Also, I, too, believe that "luck" should probably be discarded for a different term like "outlier" or "randomness." Too often, people get stuck on the semantics and therefore shut down their brain and refuse to discuss when they see results attributed to "luck," and what's the point in using a term which will repeatedly get misconstrued?

I've advocated that for years. One big problem is that "luck" to many people seems to carry with it a moral component. If you say "chance," then people will accept that as morally neutral, but you say that it was "lucky," and people begin to hyperventilate.

I've advocated that for years. One big problem is that "luck" to many people seems to carry with it a moral component. If you say "chance," then people will accept that as morally neutral, but you say that it was "lucky," and people begin to hyperventilate.

I agree with that somewhat, but I still think the bigger issue (at least to me) is the use of luck as an excuse.

Its when the non-professional sabes use "luck" to mean something that was completely external to one's control and thereby resolve people of accountability or attempt to invalidate or diminish a result.

If they truly were dealing with external factors, it might be less of a concern, but they subsume into that "will of Allah" type usage things which cannot or have not been modelled, and still are within the total or partial control of the actors.

It distorts reality b/c then you have people thinking the only thing that is real or matters is there interpretation of somebody else's model.

Its not just about precision in language or diction in connotation. If you want to be precise, then "luck" (at least from my experience) is not a statistician or a scientific term. And you can't just replace it with randomness.

When luck is invoked, it isn't invoked uniformly. There is nothing 'random' in a closed event space. It has all happened. There are only things that could have been controlled to some level by the various actors. Looking forward, there are agents that can be treated as being random, and are random from the point of view of the actor at a specified time.

Luck is used to subsume both of these things, and that is problematic. Because what is happening is we are taking things that must (or have been) taken as random by the modeller from a distant time t, and using that to mean that a certain actor did not exhibit any accountability in the outcome. And many times that is definately not true for what it is you are seeking to do.

To wit, the rebuttal to greybault's point. "rolling dice", "tossing coins", etc. are used as randomizers, not because there aren't causal forces at play, but because the human actor that rolls the dice or tosses the coin typically does not have the physical dexterity to exhibit significant control on the outcome, and everyone who is watching at the time of selection and the level of instrumentation of just our eyes, will have to treat the roll as a random event. We actually are looking for pitchers that have the physical and mental acumen to exhibit control on the outcome of an at bat. And while that locii of control is not absolute, a pitcher does have some responsibility in the outcome. And we do expect managers to understand the current physical capability of the players to be able to execute. While their locii of control is even less, they are also having some responsibility. And GM's have very little locii of control, but they did select the pitchers and managers, so they are not totally blameless. That is the critical difference.

Backlasher (in post 90) and FJ (in 93) hit on what I was trying to say.

If I had to boil it down to one sentence, it's this:

"Just because an outcome could have happened by chance doesn't mean that it did."

-- MWE

When luck is invoked, it isn't invoked uniformly. There is nothing 'random' in a closed event space. It has all happened. There are only things that could have been controlled to some level by the various actors. Looking forward, there are agents that can be treated as being random, and are random from the point of view of the actor at a specified time.

One issue, BL, is that when discussing things "that happened in a closed event space", *I think* that people mean what you say here:
"Looking forward, there are agents that can be treated as being random, and are random from the point of view of the actor at a specified time."
and
"because the human actor that rolls the dice or tosses the coin typically does not have the physical dexterity to exhibit significant control on the outcome, and everyone who is watching at the time of selection and the level of instrumentation of just our eyes, will have to treat the roll as a random event."

So when I refer to the A's losing as "bad luck", I am sying "At the time it was occuring, and looking forward *from the moment before any particular decision was made*, the decisions made were sound, hte probability of a favorable outcome was high, but luck or randomness, or the dice roll, came up a natural 20." After making decisions based on the info available at teh time, and because of factors people can't control, a low probability event *still* occurred - that is "bad luck" the way I use it.

So when I refer to the A's losing as "bad luck", I am sying "At the time it was occuring, and looking forward *from the moment before any particular decision was made*, the decisions made were sound, hte probability of a favorable outcome was high, but luck or randomness, or the dice roll, came up a natural 20." After making decisions based on the info available at teh time, and because of factors people can't control, a low probability event *still* occurred - that is "bad luck" the way I use it.

You could also take the same situation and call it luck for one persone or persons, and not luck for others. For example:



ATHLETICS 6TH: Byrnes singled to center; Byrnes stole second;
McMillon grounded out (second to first) [Byrnes to third];
Durazo walked; Tejada reached on a fielder's choice [Byrnes out
at home (catcher), Durazo to third (error by Lowe), Tejada to
second]; Byrnes did not touch the plate as Varitek blocked it
with his foot; the wild throw went to the other side of Byrnes
from the catcher; Byrnes appeared to hurt his ankle in the
collision; Varitek retrieved the ball and tagged Byrnes, who was
called out; Chavez was walked intentionally; Hernandez
reached on an error by Garciaparra [Durazo scored (unearned),
Tejada out at home (left to catcher), Chavez to second];
obstruction called on Mueller, but ruling was that it occurred
before Tejada reached 3B; therefore, by rule 7.06(b), he is
awarded 'next base', which was third and further advance was at
his own peril; Tejada stopped running, tagged out by 30 feet;
there was a long discussion by the umpires and then a long
argument from A's manager Macha; 1 R, 1 H, 2 E, 2 LOB.
Athletics 1, Red Sox 1.

Now it wasn't luck for Tejada and Byrnes that they forgot the basic fundamentals of their profession at that moment. However, Billy Beane is continually castigated by many here for his sh1t which doesn't work in the playoffs. There's nothing about Billy Beane's sh1t that makes Miguel Tejada, and outstanding ballplayer and possible HOF'er forget how to run the bases. Beane doesn't have a playoff win on his resume, and one big reason is that he was unlucky to have two players suffer severe brain cramps during the most important plays of their careers.

So Dial - whenever an underdog wins, no matter how well the play relative to the competition it was luck?

So Dial - whenever an underdog wins, no matter how well the play relative to the competition it was luck?

Of course not. What I believe Dial is saying is that anytime two teams play, the underdog will win a substantial percentage of the time. If two teams play five games, and the first team wins three of them, it does not mean that the first team is necessarily better than the second team, or better designed. If two teams are relatively evenly matched each will win a five games series at least 40% of the time, by pure random chance.

The mere fact that a team lost a five game series or two says nothing about the team or the person who created it. The sample size is much, much too small.

Bronson Arroyo hit homeruns in his first two games this year. Barry Bonds and Manny Ramirez hit zero in their first week. Should we interpret this to mean that Bronson Arroyo is now one of the greatest homerun hitters who ever lived, and that Bonds and Ramirez are no longer capable of hitting homeruns? Of course not - the sample size is much too small.

Similarly, should we declare that the A's teams of recent years were inherently flawed because they lost a couple of five games series? Of course not - the sample size is much too small.

It doesn't matter if you call it luck, or chance, or randomness, or whatever. When you are dealing with a small number of games, or at bats, anything can happen, and it doesn't mean anything.

Having said that, I agree completely with the following:

"Just because an outcome could have happened by chance doesn't mean that it did."

So when I refer to the A's losing as "bad luck", I am sying "At the time it was occuring, and looking forward *from the moment before any particular decision was made*, the decisions made were sound, hte probability of a favorable outcome was high, but luck or randomness, or the dice roll, came up a natural 20." After making decisions based on the info available at teh time, and because of factors people can't control, a low probability event *still* occurred - that is "bad luck" the way I use it.


And many times, I disagree with the fact that there is not accountability in the series.

I think you saberists have some law that says something along the line of "anything can happen in 60 AB." While I do not agree with that completely, lets say that I grant that for a moment. At least to the extent that if you define your atomic event as 1 PA, multiple series of sixty outcomes will have wide variance, and that variance will decrease when you have increased the number of events. (As an aside, this also gets into the problem with the misunderstanding and/or misuse of the term "sample").

