That's BS anti-science. If your claim is that the lineup analyzer can predict how many runs different lineups will produce, then you need to prove it by showing that it actually does that with Actual Lineups in Actual Baseball. Until it can do that at a high level, all you've got is an unproven theory that works inside of Stat World. So, if we're supposed to buy your claim that it can predict how Imaginary Lineups will do with everybody hitting in slots they've never hit in, then you oughta be able to tell us how great it has it proven to be at predicting exactly how Real Lineups do. How good is at doing that? Or is it only good at "predicting" things that never happen?

I found it awfully telling that you chose to delete the last half of the post you responded to:

All we're seeking to do is quantify how certain offensive statistics in each position of the order correlate with a team's propensity to score runs.

You can try and paint the situation as being imminently more complicated than that, with "real aspects of real baseball" that must be explicitly modeled, but really it's not.

All we really need to know to have a useful and reliable model is how much run-scoring goes up and down with changes in a properly-selected subset of independent variables. Which is precisely what this lineup optimizer gives us.

OK, I'll respond: What you just said is wrong. It's wrong to say it's not a complex problem and that they don't need a bunch of different data points. Rather, they just don't have many data points. What they have is a big sample of not-enough data points. That's what they're stuck with. The bolded part is wrong too. The concept of "independent variable" and "dependent variable" comes from the domain of doing Actual Experiments. You manipulate the independent variable, and then you see what that does to the dependent variable. You're not doing Actual Experiments. You're not changing actual batting orders and then seeing what happens to the team. What you're doing is going inside the closed abstraction of Stat World and you're finding out what happens to the Imaginary Team that lives inside of Stat World. That doesn't tell us what happens in the Real World. You're confusing Simulated Reality with Actual Reality, and acting like they're the same thing. They are not the same thing.

All of your handwringing about the complexity of the problem, and babbling about hurricane modeling doesn't change these fundamental truths.

This lineup tool is neither a crystal ball, nor a silver bullet, nor a "Magic All-Purpose Truth Machine." Nobody has suggested it is, or even is intended to be. That would be preposterous, of course, just like your suggestion that it's useless since it's not all of these things.

I'm not "handwringing", I'm simply saying that you're wrong, that's all. Because you are. You talk about "fundamental truths" but then you use the concept of independent variable to make it sound like you've proved something when you haven't. Even *you* know you're wrong, if you'd just stop and climb out of Stat World for just a minute. If you did that, you'd recognize that you're talking out of both sides of your mouth at once:

• On the one hand, you claim that Soriano hits better at the top of the order than in the heart of the order, you don't know why, you just know he does. It's one of those fuzzy human factors that we don't understand.
• But on the other hand, you claim that Stat World can tell us how everybody's gonna perform in all the different lineup slots, including the slots where we have zilch data about how they perform there. So, which one is it? We have fuzzy human factors that we don't understand, but that matter a lot anyway? Or we don't? Make up your mind, you can't have it both ways.

  .

• On the one hand, you say that nobody has suggested that it is a Truth Machine or a Crystal Ball...
• But on the other hand, you stated that having BRob leading off is gonna cost the Cubs between 15 and 26 runs. So, you *say* it's not a Crystal Ball, you pay empty lip service to that idea when somebody calls you on it, but the rest of the time you act like it *is* a Crystal Ball...and then you have a fit and start attacking people when they simply say that it's not. Make up your mind, you can't have it both ways.

This is exactly the kind of BS that gives the good work that serious people do with stats a bad name. For the most part, it's not the people who do the actual research who give it a bad name. I've known and worked with some serious stat people, and in my experience, they don't do the crazy things you and Drungo do with this stuff. For the most part, they're acutely aware of the holes in their work and, unless they're trying to sell it in a grant application or something, they're scrupulous about alerting everybody to the pitfalls. They don't BS anybody, they just work at it, and the worst thing they ever do is simply be wrong sometimes. The bad name comes from "stat-consumers", SABR-wannabes who don't actually *do* the real work, but who take the hard work done by the guys who actually do research, use it in the wrong way, and claim that it's suitable for purposes that it's not suitable for. It's not the serious SABR-research guys who get it wrong, it's the SABR-tourists who do. A prime example is what you said above: I never said that serious simulations were "useless". I said that they're a toy for the purposes that you are misusing them for. I'm sure they have utility to researchers who study what the simulations tell them, who ID where the simulations are strong vs. weak, and who take that analysis and go back to the drawing board and work hard to make the next-generation simulations better. I'm sure it's useful to them, because they're trying to do work and actually learn something; they're not just hijacking other people's work and misusing it as bogus ammunition in multi-month pissing contests about hypothetical BRob trades.

