I’ve pointed out many times some of the weaknesses of traditional statistical practices, and how the routine use of “hypothesis testing” leads to an ever-growing mountain of Broken Science.
Very interesting. The same criticism applies to the Bell Inequalities, the purported proof that quantum mechanics is a description of reality as opposed to some kind of approximation that obscures hidden variables as yet unknown.
Personal view: Bayesian reasoning is a useful challenge to some errors and omissions in classical statistics, but I agree it's not remotely going to fix what's wrong with science, and it will be a damaging distraction to even try to go down that road.
I like the piece Briggs, but while I was reading it I started thinking about Realism vs. Nominalism. The model believers, if I followed your explanation rightly, are taking something that only has a nominal existence, probability, and treating it as if it has a real existence. They are trying to fit the example parameter to the ideal parameter. This surprised me a bit as I sort of took you for a realist(in the Scholastic sense). Now, I personally am not a Realist and think it is a bit of a silly fight actually. But I was curious if you had ever considered it in that light and if you'd care to enlighten us all a bit on how you see that subject.
Probability, as much as it can be said to exist, would be a particular arrangement or order of things not a thing itself. At first glance, Aristotelian hylomorphism doesn't appear to exclude the existence of probability, but if we consider actuality or certainty as the universal and probability as the degree to which the example exhibits that quality then it seems that anything whose actuality is incomplete does not exist and any thought whose certainty is incomplete is not true. From that perspective probability is an attempt to make gradations between true and not true and should certainly be rejected.
Saying that probability is purely epistemological makes your position very clear and I would say that that is a very reasonable position. I am trying to think if the model builders genuinely think that probability has an actual existence. Certainly the adherents of quantum BS do and so that idea sort of floats around our culture like a constant fart smell in the background. I suspect that most of the modelers have never thought about it.
We cannot deal with anything symbolic for long without assuming the reality of the symbol. For example, the letters on my monitor seem to have some reality or if I am programming the variable I define seem to be real things but their existence is very doubtful. Perhaps the prohibition against 'graven images' has utility outside of the explicitly spiritual context? Calling things that do not exist into being is something that we have a very limited and broken ability to do.
There should be a *gif* of that Chappelle Show clip of Rick James that says, "Reification's a helluva drug!"
This subject always reminds me of an old Steven Wright joke from the '80s: "I have map of the United States that's actual size... the scale reads '1 mile equals 1 mile.' ...Last summer I folded it."
He had a few other riffs on that but I always thought it was a brilliant introduction to the concept of reficiation.
Maybe we just need to consider bring back some of the old corporal punishment. Violations of reification in Math class will be punished by writing (in cursive) on the board 500 times after school "Probability does not inhere in a coin. Probability does not inhere in a coin."
The data are usually just a model of reality. As an example, I have power data from the inverter on my solar system. In order to build the dataset, samples are taken every five minutes of the power being driven to the household load. My well pump runs for a couple of minutes at a time to refill the tank when the pressure in the tank gets down to a threshold value. Hence, the power used by the pump sometimes is represented in the data and sometimes not. Either way is not a true representation of reality.
Briggs, I enjoy your writing and your speaking and consider you a foremost thinker and reasoner.
But ... I'm a probability-junkie who is "hooked on the product" and I don't need to warn you:
junkies are hard to reason with
We've shortly debated before about the independent existence of probability in the world, rather than it being merely a tool of cognition. I brought up the "inherent" probability of atomic clocks, things so stable that you can literally set your watch by them.
If a process is so stable that it'd take 900,000 years of running it before it is off by even one second, then it is something that you can count on -- just as if it truly exists in the world (like a truly-existing rock that you can sit on to rest when you are tired).
But let me change gears: What about a true existence of a relationship between probabilities?
A single coin is flipped two thousand times. The first thousand outcomes are compared with the second thousand outcomes. The probability of the first thousand outcomes having more than 51% heads (~0.25 with a fair coin) is compared with the probability of the second thousand outcomes having less than 41% heads (~0.000000007 with a fair coin).
With a fair coin, the first outcome is 36 million times more likely to occur. Even when the coin is only close to being fair, there will be a stable relationship between the probabilities (one being "inherently" higher than the other).
Claim: Repeating the 2,000-flip experiment 100 trillion more times will lead to no (zero) instances where the ratio of these two probabilities "reverses" -- i.e., formerly higher one becomes lower one -- just as if the probability relationship has "true existence" in the world.
Good work on removing almost 100% of the randomness in that tiny subset of all of the coin flips which occur in the world. I agree that you made one special kind of a coin flip predictable, but you did so at the cost of stretching the definition of a coin flip beyond its normal usefulness (definitions are good when they successfully distinguish between things that exist).
I originally feared you would come back with: "Over (2000)*(100 trillion) flips, the coin becomes progressively biased toward the 41% outcome (away from the 51% outcome) until, alas, the two probability conditions outlined become at least close to equi-probable -- refuting your initial claim."
Whew, it looks like I escaped having to defend against that!
I’d love to see you have this debate with Prof Norman Fenton… he’s very much a Bayesian
Very interesting. The same criticism applies to the Bell Inequalities, the purported proof that quantum mechanics is a description of reality as opposed to some kind of approximation that obscures hidden variables as yet unknown.
Personal view: Bayesian reasoning is a useful challenge to some errors and omissions in classical statistics, but I agree it's not remotely going to fix what's wrong with science, and it will be a damaging distraction to even try to go down that road.