But the problem is that there is no analytical driver that forces you to choose a PA as your atomic event. The reason that you are choosing the PA is because is that is generally the smallest unit of information that your instrumentation is giving you.

When you are analyzing a playoff series, there are a whole range of events that are at play. More likely than not, a set of those events would have low probability outcomes if you projected their likelihood from a prior time. Some of those low probability events result in favorable outcomes, some result in non-favorable outcomes.

What many people are doing is taking the set of low probability outcomes that are unfavorable and saying the result of the whole series of events were bad luck.

Beane doesn't have a playoff win on his resume, and one big reason is that he was unlucky to have two players suffer severe brain cramps during the most important plays of their careers.


And the bigger reasons are when he had a team that had a reasonable chance of winning a playoff series, he instituted a low probability plan of station to station baseball and he didn't supplement his young talent properly with persons that were good at handling the pressure. He certainly had control of those two events.

The man lost four playoff series in a row. He fared even worse in series-clinching games.

"brain cramps" abound, including such things as Tim Hudson's bar fight. You can't just pick a couple of unfavorable low probability events and say, "Oh that's the difference". Of course it is a difference, and it wouldn't have been a difference if other low probability events didn't inure to your benefit.

So for years, rather than seeking to explore what the problem was, there was just a throwing up of hands and declarations of "Oh, just bad luck, Beane can't control Byrnes blunder."

And plenty of people did exactly like Cadbury did, they began forming hypothesis on what could be the shortcomings. And they were told about "pebbles" and not understanding "luck". Heck the argument was made that some draft pick he made the previous year was evidence that the playoff losses weren't his fault. And the best one were the people proving that because it was statistically possible for the event to occur because of luck, it meant the event did occur because of luck.

And recently it seems that a few decent studies have been done that show team compositions that do have more favorable outcomes in the playoffs. And you can find delta's between such teams and the A's teams that were playoff caliber.

It becomes a problem because the fanboys who don't know how to do anything will go ballistic because you are criticizing something they care about. And some of the analysts will get just as defensive because it might show gaps in their models, and that bothers them even if you grant their model is pretty optimal based on the existing instrumentation.

If two teams play five games, and the first team wins three of them, it does not mean that the first team is necessarily better than the second team, or better designed. If two teams are relatively evenly matched each will win a five games series at least 40% of the time, by pure random chance.

The mere fact that a team lost a five game series or two says nothing about the team or the person who created it. The sample size is much, much too small.


And the problem is that I highly doubt anyone doesn't know that an underdog will not win some times against a team that is more heavily favored.

But so many of the saberists feel like they are holding this information as a cosmic truth that other people don't understand, that when somebody MAKES AN ENTIRELY DIFFERENT POINT, you feel like you need to explain this, then they will "get it."

Because first of all, it isn't a "sample" of anything; at least not anything real. The use of "sample" by mgl and tango is correct for an artificial construct of "true talent level" and such an artificial construct is useful for them in PROJECTING FUTURE EVENTS.

But these are outcomes. No one really cares what may happen if the teams play each other one million more times; they care about the event that occurred.

And the idea that all games, all innings, all series, and all pitches are the same is completely untrue and contrary to very common sense. Such things can be treated the same if you are projecting far enough into the future that your model has entropy.

There should not be much argument that there are plans that can increase the chance of certain outcomes will reducing the odds of other outcomes. And there should not be much question that there are contexts when these different plans produce different utility. For instance, a difference in utility for a bunt in the first inning with nobody on and a slow runner against a good defense and a scrub pitcher, and a bunt in a tie game, bottom of the 9th with a runner on 1st, and no one out, against a porous defense and a dominant closer.

When you are dealing with a small number of games, or at bats, anything can happen, and it doesn't mean anything.


And this is where you are DEAD WRONG, and that is the problem. What it means is what you are looking for. It always MEANS something, in the same way as its ALWAYS CAUSAL. But you are conflating an outcome whose causes do not require you to do anything different, with "it means nothing" and then ascribing that to every series of events for which YOU do not have enough outcomes to maximize the confidence interval.

And that is a problem, and that is not semantics. That is precisely what I'm talking about as the problem. The "it could happen" so its luck.

If two teams are relatively evenly matched each will win a five games series at least 40% of the time, by pure random chance.

If your flipping coins sure.

Baseball isn't a coin flip.

Last night the Red Sox weren't unlucky to lose to Cleveland - Josh Beckett pitched like crap. If the Sox were a 700 team, and the Indians were 600 team and last night was the 7th game of the ALCS, the Indians wouldn't have won because of luck - they would have won because they executed and the RedSox didn't.

Execution is not luck. Lack of execution is not luck.

Sure does luck play a roll - yeah, when things happen outside of the teams control - an umps call or freak injury.

When you are dealing with a small number of games, or at bats, anything can happen, and it doesn't mean anything.

I agree with everything up through the last five words. When a pitcher throws a no-hitter, or strikes out 15 hitters in a game, it means something; the overwhelming majority of such games are thrown by good pitchers. When a batter puts up a 15-game hitting streak, it means something; the overwhelming majority of hitting streaks of at least that length are posted by good hitters. When a team wins its first five games of the season, it means something, as Bill James once noted - the team is, very likely, a quality team. Jmaes called these events of "signature significance"; they have little or no direct significance, in the statistical sense, but they are the "signature" of good players or teams.

-- MWE

I agree with DMN in 101, Dial in 104 and DP in 107.

"Just because an outcome could have happened by chance doesn't mean that it did."

I agree with this. The real issue is, who has the burden of proof from this point on. In my view, it rests with those who deny the role of chance.

Josh Beckett pitched like crap. If the Sox were a 700 team, and the Indians were 600 team and last night was the 7th game of the ALCS, the Indians wouldn't have won because of luck - they would have won because they executed and the RedSox didn't.

Execution is not luck. Lack of execution is not luck.

In hte micro sense. Not in teh macro sense.

*WHY* did Josh Beckett "pitch like crap"? Is he a crappy pitcher? No, what happened was he performed, within the variance of his true talent level, on the low side.

As a fan, in hte macro sense, your team lost because the 5% chance that Beckett "pitched like crap" happened. How unlucky for you. There's only a 5% chance that Beckett will pitch like crap, and he did.

If you have to roll a twenty-sided die, and only on a 1 will you lose, in one roll, if you get that 1, that's what I'd call unlucky. There's a 95% chance you will win, but that 5% got you.

Except there is no dice... what happened happened. Human beings performance is what caused the game to go the way it did. Not some big guy playing D&D in the sky.

Except there is no dice... what happened happened. Human beings performance is what caused the game to go the way it did. Not some big guy playing D&D in the sky.

That has nothing to do with it. And you didn't explain why Beckett pitched like crap. And what teh odds of that are. Proffer an explanation.

Why did he pitch bad? I don't know... I know he couldn't throw a strike, unless he grooved it - then they grooved it back over the fence.

Maybe he has a blister.
Maybe he got in a fight with his woman and couldn't concentrate.
Maybe the weather was adversily affecting his grip.
Maybe he had a sore throat.
Maybe 8,000 different things.

The bottom line is he didn't pitch well. Thats not luck - thats fact. He din't pitch well. Could you have expected it? It doesn't matter. What would he do if they replayed the game 9,000 times. It doesn't matter, they don't play the game 9000 times. It isn't a simulation - its real life where all that matters is what happens.

Why did he pitch bad? I don't know.

This isn't all that different from what people here have meant by luck, even if some lexicographer thinks it's incorrect usage of the term.

It isn't a simulation - its real life where all that matters is what happens.

I don't know many people who feel so constrained.

The real issue is, who has the burden of proof from this point on. In my view, it rests with those who deny the role of chance.


First, there is no BURDEN on anyone. There is nothing to prove. There is no null hypothesis, and analysis using that model hasn't exactly been done to much in about fifty years. Null hypothesis testing is fallacious.

YOu can choose to investigate cause, or you can bury your head in the sand and say "it doesn't mean anything." Or as others do, you can sit back and assume that all inferences (or opinions) or equal, and that someone needs to PROVE something to YOU to meet some BURDEN that does not exist.

