ABSOLUTELY, 1 million% wrong.
My statement has NOTHING to do with who plays tonight.
It has nothing to do with tonight, or the Yankees, or what actions should be taken.

This is your problem and why we are going in circles.
What I said…
DOES NOT (No No No NO NO)
“implies inherently that who plays tonight doesn’t matter.”

You are simply not getting what I’m saying, and constantly assuming assertions that I am NOT making.

Again, you are confusing my analytical statement for making out a lineup card.

But I know that you don’t believe that. I know that think it’s better to play better players than worse players, because it increases the chances of success tonight. Implicitly you DO believe that small samples are predictable, because you predict that the better players give you a better chance of success tonight, in just a few ABs.

But you refuse to give up the rhetorical structure you have created, and you adhere to certain key terms.

As for Jeter’s non-bunt, although the Old Captain is top-20 in double play percentage (17 percent of his chances, worst on the Yankees) giving away outs, as opposed to gambling on the better than 80 percent chance that a very good hitter WON’T hit into one, is not good managing. It was a poor decision by Joe Girardi which Jeter doubled down on by bunting foul with two strikes.

That strikes me as a fairly blanket judgment.

2 vastly DIFFERENT issues.

1) I say it’s random based on history (see ARod: career and PS performances) as well as seeing that day to day, a players performance can have vast fluctuations, and I can’t say which game he gets 3 hits or no hits.

2) However, who we play in an attempt to win has nothing to do with that. We can’t control or predict this randomness. However, we can make our BEST ATTEMPT at winning by putting the best players we have on the field. We play the odds. The odds are that better players have a better performance. It a best guess…. a smart guess, given what’s in our power. But playing our best players is not a prediction we will win .

Teams with the best players don’t always win and the best team does not always (less then 5o% I believe) win the WS. But knowing this doesn’t mean we don’t try.
You can’t predict baseball, right?

You are making TREMENDOUS assumptive leaps based on my statements.

I say: “In baseball, a singular large sample sized data statistic is not particularly predictive of the results of any individual singular event”. I am making an analytical statement.

And you take the meaning of this statement as:
It doesn’t matter who we put on the field, might as well start the scrubs?

Really?????? Are you just being argumentative?
I mean, I’m not that good at communicating my analytical thoughts, but man…………. how do you arrive at these conclusions????

They are. But then you take predictions that turn out to incorrect (Molina getting a couple of hits or Posada going 0-4, when the prediction suggests very different outcomes), and then declare that the process is “random” and thus largely invalidate the decision-making process.

Going back to the hypothetical situation of Posada and Molina: Posada should start tonight instead of Molina because large data sets show that he is clearly the better player. I am implicitly making a prediction that Posada will have a much better chance of contributing positively than will Molina. If they play tonight and Posada goes 0-4 or Molina starts and goes 3-4…i.e., that the prediction was incorrect on some level…does not mean that the thinking behind the prediction was wrong, or that the outcome was “random.”

I actually think that our archetypal Bill James thinks that his probabilities (or whatever you want to call them) are very broadly applicable and apply to most situations, yes.

To use the bunt, which you have adopted as an ongoing hypothetical example: yes, I think that the Bill Jame’s types will tend to think that the bunt is a “bad” play in the great many circumstances in which it has been used historically and continues to be used. They may not be “dogmatic” about it, but I bet they feel pretty strongly about such tactics based on their statistical analysis. I mean, read Rob Neyer…especially his older stuff. He was pretty willing to call out a manager for what he concluded was a bad tactic or strategy.

We average 80 inches of snow here every year (I’m making that number up). I could guess that this year, we will get at least 20″ of snow. Safe guess, yes? But I’m just guessing (even if I’m correct) and basing my guess on thay one piece of data above,

Guys who predict the weather, look at lots and lots more data, and do more complicated analysis. Weathermen don’t guess at the weather based on previous years. They study the science of weather. Their predictions are more accurate then my guesses. Yes?