BTW, according the lineup analyzer, when BRob is dancing around off of 1B, driving the P nuts and getting one of the MI's to shade towards 2B, exactly how much effect does that have on Soriano's hitting when he bats 2nd? If you can tell me how many runs that lineup is gonna cost the Cubs, then surely you can tell me how that factors into it. Right?

[Old#5fan]

I tried to read a couple of these posts and much like trying to read the Internal Revenue Code I quickly dozed off!

[TinCup]

I feel like I'm talking to the damn wall.

Look, you're talking about apples and I'm talking about oranges. I am not doubting that appropriate care was taken with their statistical techniques. That is not the issue. The issue is that statistical techniques do not manipulate Actual Ballplayers. All they do is manipulate statistics that represent a limited subset of ballplayer phenomena at limited resolution.

Everything you're talking about concerns what happens after you enter Stat World. It would be perfectly reasonable *if* we were talking about a computer game where ballplayers were automata whose behaviors were governed by statistical models. But ballplayers are not automata that are deterministically driven by stat calculation. Reality is the reverse of that : ballplayers create stats, stats don't create ballplayers. Your entire story is based 100% on the notion that stats adequately model what ballplayers do. If you could demonstrate that is actually true, then you'd have a case.

Which is 100% of the reason that I brought up the issue of predictive accuracy. Until you can demonstrate that the statistical models can predict what actually happens in a baseball season, your huge assumption is not supported. All you've got is a *theory* without empirical confirmation. Until you have empirical confirmation, it's an unsubstantiated claim. You don't know how many runs different lineups of BRob and Soriano and the rest of the bunch will score if you rearrange the lineup. Nobody does. The most you can say is that, given the data we have, and given the assumptions we are making, and assuming that there's not other stuff going on that our stats don't see, then our best guess is that they'll produce X-runs this way and Y-runs that way. But it's just a guess. Nobody knows if the assumptions are correct, and nobody knows what parts of reality matter that the stats don't see.

The way to demonstrate that the model is adequate, the way to demonstrate that stats see everything that matters, is to use the model to make predictions that hold up. Absent that, you're just making a claim that your Abstract Stat World adequately models the Actual World of Real Ballplayers Playing Real Baseball. That claim is predicated on the bogus assumption that computers in Our World act like they do on Star Trek. They don't. Look, we can't even get computers to tell UPS how to load their delivery trucks and route them in the most efficient way. We can't get computers to do that, no matter how fast they work. Even if they work at the speed of light, we still can't do it. And that's a circumstance when we *know* what all the factors are. When it comes to lineups, we don't even know what all the factors are, all we know is the small number of kinds-of-stats we have.

What the simulation guys are doing is coming up with the best thing they can using the stats they've got. They never picked the data points that baseball stats track, they're pretty much stuck with them. They're doing the best they can with what they've got, which is an admirable thing to do, and they're managing to figure out some things. But that's a completely different thing than saying the best they can do adequately models Reality. It's apples and oranges *until* you can prove it. Until you can prove it, it's just an unsubstantiated claim. What's crazy is to say, "This is our claim, and we can't prove it's right, but we're just gonna say it is, and we're gonna say that it's somehow up to other people to prove it's wrong." That's BS anti-science. If your claim is that the lineup analyzer can predict how many runs different lineups will produce, then you need to prove it by showing that it actually does that with Actual Lineups in Actual Baseball. Until it can do that at a high level, all you've got is an unproven theory that works inside of Stat World. So, if we're supposed to buy your claim that it can predict how Imaginary Lineups will do with everybody hitting in slots they've never hit in, then you oughta be able to tell us how great it has it proven to be at predicting exactly how Real Lineups do. How good is at doing that? Or is it only good at "predicting" things that never happen?