Analysis is not about convincing people of anything. You have an inference with a measurable level of support, and you have other inferences with measurable levels of support. It doesn't matter what an opinion is, except to the extent that such opinion might lead to better methods of analysis.

For instance, I have an opinion that streaks exist and are determinable and are influential in predicting outcome. I have a method to identify said streaks, and the ability to create the instrumentation to implement said method. There is a plausible inference that streaks exist, and it cannot be currently falsified.

mgl has an opinion that streaks do not exist, if they do they are not determinable, and if they are determinable, they do not influence outcome in a way that justifies the cost of the information or adds utility to the operation of baseball. He has direct evidence to support that opinion. That leads to an inference that streaks do not exist. That cannot be currently falsified.

Despite my opinion and conviction, mgl has a more supportable inference, and it is of such a higher level of validity, it deserves to be the operable inference. The continued value of the discussion is only to the extent that I bring to the table a mechanism for greater exploration.

If you want to stress BURDEN, its a game without meaning. And one that I'd play if people realize that once falsified, there opinion is not operable, and its not about convincing anyone of anything (except maybe the people that actually do some of the heavy research). We can run Bayes factor tests on everything if you like, but that really changes conversation.

Proffer an explanation.

From the POV of the analyst looking at the beginning of the season, I'm quite sure Becketts performance was within the "variance of his true talent level" Here is the problem, just about anything Beckett would have done would be within "the variance of his true talent level" as seen by an analyst projecting at the beginning of the season.

If you move up time and gather more information, that variance shrinks a little more to hopefully create something more of a decision paradigm than a dice roll.

But if you want to analyze that, the first step is to identify the cause of Beckett pitching at that particular "variance of his true talent level." And since the event space is closed, you will find causation. It may be a blister, it may be a hangover, it may be trouble with the family, I don't know.

Once you find those out, the next step is determining their level of influence, then determining how to mitigate (level and cost of control), then determining their potential for repetion, then determining the other factors likely to influence repitition.

Doing these things will greatly impact micro projections and can also influence macro projections.

When you throw your hands up and say, "its just luck"; you short circuit the conversation at causal analysis.

Now, its 100% true that you should not study everything. The cost of study and the range of expected utility are approximable for many things before you begin. I'm not sure there was ever enough of either to study using a null hypothesis framework, the null hypothesis of " pitchers do not influence hits on balls in play". But I've only seen about two people that even try to give this approximation when discussing analytics.

But the mere fact that a long event, composed of many micro events and many determinable and controllable agents, is within the range of possibility determined at a distant time WHEN said controllable agents where randomized, does not mean you should say, "its expected and it means nothing."

This isn't all that different from what people here have meant by luck, even if some lexicographer thinks it's incorrect usage of the term.


Its a hell of a lot different than "it means nothing" Its a hell of a lot different than "its nobodies fault".

The absence of information, the lack of instrument precision is not the same thing as lack of accountability, and inability to discern useful information.

The bottom line is he didn't pitch well. Thats not luck - thats fact. He din't pitch well. Could you have expected it? It doesn't matter. What would he do if they replayed the game 9,000 times. It doesn't matter, they don't play the game 9000 times. It isn't a simulation - its real life where all that matters is what happens.

It's luck from the point of view of everyone except Beckett.
Luck and fact are not mutually exclusive

The fact that a habitual drunk driver gets into an accident is not "luck"
the fact that he has an accident on any particular day and runs down a specific individual is random chance (and pretty crappy luck for the individual run down).

When one playoff series is over and done with- you are right it's over and done with. But what about the next one? and the one after that? If you are running a team you want to know what is the right strategy to use the next time. Then all that matters is what you should be able to expect from Beckett (10% of the time he does this- 5% he does that etc.)- what matters most to teh decision makes (or what should matter) is precisely what you say doesn't matter.

Its a hell of a lot different than "it means nothing" Its a hell of a lot different than "its nobodies fault".

With those remarks taken out of context, I suppose that's true. If being taken seriously by a professional statistician is someone's great goal in life, then this could be very important.

The real issue is, who has the burden of proof from this point on. In my view, it rests with those who deny the role of chance.

It's not about proof - it's about assessing evidence and drawing inferences from evidence. The more evidence you have, the more likely it is that you can draw inferences from the evidence that closely reflect the real underlying causes of the real-world events that you see. But there's always a level of uncertainty.

Too many people apply the term "luck" or "random variation" or "chance" to PAST events - and accept the conclusion that because something could have happened by "chance", there's no reason to look at possible scenarios that could explain why it DID happen that way (and what might have been done to prevent what happened before it actually did happen, or at least to prevent a recurrence next time around). I don't think there's anything wrong with admitting that "I can't tell why this happened"; where I think we go astray is when we move from "I can't tell" to "it just did; dem's da breaks".

-- MWE

There is no null hypothesis, and analysis using that model hasn't exactly been done to much in about fifty years. Null hypothesis testing is fallacious.

That's simply completely untrue.

It's not about proof - it's about assessing evidence and drawing inferences from evidence.

Not proof, no, burden of proof: who needs to come forward with evidence and how much evidence do we need/want. I agree we should assess evidence and draw inferences. My point was "who is this we?".

Too many people apply the term "luck" or "random variation" or "chance" to PAST events - and accept the conclusion that because something could have happened by "chance", there's no reason to look at possible scenarios that could explain why it DID happen that way (and what might have been done to prevent what happened before it actually did happen, or at least to prevent a recurrence next time around). I don't think there's anything wrong with admitting that "I can't tell why this happened"; where I think we go astray is when we move from "I can't tell" to "it just did; dem's da breaks".

In principle, we're always better off trying to identify a cause. Sometimes, though, we can't. I do not know, and no one will ever know, why sometimes a blooper drops in or why sometimes a screamer goes right at someone. We certainly could investigate it, but we're better off IMO if we just ascribe it to "chance" and move on to a problem we might be able to solve.

Just to be clear here, too, someday we may be able to solve problems like this. We can't now. I'm making a pragmatic argument not a metaphysical one. Too much of the objection to use of the term "luck" seems metaphysical to me. I don't care for the term myself, but using it is not a sign of moral degeneration.

I agree with Chris in 113.

I agree with everything up through the last five words. When a pitcher throws a no-hitter, or strikes out 15 hitters in a game, it means something; the overwhelming majority of such games are thrown by good pitchers. When a batter puts up a 15-game hitting streak, it means something; the overwhelming majority of hitting streaks of at least that length are posted by good hitters. When a team wins its first five games of the season, it means something, as Bill James once noted - the team is, very likely, a quality team. Jmaes called these events of "signature significance"; they have little or no direct significance, in the statistical sense, but they are the "signature" of good players or teams.

I generally agree with everything Mike has said in this thread. I also agree that striking out 15 hitters in a game, or having a 30 game hitting streak, or hitting four homeruns in a game are all meaningful. Winning several games in a row is meaningful. Doing something that is relatively rare, or that mostly good (or bad) players do, is generally meaningful.

What is generally not meaningful is doing something that is relatively common, and achieved by players and teams of all levels of ability. Winning three games out five from one team, for example, is extremely common and is accomplished by teams of all levels of ability. Getting a hit in seven out of eight games is very common and is accomplished by lots of players of varying abilities. It is these cases that we should be very careful not to overanalyze. Its true that something else other than chance may be involved, but its so common that almost silly to make inferences based on this kind of event - this is why I'm so opposed to the "Why can't the A's win in the playoffs?" threads.

What Jeff Franceour did over his first several games is relatively rare; the fact that you can think of a few other guys of limited ability who put of similar numbers does not mean that it was luck - something else might have been going on. In a case like Franceour, it is clearly worth having the conversation about what is going on, and I appreciate the original article that started this thread.

It's luck from the point of view of everyone except Beckett.


No, its not. More, infra.

Then all that matters is what you should be able to expect from Beckett (10% of the time he does this- 5% he does that etc.)- what matters most to teh decision makes (or what should matter) is precisely what you say doesn't matter.