What is outdated or mistaken? In who’s opinion? Is the sac bunt a mistake? Always? Never? Does it depend on the situation? These guys are developing formulas and crunching numbers to come up with general probabilities. Do you think Bill believes that his probablitlies are absolute, and apply in any and all situations? I don’t.

Yankster:

Average from large samples is more useful for predicting subsequent large sample averages. But the distribution of observations (which can be indicated by deviation from the mean) gives you a much better sense of the probability of a subsequent single event.

No, that’s not right. The average gives you a much better estimate of the probability of a single event than the standard deviation does. The standard deviation is useless.
If one player has batted .350 for the past five years and another has batted .200, and you want to know which one is more likely to get a hit tomorrow, you should rely on the averages. It makes absolutely no difference whether the player with a higher average has a large standard deviation in his average from year to year.

It’s random. We have NO idea how ARod will do in the next 5 games. It’s random. It’s random. It’s random. It’s random.

But it’s not “random.” If it’s random, then the Yankees should just pull names out of hat when writing out the lineup card for tonight’s game.

But you don’t REALLY believe that, do you? You know that the odds are better if they play better players than worse players—indeed, you admit as much above, when you say “Of course you play ARod, because after years of watching him and collecting data, we know he is a superior ballplayer.”

That very statement, the underlying assumption assumption denies your claim that it’s all “random” even in small samples.

Maybe we are just speaking past each other in terminology. I don’t think that you are using words like “random” or “predictive” or “small sample sizes” properly, but then maybe I am using them improperly.

To read most blogs, the answer seems to be between 50% and 100%
(Yup… Girardi made the RIGHT move there, because I said so. Yup, Girardi made the WRONG move there, because I said so.)
I object to people not only thinking they know more then Joe (and every other manager) and then actually being dogmatic about it, if/when someone provide another point of view.

One final comment on this thread for me. It is worth noting that some of the major advances in the analysis of the game (the development of new and better statistics, which you cite in one of your longer threads) have been developed by guys like Bill James who—before he was hired on by the Sox—had no major league experience.

I reject the implication that ONLY insiders are allowed to analyze and critique.

Former players like Timmy say stupid things when they are announcers—we all recognize this—and I see no reason why former players who are managing are immune from outdated or mistaken modes of thinking, or blinded by loyalty, etc.

Large sample sized Data allows us to make a ‘Best Guess’ for an action. Making a Best guess is indeed far, far better then making a poor guess, so it does matter and does have value. If I have no other qualifying data, I will ALWAYS play ARod over JHJr. Always. It’s a great guess. But it you want to be predictive… meaning actually being able to predict with some reasonable degree of accuracy the outcome of ONE very small sample…. forgetaboutit.

Of course you play ARod, because after years of watching him and collecting data, we know he is a superior ballplayer. Of course you play him….. and the Mick too.

But lets look at some REAL data. History. Stuff we don’t have to guess at, because it has already happened.

ARod has a career OPS of .965. This is a large sample size.
ARod had a 1.500 OPS in his last 2 PS series (small sample).
STATISTICALLY speaking, please show me a ‘predictive’ analogy to account for this.
ARod had a 525 OPS in his previous 2 PS series (small sample).
STATISTICALLY speaking, please show me a ‘predictive’ analogy to account for this.

It’s random. We have NO idea how ARod will do in the next 5 games. It’s random. It’s random. It’s random. It’s random. Wanna guess somewhere between .900 and 1.050? great. I agree. Good guess. Good. Guess.
But don’t bet the farm on it.

His career stats will not predict anything over a very small sample size. He could have a .525 OPS, or (coincidentally) have a .965 OPS, or a 1.500 OPS, or anything in between. It can’t be predicted.

However, over the next 5 years (large sample size), if ARod stays healthy, then my guess he will post similar numbers as his current career numbers, decremented by some sort of aging factor. I believe a large sample size DOES have some predictive accuracy when applied to another large sample size.

In the ALDS, Nick Pinto had a 1.139 OPS. How predictive was his .647 career OPS?
Jeff Mathis. 1.400 in the PS vs a .597 career OPS. Anybody predict that?