OK, I'll respond: What you just said is wrong. It's wrong to say it's not a complex problem and that they don't need a bunch of different data points. Rather, they just don't have many data points. What they have is a big sample of not-enough data points. That's what they're stuck with. The bolded part is wrong too. The concept of "independent variable" and "dependent variable" comes from the domain of doing Actual Experiments. You manipulate the independent variable, and then you see what that does to the dependent variable. You're not doing Actual Experiments. You're not changing actual batting orders and then seeing what happens to the team. What you're doing is going inside the closed abstraction of Stat World and you're finding out what happens to the Imaginary Team that lives inside of Stat World. That doesn't tell us what happens in the Real World. You're confusing Simulated Reality with Actual Reality, and acting like they're the same thing. They are not the same thing.

I'm not "handwringing", I'm simply saying that you're wrong, that's all. Because you are. You talk about "fundamental truths" but then you use the concept of independent variable to make it sound like you've proved something when you haven't. Even *you* know you're wrong, if you'd just stop and climb out of Stat World for just a minute. If you did that, you'd recognize that you're talking out of both sides of your mouth at once:

• On the one hand, you claim that Soriano hits better at the top of the order than in the heart of the order, you don't know why, you just know he does. It's one of those fuzzy human factors that we don't understand.
• But on the other hand, you claim that Stat World can tell us how everybody's gonna perform in all the different lineup slots, including the slots where we have zilch data about how they perform there. So, which one is it? We have fuzzy human factors that we don't understand, but that matter a lot anyway? Or we don't? Make up your mind, you can't have it both ways.

  .

• On the one hand, you say that nobody has suggested that it is a Truth Machine or a Crystal Ball...
• But on the other hand, you stated that having BRob leading off is gonna cost the Cubs between 15 and 26 runs. So, you *say* it's not a Crystal Ball, you pay empty lip service to that idea when somebody calls you on it, but the rest of the time you act like it *is* a Crystal Ball...and then you have a fit and start attacking people when they simply say that it's not. Make up your mind, you can't have it both ways.

This is exactly the kind of BS that gives the good work that serious people do with stats a bad name. For the most part, it's not the people who do the actual research who give it a bad name. I've known and worked with some serious stat people, and in my experience, they don't do the crazy things you and Drungo do with this stuff. For the most part, they're acutely aware of the holes in their work and, unless they're trying to sell it in a grant application or something, they're scrupulous about alerting everybody to the pitfalls. They don't BS anybody, they just work at it, and the worst thing they ever do is simply be wrong sometimes. The bad name comes from "stat-consumers", SABR-wannabes who don't actually *do* the real work, but who take the hard work done by the guys who actually do research, use it in the wrong way, and claim that it's suitable for purposes that it's not suitable for. It's not the serious SABR-research guys who get it wrong, it's the SABR-tourists who do. A prime example is what you said above: I never said that serious simulations were "useless". I said that they're a toy for the purposes that you are misusing them for. I'm sure they have utility to researchers who study what the simulations tell them, who ID where the simulations are strong vs. weak, and who take that analysis and go back to the drawing board and work hard to make the next-generation simulations better. I'm sure it's useful to them, because they're trying to do work and actually learn something; they're not just hijacking other people's work and misusing it as bogus ammunition in multi-month pissing contests about hypothetical BRob trades.

BTW, according the lineup analyzer, when BRob is dancing around off of 1B, driving the P nuts and getting one of the MI's to shade towards 2B, exactly how much effect does that have on Soriano's hitting when he bats 2nd? If you can tell me how many runs that lineup is gonna cost the Cubs, then surely you can tell me how that factors into it. Right?

Inquiring minds want to know Shack....how long does it take you to come up with these posts and type them? Seriously. I am curious.