Yep, and also what you should do when Beckett is pitching in certain circumstances, what things should you do maximize your goal of winning the game. And under what circumstances do you pitch Beckett to maximize your long term goal of winning a world championship (or keep a dynasty, or lesser goal of improving by x things).

And what you don't do is pretend that a model once invented with x information at a distant time t is the model.

The further away in time you are, and the more your abstract and randomize, the more it looks like a crapshoot.

If you know that Beckett had a blister, and when he has a blister, the distribution of pitches will do y, you might have a different defense, you might decide to skip a turn (depending on heal rate). If you find those blisters recur often, you might try to remediate or find cause, and you might determine you alleviate them if you don't throw y pitch. Not throwing y pitch may change his role with the team, etc.

My not knowing these things doesn't change the fact that knowing these things and investigating them can lead to different plans that have more utility for goal achievement.

If the event space is shrunk down to pitch, you increase the number of events whereby you can make accurate inferences with low variance. If the event space is shrunk down to mechanics at various points in the delivery, then you can make even more inferences with low variance.

And the human actors involved do have those instruments, they are called the eyes; they do have a lot of the causal equations, that is called the base of experience or baseball knowledge; and they do have the processors, they are called the brain.

But that brain operates with time constrained anytime algorithms, baysian convergence equations, and/or pattern recognition methods. Baseball analysts don't have those constraints. They can model without worrying about the impact of time. They can use the most precise algorithms. And if they do, they can reduce the influence of bias, misinformation, and output plan that reduces even better utility. Of course, they can never get to that if they get distracted with "it could happen", "it means nothing" and "burden".

In fact, on the latter, if we want to assign "burdens", considering that every event means something, let's put the burden on those that say in a roundabout way "its not significant enough to discuss" to prove "its not significant enough to discuss" That is an obtainable output. I can falsify it means nothing in about four sentences.

</b>Why did he pitch bad? I don't know.</b>

This isn't all that different from what people here have meant by luck, even if some lexicographer thinks it's incorrect usage of the term.

Complete apples and oranges. Someone might be in a position to know why he pitched poorly - for example, Beckett's pitching coach, the medical staff, an expert analyst who looked at Beckett's motion, a psychologist, a dietician, Beckett's "woman", Beckett himself, etc. etc. Rauseo doesn't happen to know because he's a fan like you and me. We lack the access and the expertise. But it is, in some sense, knowable. In the same way that I don't know why Francouer is doing worse now than he was, but a hypothetical expert who looked at all his at-bats, diet, training regimen, etc etc would be able to tell you.

Now, in the real world it may not be possible to access the information quickly enough to be in a position to do anything about it - eg Beckett had a fight with his "woman". And even if we have the information, we may not be able to do anything abou it - eg we knew Beckett wasn't going to be great, but he was still the best option to start. And it may be very expensive and time-consuming to get the information, and so it may be judged not worth the effort to obtain. For example, I had to choose between Joe Mauer and Victor Martinez for my AL-only fantasy baseball team. If I had watched every plate appearance each made last season before making my choice, I might have been in a better position to judge. But obviously I don't have time for that.

None of the above is luck. It is all knowable. If I win the fantasy baseball because I picked Victor Martinez over Mauer, that is in some sense luck, because I was essentially picking randomly. My choice between the two added no value. But, as BL so rightly pointed out, we want GMs, managers, hitting coaches, etc to be able to add value with their choices. And the more value they can add for given expenditure, the better they are at their job. To take it back to fantasy baseball, I could also have picked Raul Chavez. If I win because I picked Martinez over Chavez, that's not luck, because I added value by my choice. And a better fantasy player than me would have added value to the choice between Martinez and Mauer. And so on.

I am glad this discussion took this turn, one of my fascinations from way back is game theory.

One of my joys in watching baseball is the intricate strategic puzzle of every batter/pitcher matchup. The pitcher with each pitch tries to throw something that the hitter is unable to hit hard, and the hitter on the other hand is trying to be prepared for where he thinks the pitcher will attack him. The key, and where we get to game theory, is that there is no absolute right thing for either to do. The worst strategy though is easy to explain; being predicatble is a loser every time. The pitcher and catcher try and have a strategy that keeps hitters "off-balance", using those pitches which the pitcher best executes in a way that the hitter is not able to prepare for a specific pitch in any situation.

Five or ten years from now we will wonder how people could not perceive the importance of the proper pitch selection strategies, and sites like BP will grade catchers on their pitch calling from a game theory perspective. Likewise it will be abundantly clear which hitters have never been able to hit a breaking pitch, and how foolish pitchers are that throw them nothing but fastballs.

But for now we don't have the tools to break this down...but thats ok. We can still enjoy the seat-of-their-pants brilliance of pitchers, catchers, and hitters who figure it all out.

Complete apples and oranges. Someone might be in a position to know why he pitched poorly - for example, Beckett's pitching coach, the medical staff, an expert analyst who looked at Beckett's motion, a psychologist, a dietician, Beckett's "woman", Beckett himself, etc. etc. Rauseo doesn't happen to know because he's a fan like you and me. We lack the access and the expertise. But it is, in some sense, knowable. In the same way that I don't know why Francouer is doing worse now than he was, but a hypothetical expert who looked at all his at-bats, diet, training regimen, etc etc would be able to tell you

I disagree with this. In many cases, no one is in a position to know why something happened, no matter how much information they have. Beckett's pitching coach, or some other expert may have some ideas or educated guesses about what happened, but in most cases its impossible to say for sure. No hypothetical expert can say for sure why Francouer is doing worse than he was - the information simply doesn't exist. It is not, in any sense, knowable.

In fact, people close to the situation often get in trouble because they think they know the cause of a situation, and they are just wrong. Hitting coaches for example, sometimes point to a change in the swing, and say "if he just stops wriggling his elbow" he'll get back to hitting the way he did before. In many cases, they're just wrong.

That's simply completely untrue.


I'm sorry it is true. Bayes factors replaced null hypothesis testing before I was even born, and Null hypothesis testing can lead to fallacious output. All you can know is the power of the inference, and from that power decide how you wish to weigh it.

I'm making a pragmatic argument not a metaphysical one. Too much of the objection to use of the term "luck" seems metaphysical to me.

Ok, pragmatically, any outcome that was modelled with a random variable will have a percentage chance of the entire set for the range of outcomes.

Or put more distinctly, with just about anything we are talking about, there is some percentage chance the outcome will occur.

So you want the default position to be that all outcomes are to be considered 'luck' unless someone meets a BURDEN otherwise.

Player A strikes out 50 times in a row. Could happen, its going to be luck and we plug him back in until somebody shows X to what degree.

Because you can't falsify the position that "using the information at the beginning of the season and randomizing the opponents, and injuries there is .000001% chance that he would strike out that many times."

Or on its face, common sense would say, "something is wrong with this guy", but that is not how you want to proceed. You want somebody to bring forth evidence of what is wrong with this guy. That is not pragmatic.

Nobody is talking about pebbles and stones in the infield. We are usually talking about players and organizations in defined environments. We are usually talking about information obtainable from medical authorities.

The worst strategy though is easy to explain; being predicatble is a loser every time.

I agree, (unless you are Mariano Rivera.)

I disagree with this.

Disagree all you wish, its not a matter of opinion.

In many cases, no one is in a position to know why something happened, no matter how much information they have.

BZZZT. It may be very possible that it is not feasible to get the information. But if its a closed space, and all events are causal, then the information does exist, and if you can get the information, then you will know. Its a matter of the cost of the information and whether the instrumentation presently exists.

That's why all this "lexicographer" stuff is important.

What you say is demonstratably false, but even if it were accepted as being pragmatic, you don't have ASMI building pitching motion machines, you don't have investigation into elements where we have affordable instrumentation to get information to influence the decision.

In fact, people close to the situation often get in trouble because they think they know the cause of a situation, and they are just wrong. Hitting coaches for example, sometimes point to a change in the swing, and say "if he just stops wriggling his elbow" he'll get back to hitting the way he did before. In many cases, they're just wrong.