And who, precisely, has done that? Every person who posted for or against the move—which really seems to be at the core of your complaint about predictions and small sample sizes—cited (as I recall) several pieces of evidence:

On the offensive side: various averages, RC totals (which encompass several stats), probable numbers of ABs, short term trends (Posada was scuffling some), etc.

On the defensive side: AJ’s ERA with Molina catching, AJ’s best games caught this season (w/various catchers behind the plate), short term evidence (good and bad starts in the play offs), Molina’s superior ability against the running game, subjective evidence (trips to the mound, cross-ups) that suggest the quality of the relationship between C and P, and so forth.

Who criticized Girardi for using “one piece of information”? That’s a straw man.

That was a paraphrase, an interpretation of various statements like:

[51] “…that any small sample is somewhat random.”

[51] “he might have a .400 OBP (Large sample size), but it has little bearing of what might (ACTUALLY, not PROBABLY) happen in his next 20 ABs (small sample size). Yes, MATHEMATICALLY he SHOULD get on base 8 times…. but 4 times, or 12 times, is just as likely .”

[51] “Historical Actuality does not equal future probability in small samples.”

[61] “I agree, based on historical data, .925-.975 might be the best guess, but not necessarily an accurate guess the majority of the time.”

[68] “I’m just saying in any ONE instance, the odds of not conforming to the relationship may be close to 50%. Isn’t it widely concluded that small samples sizes have little predictive value?” [note: I responded to this above; you have confused terms here, I think, but that is not important at this juncture.]

[68] “But your assertion is basically correct more then incorrect, as stats do have some predictive value. But…. NOT necessarily on small sample sizes…”

[74] “That stat is one giant smear of data, all of which happened under different circumstances.”

[74] “And we know baseball is chaotic… that there doesn’t appear to be a pattern to just what effects a player, and what might get him a hit in THIS ONE AB.”

and for fun…

[80] “I don’t think you have a very high degree of accuracy doing this (bearing in mid that a 50% accuracy rate is totally meaningless…. random guessing yields the same rate.”
====

So, you have consistently argued, with varying degrees of intensity, that large data sets cannot be used to predict what will happen in a small number of events (one AB or a few ABs, for example). You posit thatbaseball is chaotic and lacking predictable patters in such cases. You posit that in any given even (e.g., one AB) the odds of getting a prediction wrong are at least as likely as it is getting it right (in other words, the prediction itself is no more or less certain than random chance). And so on.

And yet you take umbrage at me interpreting your basic argument as saying that large data sets are essentially meaningless in making a tactical decision (a “prediction”) about a single AB or even a single game?

How else am I supposed to interpret your argument?

I not sure what you are ‘understanding’ about what I said, but you are saying words I never said, and making assertion I never made. And I have said in every post that Statistic analysis IS important, and plays a role in decision making. I believe in your effort to reenforce your own view, that you are NOT getting what I am saying.

Yankster pointed out some flaws in the ‘simple’ way people here are analyzing data, and while presenting it differently (and better) he is basically making the same point I am trying to make about applying large sample size to a SPECIFIC event.

Again, to predict something (as ossposed to guessing) implies a certain degree of accuracy. When you say a bunt is (always?) a bad play because statistically it gives away 0.25 runs, you are assuming that a huge amount of NON specific data can meaningful be applied to ONE SPECIFIC event with a number of specific contingencies, and be somewhat accurate.

This is where I disagree with you. I don’t think you have a very high degree of accuracy doing this (bearing in mid that a 50% accuracy rate is totally meaningless…. random guessing yields the same rate.

I brought up a half dozen or more examples of specific contingencies that could effect ‘how smart/successful’ a bunt might be.

But here’s what I am REALLY objecting to.
Many people here call Girardi STUPID on a certain play, and use 1 or 2 stats, without even knowing how accurately they apply, to give weight to their opinion. Some people (who shall remain nameless) even glue all kinds of assumptions to their stats in hopes it further qualifies their opinion.

