What is intelligence, and is there a genetic basis for it?

[Alisa Zinov'yevna Rosenbaum]

Someone asked me:
> Let's say your goal was to push your country to be better at football? Would you try to encourage large people or small people to reproduce? Black people or white people? Why?

In order for the question to be relevant, there must be some theories
about how encouraging certain people to reproduce could possibly make
a country better at some sport. Maybe you are thinking of something
like the following:

Football is an artificial game, a human-created problem, and the rules
prohibit players from augmenting themselves with technology in all but
a few limited ways. This makes important the physical characteristics
of the players like height, weight, strength, and speed. (Note that
some of these can be improved with ideas and training, and there are
drugs people can take to improve at all these things, though only some
are allowed under the current rules.) We have good explanatory
theories for how physical characteristics like size are inherited, and
so if we want bigger players, and our only way of influencing this is
who reproduces, then perhaps we want to encourage big people to
reproduce. Similarly, if not being seen at night was important in the
game, and players had to play naked, then we might want more black
players.

By broadening the range of football improvement strategies we
consider, we open the door to drastically more effective strategies.
What then is the most effective strategy?

In general, if some person or group of people want to solve some
problem, then the most important thing is for the people in that group
to learn better ideas, so they can create more wealth and knowledge,
which they can apply towards solving the problem. So the best way for
a country to get better at football is to improve their ideas, become
wealthier, and apply those wealth and ideas towards creating better
height/mass/strength/speed-enhancing drugs, plays, training regimes,
ways of analyzing opponents, etc. (This idea is based on my reading of
http://curi.us/1169-morality; scroll down to section about squirrels.)

[Blue Yogin]

Hi. I am the original asker of this question, and I'm new to the list. Happy to be here. Some of the context was lost here - we had been talking about the heritability upper and lower bounds for some common phenotypes, chief among them intelligence. (Or, if you prefer, "the ability to do well on IQ tests"). Whenever I talk about this I try to switch to sports analogies as quickly as possible. I think we have better intuitions about athletic ability than about mental ability, and they have the same basic statistical properties, so it's easy to switch between them. So, that's the context. 

To make this question harder and to avoid sneaking away from the pain I'm trying to inflict, allow me to add another proviso.  Suppose that any good idea in football can be quickly copied by other teams. Thus, though individual coaches might have some good ideas about how to train better or play better, whenever a good idea like that is shown to be successful, it quickly sweeps the entire league, and the game returns to an equilibrium state in which every team employs the best training and playing strategies, so that the only difference between teams are physical differences that can't be easily copied. 

And let me propose another thought experiment designed to evoke the same sort of question, but which makes, I feel, more sense absent the original context. The point of the question is to force you to make a decision without the benefit of a satisfying "explanatory theory" beyond a statistical likelihood calculation. 

To me, statistics is in the business of explanation - whatever that means. A statistician is perfectly happy to run a regression analysis and then assert that IQ explains 65±5% of a US citizen's lifetime earning capacity, without having a fully baked idea of why that might be so. It seems from your usage that you, with David Deutsch, assign a different meaning to the term explanation. The point I am driving at is that neither of you seem to have given a bright line past which something can be said to be explained. It seems to me that there is a just a continuum of explanation quality, and contra Deutsch, statistics is the quantification that continuum. I need to understand how an explanation that you consider "good" differs in kind from a statistical explanation. Deutsch has a little bit of this with "hard to vary" and "reach". Let me try to defeat these right now.

Let's say that twenty of us live in a village.  We find some red mushrooms and some blue mushrooms in the forest near our village. We've never seen either kind before. Ten of us eat the red mushrooms, and ten of us eat the blue mushrooms. A day later, we find that four people who ate the red mushrooms have died in their beds. Everyone who ate blue mushrooms feels fine.

We don't have the benefit of gas chromatography, and can't do an assay to find out the various exotic compounds in the mushrooms. We don't have the benefit of forensic medicine that would allow us to determine the cause of death for the red mushroom eaters. But we can run a p-test, so let's do that.

The null hypothesis is that neither red nor blue mushrooms affected anybody's physiology. The deaths occurred for some other reason. We assume for our test that the deaths were randomly distributed throughout the population of 20. 

This is ye old balls and urns. The number of distinguishable ways that all four deaths could have fallen into the red camp is 10 chose 4.

The number of distinguishable ways that the deaths could have been distributed is 20 chose 4.

Dividing, we find that the probability of only reds having fallen, assuming all configurations are equality likely, is 4.33e-2, significant to within the p > .05 level. (And someone please chip in if I've got this math wrong, but even if I have, I think the point stands. You can change the numbers to make this scenario work).

So, we have statistical significance, but no real theory - right? Maybe it's something else. We don't have a mechanism. Maybe we should keep eating red mushrooms. Should we, or should we not? And how are we to make that decision? We have to eat something. We have to eat something today. We can't wait until the invention of gas chromatography and forensic toxicology, right?