In many cases they are wrong, because you don't try to get complete and perfect information. You try to get information whose cost creates a benefit in the return on the decision.

If you could get perfect information with no cost and no time constraint, you would always get it before you make a decision. But there is time and money.

And at the end of the day, some hitting coaches are better than others. They are better observers, or better processors, or better communicators. You want the one that balances cost and the best chance of utilitarian outcome.

I disagree with this. In many cases, no one is in a position to know why something happened, no matter how much information they have. Beckett's pitching coach, or some other expert may have some ideas or educated guesses about what happened, but in most cases its impossible to say for sure. No hypothetical expert can say for sure why Francouer is doing worse than he was - the information simply doesn't exist. It is not, in any sense, knowable.

In fact, people close to the situation often get in trouble because they think they know the cause of a situation, and they are just wrong. Hitting coaches for example, sometimes point to a change in the swing, and say "if he just stops wriggling his elbow" he'll get back to hitting the way he did before. In many cases, they're just wrong.

You miss the point. But your example of the hitting coach is a good one, even if it illustrates the exact opposite of what you want it to. A good hitting coach will be able to tell what's wrong most of the time, in the same way that a good doctor will be able to diagnose most illnesses. Even better, the hitting coach may be able to "fix" the problem, just like the doctor may be able to prescribe the cause of treatment. No doctor and no hitting coach is perfect, they all make mistakes, of course. There is no "fix" for some problems, like there is no cure for some diseases. Sometimes there is a "fix", but the player is unwilling/unable to implement it, like a patient refusing to take his medicine. And some errors in your hitting, like some diseases, may not have been discovered yet.

That's why I say a hypothetical expert. I don't dispute that it's theoretically possible that no-one currently knows the cause of the problem, but it's knowable in that it's possible to find out. No-one knows how many grains of sand there are in the Sahara, but it's knowable. In fact, though, I don't think this is like sand in the Sahara. The Braves are a well-run organisation, I reckon they've probably got it figured out. Of course, that doesn't mean it's in any way fixable. If the main problem is as suggested above, that Francouer lacks the eyesight to identify and hit breaking pitches, then it seems there's little that can be done.

BL,

Strangely i am not surprised we agree on this subject...

but even with Mariano, his pitches may be predictable...location isn't.

As a pitcher your choices are limited to what you can do, Kirk Rueter's best options were not as good as Mariano's worst, however Kirk was a master of understanding the theory, and was pretty good at the execution as well. But at a certain point you need to be able to have a good option that gets people out....

Dizzypaco's points made a lot more sense in the era before universal taping of every aspect of a game....now the team DOES know exactly what a pitcher was doing right and wrong, and they have done research into what to look for, the information is there even if it is not available to the outsider.

My not knowing these things doesn't change the fact that knowing these things and investigating them can lead to different plans that have more utility for goal achievement.

Go back. You said (and I agree):
"Looking forward, there are agents that can be treated as being random, and are random from the point of view of the actor at a specified time."
and
"because the human actor that rolls the dice or tosses the coin typically does not have the physical dexterity to exhibit significant control on the outcome, and everyone who is watching at the time of selection and the level of instrumentation of just our eyes, will have to treat the roll as a random event."

Beckett pitched like crap. Why? At our level (you me and everyone but Mr. HiStds) the best we can do is "random variation". Beckett isn't going to have "it" everyday.

There is some *why*, but even with your invocation of "medical personnel", that doesn't mean there was a blister, a fight with a girlfriend. It *can* be as simple as he just didn't throw well.

You hear it all teh time - "I had good stuff warming up, but didn't have it on the mound." And vice versa. You hear that from the players that had hte performance. If you ask Beckett he'll say he just stank (probably). He's most likely to call any "Why" that MHS put forward as an "excuse, and he doesn't make excuses".

So back to your quotes:
"there are agents that can be treated as being random, and are random from the point of view of the actor at a specified time"

Where *we* are the actors, barring some given explanation (and there isn't one as yet), isn't the best assumption that it is a random fluctuation?

If not, what is the "best" assumption?

Disagree all you wish, its not a matter of opinion.

I agree with you, its factual that some things are not knowable. What you are saying is demonstrably false.

It may be very possible that it is not feasible to get the information. But if its a closed space, and all events are causal, then the information does exist, and if you can get the information, then you will know. Its a matter of the cost of the information and whether the instrumentation presently exists.

This is just incorrect, at least as I understand your point. It is often virtually impossible to know why something caused something else, even with all the information in the world. We might know that Beckett stubbed his toe, or had a fight with his girlfriend, or whatever, but we will never know if each of these factors caused a specific performance. Knowing that two events occured (Beckett had an argument, and he pitched badly), does not mean one caused the other. You are confusing correlation with causality.

In many cases they are wrong, because you don't try to get complete and perfect information. You try to get information whose cost creates a benefit in the return on the decision.

In many cases they are wrong for the same reason that you are wrong - they confuse correlation with causality. Sometimes, the reason that a player performs worse one week versus another is random chance, or is simply unknowable. Yet by having some information, they make an incorrect assumption about causality. The problem isn't that they didn't have enough information; its that there is no way of knowing for sure at all.

Dizzypaco's points made a lot more sense in the era before universal taping of every aspect of a game....now the team DOES know exactly what a pitcher was doing right and wrong, and they have done research into what to look for, the information is there even if it is not available to the outsider.

Being able to watch a videotape of a performance does not give you as much information as you think it does. Sometimes it will show you a change in motion, or something similar, but many times videotape will not show you why a pitcher pitched better in one game than another.

But if its a closed space, and all events are causal, then the information does exist, and if you can get the information, then you will know.

Actually most physicists would say that some of the information doesn't actually exist (Einstein was in the minority here) and I believe that all physicists would agree that there is information that is unknowable.

You might think that talking about the position and momentum of electrons is missing but point but all objects have some quantum nature and, if you buy into Chaos theory (and I do), you might believe that the unknowable position of a small particle could certainly determine the outcome of a playoff series (not ot mention much grander events). My point is that it's quite possible that there is true randomness (for all actors) at work here.

That said, yes, we need to use a Bayesian model where the more series the A's lose the less credence we give to the theory that they've merely been unlucky in the playoffs and the more credence we give to the "flaw that shows itself in the playoffs" camp. It would take a couple more convincing playoff loses for me to pay too much attention to that camp.

but even with Mariano, his pitches may be predictable...location isn't.


I agree Tom, I was just making a joke. Sorry.

If being taken seriously by a professional statistician is someone's great goal in life, then this could be very important.


I don't know a professional statistician, nor do I imagine such a person would have the necessary training to deal with what we are discussing.

If what you mean is the normal person. That's fine. Your average person isn't going to think a 40 game hitting streak is "luck" They aren't going to think "Pitchers have no effect on balls in play and that it is luck."

If you are talking about professional analysts, they are going to have tools at their disposal not to make big mistakes.

But what we are talking about is this environment, which is kind of weird. Out here, people are going to have some modicrum exposure to classical statistics. They are going to understand how to perform operations on large sets of data that has normal distributions. They are going to have knowledge of trivial things like the Law of Large Numbers, Gamblers Fallacies, or little buzz phrases like "correlation isn't causation" then they are going to let those things creep much further into decision then they ever should.

They also are going to think the first set of people are dumb because they just don't get it, and they are going to intefere with the second set on having discussions on causal hypothesis.

That is where the problem lies. Its not semantics; I don't care what word people choose; its the application that is usually the problem.

Bayes factors replaced null hypothesis testing before I was even born, and Null hypothesis testing can lead to fallacious output.

You're not that young. They were still teaching null hypothesis testing in my college stat classes, and I assure you that was after the Dark Ages. Even today, I'd describe it as "controversial" or "to be used with care and a full understanding of its limits" rather than "replaced". It's not as if Bayesian analysis is without its critics either.

... Where *we* are the actors, barring some given explanation (and there isn't one as yet), isn't the best assumption that it is a random fluctuation?

If not, what is the "best" assumption?