"Yes Johnnie, you have watched hundreds of hours of Porn, but you are still a virgin."
Or
"I agree Nigel dear, women are OK but they are not like the real thing."`
Or
"I know the weather forecast says sunny, but I have just fetched the newspaper and I am soaking bloody wet."
I like the piece Briggs, but while I was reading it I started thinking about Realism vs. Nominalism. The model believers, if I followed your explanation rightly, are taking something that only has a nominal existence, probability, and treating it as if it has a real existence. They are trying to fit the example parameter to the ideal parameter. This surprised me a bit as I sort of took you for a realist(in the Scholastic sense). Now, I personally am not a Realist and think it is a bit of a silly fight actually. But I was curious if you had ever considered it in that light and if you'd care to enlighten us all a bit on how you see that subject.
Well, there are various forms of realism, and I fall into the Aristotelian camp.
But probability is purely epistemological. Which is not the hard-core realist position, that's for sure. I think that's more like a form of idealism.
In seriousness; what is the ~ism name for modeling based on reality?
Are you saying the Box of Grid squares I’ve been on a lifelong search since PFC was ...wasn’t real?
Are you suggesting that the terrain benchmarks I have spent years looking for aren’t there?
And the squelch oil?
It was all a lie?
"And Lo, did I search 29 palms for the ST-1s (st-ones) in vain because I lacked Wisdom..."
Probability, as much as it can be said to exist, would be a particular arrangement or order of things not a thing itself. At first glance, Aristotelian hylomorphism doesn't appear to exclude the existence of probability, but if we consider actuality or certainty as the universal and probability as the degree to which the example exhibits that quality then it seems that anything whose actuality is incomplete does not exist and any thought whose certainty is incomplete is not true. From that perspective probability is an attempt to make gradations between true and not true and should certainly be rejected.
Saying that probability is purely epistemological makes your position very clear and I would say that that is a very reasonable position. I am trying to think if the model builders genuinely think that probability has an actual existence. Certainly the adherents of quantum BS do and so that idea sort of floats around our culture like a constant fart smell in the background. I suspect that most of the modelers have never thought about it.
We cannot deal with anything symbolic for long without assuming the reality of the symbol. For example, the letters on my monitor seem to have some reality or if I am programming the variable I define seem to be real things but their existence is very doubtful. Perhaps the prohibition against 'graven images' has utility outside of the explicitly spiritual context? Calling things that do not exist into being is something that we have a very limited and broken ability to do.
There should be a *gif* of that Chappelle Show clip of Rick James that says, "Reification's a helluva drug!"
This subject always reminds me of an old Steven Wright joke from the '80s: "I have map of the United States that's actual size... the scale reads '1 mile equals 1 mile.' ...Last summer I folded it."
He had a few other riffs on that but I always thought it was a brilliant introduction to the concept of reficiation.
Maybe we just need to consider bring back some of the old corporal punishment. Violations of reification in Math class will be punished by writing (in cursive) on the board 500 times after school "Probability does not inhere in a coin. Probability does not inhere in a coin."
The data are usually just a model of reality. As an example, I have power data from the inverter on my solar system. In order to build the dataset, samples are taken every five minutes of the power being driven to the household load. My well pump runs for a couple of minutes at a time to refill the tank when the pressure in the tank gets down to a threshold value. Hence, the power used by the pump sometimes is represented in the data and sometimes not. Either way is not a true representation of reality.
Briggs, I enjoy your writing and your speaking and consider you a foremost thinker and reasoner.
But ... I'm a probability-junkie who is "hooked on the product" and I don't need to warn you:
junkies are hard to reason with
We've shortly debated before about the independent existence of probability in the world, rather than it being merely a tool of cognition. I brought up the "inherent" probability of atomic clocks, things so stable that you can literally set your watch by them.
If a process is so stable that it'd take 900,000 years of running it before it is off by even one second, then it is something that you can count on -- just as if it truly exists in the world (like a truly-existing rock that you can sit on to rest when you are tired).
But let me change gears: What about a true existence of a relationship between probabilities?
A single coin is flipped two thousand times. The first thousand outcomes are compared with the second thousand outcomes. The probability of the first thousand outcomes having more than 51% heads (~0.25 with a fair coin) is compared with the probability of the second thousand outcomes having less than 41% heads (~0.000000007 with a fair coin).
With a fair coin, the first outcome is 36 million times more likely to occur. Even when the coin is only close to being fair, there will be a stable relationship between the probabilities (one being "inherently" higher than the other).
Claim: Repeating the 2,000-flip experiment 100 trillion more times will lead to no (zero) instances where the ratio of these two probabilities "reverses" -- i.e., formerly higher one becomes lower one -- just as if the probability relationship has "true existence" in the world.
What say you, Briggs?
Here's something very stable:
https://www.wmbriggs.com/post/48446/
That machine is cool.
Good work on removing almost 100% of the randomness in that tiny subset of all of the coin flips which occur in the world. I agree that you made one special kind of a coin flip predictable, but you did so at the cost of stretching the definition of a coin flip beyond its normal usefulness (definitions are good when they successfully distinguish between things that exist).
I originally feared you would come back with: "Over (2000)*(100 trillion) flips, the coin becomes progressively biased toward the 41% outcome (away from the 51% outcome) until, alas, the two probability conditions outlined become at least close to equi-probable -- refuting your initial claim."
Whew, it looks like I escaped having to defend against that!
William,
Are you a mathematician by trade?