How about this c example. Here are some facts.
1) Girardi has a number of years of MLB experience as a player.
Most commenters (on all blogs) have none.
2) Girardi has 2 years of MLB experience as a manager.
Most commenters (on all blogs) have none.
3) Girardi has meeting with the Yankees FO, coaches, scouts and other personel involved with baseball decisions.
Most commenters don’t.
4) Girardi has personal relationships with the players, sees them everyday, talks to the frequently and witnesses the batting/pitching practice daily.
Most commenters don’t.
5) Girardi is paid a lot of money, is carefully watched by his bosses, and most take responsibility for his actioms
Most commenters aren’t and don’t.
6) Commenters had access to BR.com and other websites to view statisical data in various forms.
Girardi does also, but I’m guessing he ALSO has additional statisical data and analysis provided by the Yankees.

So my question is, statisically speaking:
What are the odds that any given commenter can make better managerial decisions then Girardi?

To read most blogs, the answer seems to be between 50% and 100%
(Yup… Girardi made the RIGHT move there, because I said so. Yup, Girardi made the WRONG move there, because I said so.)
I object to people not only thinking they know more then Joe (and every other manager) and then actually being dogmatic about it, if/when someone provide another point of view.

And frankly, EVERYONE knows Posada is a far better hitter then Molina. This is beyond obvious, and I believe Girardi is even aware of this (do ya think?). To criticise Joe for playing Molina over Posada, using only this ONE piece of information, is beyond shallow. If you don’t offer all the many other pieces of qualifying data that apply to the SPECIFIC situation (because the exact same decision may be much more right or wrong depending on the given situation), and analyze the data correctly, then I don’t believe your opinion (that’s a collective your) carries any weight.

That’s not what I am arguing against. Rather, I understood OYF’s argument to be: it doesn’t matter whom you pick, because how they did all season doesn’t tell us much of anything about how they will do this game.

Again, I think that is a problematic approach. Unless I have grossly misunderstood what was being argued, which is distinctly possible.

Interesting!

The point is that the season’s batting average or OPS is less predictive of two world series at bats than the distribution of at bats and definitely much less predictive than the median event in the single pitch conditional probability…

I agree. But all we have are larger data sets on which to make decisions (or predictions, as all managerial decisions are effectively predictions)…unless we defer to “gut instinct” or other nebulous concepts.

Where I disagree with OYF is that he (it seemed to me) glided too easily from “large data sets have less predictive value over a couple of WS at bats” to “and as such they are meaningless, so it doesn’t matter who starts.” If the larger data sets are so un-predictive as to be meaningless, then it really doesn’t matter who starts or what the lineup is for single game. And I don’t buy that.

All of the contingencies that you pose are measurable, or nearly all of them. And I have no problem with someone bringing additional factors into the equation (e.g.: yes, Hinkse slugs more than Hairston, but Hairston hits lefties much better, etc, etc.).

Again, that is not what you were arguing before. You were arguing that all of these differences are essentially small and meaningless, and that we shouldn’t really bother questioning decisions because larger data sets dont really tell us about what is going to happen next.

But even here, you appeal to larger data sets (what a player does against lefties, or in day games, or in certain circumstances). You are simply making the case that certain data sets are more relevant to particular tactical decisions than are other data sets. Still, your entire argument in this post fundamentally contradicts the reasoning you presented in previosu posts.

In fact, you DO believe that larger data sets have predictive value in “small samples,” so long as we isolate the most relevant data sets!

I agree!!

The problem is going from the abstract to the specific: What’s the event? In my view the reliable elementary event statistically is a single response to a single kind of pitch, not an at bat. But stats are generally discussed at the at bat event level and then that’s combined into numbers that are to me confusing in their value, like batting average. OPS is even more confusing given its (in my mind underweighting of on base). (batting average clearly has some strong relationship to the results of individual pitches – I’m just saying I don’t know exactly what that relationship conceals).

I don’t think that anyone is arguing that given the choice of who to bat between Posada and Molina you pick Molina. The point is that the season’s batting average or OPS is less predictive of two world series at bats than the distribution of at bats and definitely much less predictive than the median event in the single pitch conditional probability (if this probability of pitch and this probability of batter reaction, then this probability of the single pitch event outcome).