The best action is to ask "why?" The best deterimination is that "I don't have enough information to answer the question." And if you or I need to make a decision the more powerful inference may be "not based on anything I have control over."

But what decision are you or I making. Whether to start him in a fantasy league. Then have at it.

This persona actually evolved because I'm more naturally interested to ask the "why?", but you run into people that insult you or shout you down for asking the question or speculating a plausible hypothesis. These "primery" like threads are closer to my "true personality level." Its just it doesn't often happen. To many people will poison the thread, and it really takes just one to go into the "I'm right mode" to take things down the wrong path.

You are confusing correlation with causality.


Ah, cute. I'll repeat, if everything in the space is causal then its causal. I can't "confuse correlation with causation" until I say that x influenced y amount of influence over z. So its nice that you think you know why i don't understand, but its simply not true.

With perfect information in a closed event space, then by definition everything is known. You can't falsify that no matter how much you try.

You can make a seperate point about the feasibility of information, but that becomes relevant only at the time that specific information is sought.

And I don't think you can falsify my express statement,
"But if its a closed space, and all events are causal, then the information does exist, and if you can get the information, then you will know.
"

You might be able to prove that my ability to know is intractable, but that is about it.

My point is that it's quite possible that there is true randomness (for all actors) at work here.


Well I don't think the position of electron z258 affects the utility of action. I don't think the quantum nature of particles effects this decision space. But that is an opinion.

And I've already had this quantum theory discussion once before. My ability to know something is not the same as what actually occurred. But more important, since the desired outcome is fixed and not a quantamized particle, do you think my looking at it again changes that outcome.

As long as we do not have information, there will be uncertainty, but the goal is not 100% foolproof outcomes, the goal is more powerful inferences that increase a utilitarian outcome.

They were still teaching null hypothesis testing in my college stat classes, and I assure you that was after the Dark Ages.

I imagine they still teach it now.

Even today, I'd describe it as "controversial" or "to be used with care and a full understanding of its limits" rather than "replaced".

Fair enough. It will still work for simple hypothesis. But the applications that the saberists now want to use are more complex. It would be like using Newtonian mechanics for long range missle telemetry and not accounting for the corolias effect.

Everything is hunky doory up to a point.

It's not as if Bayesian analysis is without its critics either.


I'm not sure that is "critical." That is just an application of Garbage In Garbage Out, and would apply equally to the prior distributions used for analysis with only classical statistical models.

I consider myself fairly intelligent, but I still can't tell what the difference is between Bayesian and other statistics. Something about the topic makes my eyes glaze over. I have to answer the damn phone, but I'll try rereading the link after that.

I learned all about H sub O and H sub N about 5 years ago, so I agree that it's still being taught (to my detriment?)

If what you mean is the normal person. That's fine. Your average person isn't going to think a 40 game hitting streak is "luck" They aren't going to think "Pitchers have no effect on balls in play and that it is luck."

BL, don't completely agree here; I believe that in examining most long hitting streaks, one could point to events within the closed space that can only be described as fortunate, at least from the point of view of the hitter. For example, a ball being judged a hit rather than an error; though there might be a reason for that (i.e. the scorer wants to prolong the streak), the outcome is fortunate for the batter. Now, that said, any hitter who achieves a 40 game streak, while the recipient of good fortune, is at the same time a very talented batter. IOW, 40 game hitting streaks don't happen to Joe Blow.

Is a pitch not breaking a random event?

Is a pitcher not throwing a pitch where he wants a random event?

No they aren't random - they are not influenced by "luck" unless his shoe is untied and he trips or something.

When an event is completly under a persons control (to throw a strike or a ball) and throws a ball the outcome is not luck.

BL, don't completely agree here

I'm not sure how far we are apart.

I don't think the average person is going to take the position, "It could happen to anybody." even though any major league regular has some probability of obtaining that streak taken from the information available at this point in time.

My point was limited to where "luck" causes a problem on this forum.

I do not disagree that any person that maintains a 40 game hitting streak will not be the beneficiary of some low probability outcome. I would not agree that Jimmy Rollins hitting streak makes him a better hitter, or a hitter more likely to get a hit than Albert Pujols in a random at bat encounter.

Conversely, the player is also going to be the unfortunate recipient of some low outcome events in that time space.

What is important is the fact that the hitter was able to maintained sustained production. Was there anything in that time span that created that sustained production that can be cheaply replicated for the future.

Now in asking the why, you may get tons of red herrings. And the streak itself is probably less important as aggregate production during the time period. but plausible theories should not be dismissed; they should be explored; and they should be discussed.

I consider myself fairly intelligent, but I still can't tell what the difference is between Bayesian and other statistics.

The biggest difference is that things classified as Bayesian deal with a system of analysis that at some point uses Bayes theorem. I think its charitable to call that "statistics".

But applications of Bayes theorem, or the use of Bayes factors, etc. are not the only alternative method of analysis. And I don't think analysis=statistics.

And if you or I need to make a decision the more powerful inference may be "not based on anything I have control over."

But what decision are you or I making. Whether to start him in a fantasy league. Then have at it.

"not based on anything I have control over".

"because the human actor that rolls the dice or tosses the coin typically does not have the physical dexterity to exhibit significant control on the outcome, and everyone who is watching at the time of selection and the level of instrumentation of just our eyes, will have to treat the roll as a random event."

That's how I use "luck".

Is a pitcher not throwing a pitch where he wants a random event?

When an event is completly under a persons control

Uh, do you not see a problem with those two statements?

MHS,

If I can answer your questions with a question...

If my 3-hole hitter hits a line drive at the shortstop (where the SS always plays other than DP depth), it will be an out. If the SS decides to break to the bag during the pitch in order to get ready for a backdoor throw from the catcher (the runner at 2 has been getting aggressive leads for three pitches now) and that same line drive is a run-scoring single, is that random? Or to follow up on your last statement, is my team lucky?

#50, it certainly isn't wholly random as you describe a pretty clear cause (the runner on 2nd) for the re-positioning of the SS. The runner isn't a randomly bouncing quark, right?

FB,

From the vantage point of the hitter, doesn't it seem pretty random? Why did the SS vacate his position on that particular pitch, and not the next one - or for that matter the previous one?

I can't speak to the math/definition of it (what with closed event spaces and such, I have no reason to doubt BLs veracity on all of this), I'm referring more to the conversation that happens between people on the bench or in the field or in the stands. How many times a game do people look at each other and think, "Damn, we were lucky there..."

That's why I asked MHS my follow up question - I can say that in the thousands of games I've been on the field, almost every game at least one event occurs that strikes me as being pretty dang random. Again, from my perspective, the perspective of an actor, in real time.

Chris (148), that's how I understand most people to use the term "luck". Again, I'd prefer the term "unpredictable" or "chance" or something like that, but I don't have a problem with "lucky".

Something about the topic makes my eyes glaze over.

Right there with you on this one. Wikipedia has a discussion which I didn't link, but it's about as clear as your ordinary textbook. Or mud.

That is just an application of Garbage In Garbage Out

I'd say it's a little more than that. Sometimes the validity of the prior distribution just can't be assessed. I think that's part of the issue here, which is why I'm more concerned with pragmatism than theory. In my view, what people are calling "lucky" are precisely those situations. Now, they probably decided that intuitively rather than formally, and you and Mike are right that, strictly speaking, a formal approach is better. But life is short.

And I've already had this quantum theory discussion once before.

That was with me. JCross, if you search the site (good luck) for "Stephen Hawking", you might find the previous thread.

Max, but the batter's perspective isn't the only one relevant to whether the event is 'random'. The runner on 2nd could be looking for catcher signals and knew the batter's 'pitch' was coming. The manager could have ordered the 3rd base coach to have the runner take a bigger lead to cause the SS to move. Of course it COULD be random, but I think you are making an ASSUMPTION that it was in your instance.

And you know what they say about assuming...

That's why I asked MHS my follow up question - I can say that in the thousands of games I've been on the field, almost every game at least one event occurs that strikes me as being pretty dang random. Again, from my perspective, the perspective of an actor, in real time.

And there is truth to that, that is why its a game and not a puzzle.

You can have the best team, do everything right, and still lose.

But the bigger question is when do you send the shortstop to cover; when do you try to exhibit influence to pull the ball (assuming RH).

And if you apply it to 1b, when do you hit and run, etc. How does that change with the pitchers. For the most part, the onfield personnel know most of these things pretty well without a model, and have made decisions without a written model. Sometimes they may not make the best decisions and that is where an analyst could help in improving the decision making.

I agree with Max - there are lots of things that occur in a baseball game that seem pretty random to me. If a pitch breaks or doesn't break isn't random. If a pitch breaks 5.35 inches or 5.56 inches is pretty random - but can easily make the difference between a home run and an out once the ball is hit. In fact, the exact same pitch thrown to the exact same hitter in the exact same situation at two different times can easily yield different results.

If this was track and field, I would say that very little is random. But in a sport where milimeters of difference at the beginning of an action can lead to dramatically different results, it is hard to escape the conclusion that randomness comes into play.

Maybe this is just me, but I think "luck" and "randomness" are not necessarily synonymous.

Maybe this is just me, but I think "luck" and "randomness" are not necessarily synonymous.

I agree, but I think in common usage they sometimes are.

When an event is completly under a persons control (to throw a strike or a ball) and throws a ball the outcome is not luck.

Well- what if-

The pitcher throws a beautiful pitch right on the corner- and the Ump blows the call.

The pitcher throws the ball a foot outside- but the batter chases it.

Even assuming the pitcher can perfectly control where the ball goes- the "outcome" still has random variables that are not under the pitcher's control.

You can have the best team, do everything right, and still lose.

was it you or Rauseo who once declared that "by definition" the best team is the team that won?

Well- what if-

That's easy, JPWF13. Score it E-ump.

I agree, but I think in common usage they sometimes are.

I think that was the main argument earlier in the thread, where people wanted to semantically disassociate luck from randomness. factors which can't be controlled are normalised, the more factors you can incorporate, the better obviously!

BackLasher and Mike : I agree with you in the greater sense that better instrumentation might help with better projections, but disagree that baseball is a closed event space. the god figure in the event space might have access to data like weather, wind, how many times a pitcher missed his location, what the pitch called was, positioning quirks et al, but still human reaction to external stimuli is non-deterministic. And since we have human actors involved, this can never be a truly closed event space. This IMO is what makes the game so interesting.
The question which is interesting is that whether we can project Francouer's career curve based on Francouer's last season stats. Shane Spencer or Victor Martinez?
MLEs are a nice benchmark to have, but are not very good at predicting the future. off the top of my head, an example would be Beltran. I am sure there are many other such cases.
The other question is when does available data become meaningful. To take a recent example, last season Andruw had an awesome spring, and there were people predicting a breakout season. When Andruw hit a huge slump in April/May, people were saying "same ol, same ol". Did anything in his approach or stats predict the offensive explosion that was imminent?

but disagree that baseball is a closed event space.

As I said, disagree all you want. The probability of the Atlanta Braves winning the 1995 World Series is 1. The probability of anybody else winning the 1995 World Series is 0. What happened, happened.

the god figure in the event space might have access to data like weather, wind, how many times a pitcher missed his location, what the pitch called was, positioning quirks et al, but still human reaction to external stimuli is non-deterministic. And since we have human actors involved, this can never be a truly closed event space.

Absolutley nothing you said, leads to the conclusion in the last sentence.

And people think, we are getting metaphysical.

What one can know is seperate from what happened. What is economically practical to know is seperate from what one can know.

This isn't metaphysical or semantical; it isn't determinism and it isn't about pre-destiny. Its not about Bell's Thereom or Heisenberg's Uncertainty Principal.

With perfect information, you can do lots of things. The ability to get perfect information is not available, and that is not being contested. And I don't need to talk about a God actor.

The event space is closed and that has nothing to do with what information you can get out of it.

Primarily people are talking about things that have been observable and recordable and using that information. And then other people are saying "no that's luck, here is my preseason projection and there is an x% chance it could happen"

Unless you want to get into a discussion of regularity, its not about "human's wanting to look for causes" because there are causes.

And the reason it becomes semantic, is because the logic being used is like null propogation. You are redefining terms, introducing something in that has intrinsic properties, and allowing it to propogate to conclusions that present a severly warped version of reality.

I agree, but I think in common usage they sometimes are.

I've got one word for you -- DIPS.

In common usage nobody is ever going to say that a pitcher getting tattooed is just unlucky. Its only in saberspace where this occurs.

was it you or Rauseo who once declared that "by definition" the best team is the team that won?


I don't know. When did you stop killing puppies?

What a loaded question. I tell you what, if you find that either of us said something contradictory in our statements, you can post it and get your "gotcha." Otherwise, I don't have a clue how even if one of us did do such a thing, it has anything to do with the validity of what is being said.

the "outcome" still has random variables that are not under the pitcher's control.

No, the outcome was caused by agents not within the pitchers control. When the pitcher releases the pitch, there are multitude of factors beyond the pitchers control the pitcher may treat as random.

Bayes factors replaced null hypothesis testing before I was even born

Backlasher's a Bayesian?

Holy ####!

I was have thought that at times when reading your various comments about how "what happened has happened and it's not luck, it's not random anymore", etc., but you invoked the almighty Bayes factor.

However, if you were testing a hypothesis that Billy Beane was a good general manager (even in the top 5% of General Managers of the laast 30 years), I find it difficult to believe that the BF in that situation would weigh heavily in his favour.

Because that's what we're really arguing about with respect to Beane. He's great vs. he's not that great. If we came up with a parameter that actually indicated how good a GM was and then measured the correct data for that parameter, I really think we'd find that the playoff losses would weigh in less than his regular season wins, unless you have by definition some 0-1 loss assigned to whether someone wins the World Series .

The pitcher throws a beautiful pitch right on the corner- and the Ump blows the call.

That's easy. First, you stare at him for 30 seconds. Then, you walk away. Finally, you throw your bat at him.

The probability of the Atlanta Braves winning the 1995 World Series is 1. The probability of anybody else winning the 1995 World Series is 0. What happened, happened.

and how is that relevant to what I was saying?
couple of clarifications first. by "god figure in an event space", I am not getting metaphysical, but just referring to the hypothetical expert who has all the available physical information. And in your definition of end result, yes the event space is closed. But if you consider the result to be a series of events, it won't be closed.
The braves won in 95, but my argument is that whether there is enouugh information in the event to be leveraged to predicting the next successful Braves team. In other words, it is the process which is of more interest. As I said in my previous post, there is a huge degree of uncertainity introduced by human reactions, and the best approach is to identify a process which will give you maximum chance of success.
Whether they were the best Braves team during the run or the luckiest is a matter of POV.

And that is what I gathered from your first post (#90), where I THOUGHT that you said that given a dataset, whether we can identify if there are two different curves, and if such a thing exists, identify the causal relationships between the dataset and the curves..

I really think we'd find that the playoff losses would weigh in less than his regular season wins

why? As a GM, you know your job entails not only getting your team to the playoffs, but also winning playoff series. If you are not putting your team in the best possible situation to win a playoff series ( not that I am saying he did or didn't ), you have failed at doing your job..

Because that's what we're really arguing about with respect to Beane. He's great vs. he's not that great. If we came up with a parameter that actually indicated how good a GM was and then measured the correct data for that parameter, I really think we'd find that the playoff losses would weigh in less than his regular season wins, unless you have by definition some 0-1 loss assigned to whether someone wins the World Series .


At a high level, that is close to what I am discussing. The problem is that people get in on the end of conversations, and you can't correct everyone.

My contentions wrt Beane have usually been he is an above average GM; severally overrated by a subset of the population; and that really gets in the way in both regular discussion and analysis.

There isn't much you can do when every action has this hidden supposition "that he's a genius and he must have figured out the market inefficiency with so-and-so."

I also do think that playoff series have different criteria that are worth studying; Beane should have, during the Alderson talent run, should have had the goal of achieving a World Series victory.

And as for being a "Bayesian", at least in the context of how this was applied in my space, I fell more into the Belief Theory crowd for the CCS stuff.

But Bayesian analysis is like Beethoven compared to the Britany Spears of classical statistical analysis.

I don't know. When did you stop killing puppies?


Bad analogy, usually your reasoning is better than that.
My question was leading but did not assume facts not in evidence" which is the flaw of the classic, "When did you stop beating your wife question".

But then again I knew full well that it was MHS who previously made the statement I tried to infer to you... :-)

Is a pitch not breaking a random event?

The pitch breaking or not breaking will be caused by something (or some combination of things) but that doesn't mean it isn't (or shouldn't be treated as) a "random" event. A gust or puff of wind could change the ball's airspeed by 20-30 mph which could effect the break of a pitch. The change in break had a cause, but that cause was effectively "random" from the perspective of the baseball players involved. Even a pitcher's control over his own body isn't absolute -- involuntary reflexes and feedback mechanisms influence a person's voluntary control over his body.

"Randomness" and "luck" in this context aren't meant to answer the question of whether an event had a cause, but whether that event was predictable based on known information and whether it is repeatable based on what is known after the fact. Even the hypothetical personal causes advanced above (hung over, lady problems, for example) have a luck element from our perspective -- to us it's "random" that Beckett was hung over or having lady problems before this particular start (if he was) rather than the previous or subsequent start while to him there was no doubt some clear explanation.

"Luck" is an admission of ignorance as to the cause of an event or a particular sequence of events rather than an affirmative explanation. (At least when offered as an answer for a "why" question. Using "luck" as a counterargument when someone advances a plausible causal explanation for an event is just laziness.)

Using "luck" as a counterargument when someone advances a plausible causal explanation for an event is just laziness.

Well, anyone can make up a theory that's possible but I've yet to see a reasonable causal explanation for why an A's team that has success in the regular season would fail in the post-season. If there is one it would give that way of thinking a lot more credence. Failing that, luck is still the MOST plausible explanation.

Well, anyone can make up a theory that's possible but I've yet to see a reasonable causal explanation for why an A's team that has success in the regular season would fail in the post-season.

Not to mention, that if the A's "station to station" style was so unsuited to the post season, how did they win 8 games? Every series went the limit, and most games were close.

Were Jackie Robinson's Dodgers similarly unsuited?

But if you consider the result to be a series of events, it won't be closed.

I can't gather or presently test any information on future events that have not yet occurred, so "my definition" is the only definition that creates a meaningful difference.

The braves won in 95, but my argument is that whether there is enouugh information in the event to be leveraged to predicting the next successful Braves team.

There is certainly enough information in the event space to make a determination. Whether that information is available to us, and whether the result of the causality is something we can exhibit control over are different questions.

And if you rewind a little bit, that is what is at issue, people cutting off plausible inferences that could be tested with information gathered in the future because the outcome is assigned to luck.

Well, anyone can make up a theory that's possible

And anyone can say its luck.

If there is one it would give that way of thinking a lot more credence.

You have a couple of contemparaneous studies on team composition and there varying levels of success in the postseason. You can start there and start identifying delta's if you like. Of course there is not much to the exercise if you determine "check, they had a dominant closer in Koch or Blotel" and then keep using your old models to validate it.

In fact, with the A's that's part of the problem. The A's during that period were probably maximized based on the known body of sabermetric research. So when people wanted to validate the A's, they validated them against that model. DIPS scores and everything. That's just circular.

Failing that, luck is still the MOST plausible explanation.

Says who? If that is your contention, I'd love to see the proof. Step 1 would be defining what you mean by "luck".

Because in this thread, despite all the crowing about semantics, you got Field saying "everybody means randomness" I can prove that's invalid and false pretty quickly. You got Dial saying it means events beyond the control of the actor. If that is the case, which actor are you talking about. Are you talking about the one actor that exhibits partial control over EVERY EVENT. I don't think you'll get a good power output on that inference. And then you've got the definition from PRP of laziness.

And then you've got the one that I see most often, the one that features some actor named luck that comes down and just ##### things up for ol Billy Boy.

Were Jackie Robinson's Dodgers similarly unsuited?

I don't know, you tell me. I saw a young team, prone to mental errors in crucial situations that manifested itself in blown saves, fielding errors and baserunning errors, that had trouble manufacturing a run, had trouble in close games, and couldn't do anything in big games. I saw a team that played station-to-station baseball. I saw a team that had trouble with the running game. I saw a team with players that had trouble against elite pitchers.

You saw bad luck. But the bigger problem is that rather than discuss the issue of process, you want to change this into "a defense of Billy Beane"

Its pretty simple on that front. If you want to be an analyst, it makes since to explore plausible theories rather than just keep shouting an opinion that "its luck." If you want to be a prognasticator, you lured me out into that one, I gave my prediction for 2003 and I was right, I gave my prediction for 2004, and I overestimated their prospects; and I've given my prediction for 2005 and 2006 and I'm looking pretty good on that one.

If you want to believe its "all luck" more power to you. I have no expectation that I will convince you otherwise or have you join in a conversation about the causal elements. Or do you think luck itself is a cause.

ou got Field saying "everybody means randomness" I can prove that's invalid and false pretty quickly.

By that, i don't mean I can prove Field is invalid, I mean with that definition of luck, I can prove pretty quickly that the A's loss is not because of randomness, because there isn't any randomness in the event space.

Well, anyone can make up a theory that's possible but I've yet to see a reasonable causal explanation for why an A's team that has success in the regular season would fail in the post-season. If there is one it would give that way of thinking a lot more credence. Failing that, luck is still the MOST plausible explanation.

I don't find any explanations I've read for the A's failures plausible. (I had another sentence after the one you quoted but took it out in favor of adding the word "plausible" in front of "causal explanation." I probably should have emphasised that qualifier more.) Explanations implicating "pressure" seem to me to substitute one nebulous term for another, fostering debate without advancing knowledge. Even if a luck person wins an argument against a pressure person (or vice versa), we learn nothing at all about how to construct a team better able to overcome the A's failures. The proponents of such arguments also tend to substitute the stress/pressure felt by the fan during an important moment for the stress/pressure felt by the player.

Even if a luck person wins an argument against a pressure person (or vice versa), we learn nothing at all about how to construct a team better able to overcome the A's failures.

No, but whether than winning an argument, if one takes a plausible theory of pressure, then one can certainly start doing the things needed to learn how to construct a team for better success, and our build identification and selection algorithms to test whether the hypothesis is correct. If one says \"#### it, it was luck," there isn't anywhere to go.

The proponents of such arguments also tend to substitute the stress/pressure felt by the fan during an important moment for the stress/pressure felt by the player.

Really. I don't imagine too many people other than A's fans, Yanks fans, Twins fans and Bosox fans were feeling too much pressure.
But I'll give you some props. Taken together, this is the modified Field argument. Until and unless you present a complete unified theory on the outcome that no one can cast any aspersions of doubt, we're just going to say it was x.

But your x is "unknown" and not luck, and as long as you aren't shouting down exploration of plausible theories, then You Win. You likely do have the best operable theory.

Now after I type that, I

I was a bit worried that'd be interpreted as more adversarial than I meant it. I just don't know what exactly pressure is in this context -- I don't know how to identify when a player is under its influence and when a player isn't. It's not so much a demand for a complete theory as it is a request for something identifiable. As interested observers I think we as fans are more apt to project our emotions and desires onto players, making it especially hard for us to determine when a player is feeling pressure and when he isn't. Being able to identify when a player is feeling pressure (beyond the normal level of pressure involved in being a major league baseball player) and when he isn't seems to me to be necessary before we can identify and select players resistent to the deleterious effects of pressure.

MLEs are a nice benchmark to have, but are not very good at predicting the future.

MLEs are a translation of the past not a look at the future.

There are fairly large year to year variations in player stats but they're basically the same whether you're talking MLE to major league or consecutive major league seasons.

You got Dial saying it means events beyond the control of the actor.

I was quoting *you* when I said that.