"Eating red mushrooms might kill you" is a theory of reality. It is hard to vary: if we changed it to blue mushrooms, it wouldn't fit the facts. If we changed it to "toads", it wouldn't fit the facts: it wouldn't explain why only the red-mushroom-eaters died. It has "reach": the theory implies that if the tribe traveled a thousand miles away and encountered red mushrooms there, it still shouldn't eat them because they would still be poisonous. It might even imply that we shouldn't let our beloved pets eat them, etc. How does it differ in kind from explanations that a fallible ideas proponent would call "good"? Is it not just another theory, somewhere on the quality spectrum?

--
You received this message because you are subscribed to the Google Groups "Fallible Ideas" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fallible-idea...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[Alisa Zinov'yevna Rosenbaum]

> Hi. I am the original asker of this question, and I'm new to the list. Happy to be here. Some of the context was lost here - we had been talking about the heritability upper and lower bounds for some common phenotypes, chief among them intelligence. (Or, if you prefer, "the ability to do well on IQ tests”).

First, a question about heritability: If a trait is 100% heritable, what does that say about the intervention strategies for changing that trait? How about if a trait is 0% heritable?

Second, a note on the meaning of intelligence. I don’t consider the ability to score highly on IQ tests to be equivalent to intelligence. IQ tests and the Turing test are both operational definitions that don’t explain the underlying phenomena. There’s already a convenient word for the ability to score highly on IQ tests, so, for clarity, let’s use it (“IQ”) rather than "intelligence", if that is indeed what you mean.

> Whenever I talk about this I try to switch to sports analogies as quickly as possible. I think we have better intuitions about athletic ability than about mental ability, and they have the same basic statistical properties, so it's easy to switch between them.

By shared statistical properties, I guess you mean that tests can be devised (such as a baseball batting average, golf score, or IQ test score) for which the results follow a normal distribution across some population.

> And let me propose another thought experiment designed to evoke the same sort of question, but which makes, I feel, more sense absent the original context. The point of the question is to force you to make a decision without the benefit of a satisfying "explanatory theory" beyond a statistical likelihood calculation.
>
> To me, statistics is in the business of explanation - whatever that means.

This is an idiosyncratic usage of either statistics or explanation. (Luckily you qualified it with “to me”.) Statistics has a rich vocabulary, so how about if we use the specific terms from statistics?

> A statistician is perfectly happy to run a regression analysis and then assert that IQ explains 65±5% of a US citizen's lifetime earning capacity, without having a fully baked idea of why that might be so.

Aren’t there are a lot of assumptions (theories) that go into a doing a regression analysis properly? (http://www.ma.utexas.edu/users/mks/statmistakes/regression.html) For example, think about elementary school students. As they get older, their shoe size increases, along with their reading scores. Would a statistician also be happy to assert that shoe size explains a large percentage of reading score among that population?

> It seems from your usage that you, with David Deutsch, assign a different meaning to the term explanation.

As far as I can tell, that statistical use of the term “explanation” is technical and has nothing to do with the common meaning of the word as used by Deutsch and the rest of the English-speaking world. I think this discussion would be easier if we reserve the word “explanation” for things like the axis-tilt explanation of seasons on Earth.

> The point I am driving at is that neither of you seem to have given a bright line past which something can be said to be explained. It seems to me that there is a just a continuum of explanation quality, and contra Deutsch, statistics is the quantification that continuum.

No, we need theories/explanations in order to interpret our statistics correctly in the first place.

> I need to understand how an explanation that you consider "good" differs in kind from a statistical explanation.

This is a bit hand-wavy, but an explanation is a model of the world which helps us understand cause and effect. A good explanation, like a good idea, is simply one that is un-refuted. In a better world, it would be common to criticize explanations that are easy to vary or have limited reach.

> Deutsch has a little bit of this with "hard to vary" and "reach". Let me try to defeat these right now.
>
> Let's say that twenty of us live in a village. We find some red mushrooms and some blue mushrooms in the forest near our village. We've never seen either kind before. Ten of us eat the red mushrooms, and ten of us eat the blue mushrooms. A day later, we find that four people who ate the red mushrooms have died in their beds. Everyone who ate blue mushrooms feels fine.
>
> We don't have the benefit of gas chromatography, and can't do an assay to find out the various exotic compounds in the mushrooms. We don't have the benefit of forensic medicine that would allow us to determine the cause of death for the red mushroom eaters. But we can run a p-test, so let's do that.
>
> The null hypothesis is that neither red nor blue mushrooms affected anybody's physiology. The deaths occurred for some other reason. We assume for our test that the deaths were randomly distributed throughout the population of 20.
>
> This is ye old balls and urns. The number of distinguishable ways that all four deaths could have fallen into the red camp is 10 chose 4.
>
> The number of distinguishable ways that the deaths could have been distributed is 20 chose 4.
>
> Dividing, we find that the probability of only reds having fallen, assuming all configurations are equality likely, is 4.33e-2, significant to within the p > .05 level.
>

> So, we have statistical significance, but no real theory - right?

We have a theory that the mushrooms are poisonous. It’s a rough theory, in that we don’t understand the exact mechanism of the poison, but it falls into a class of physical/biological theories in which some substance disrupts the normal functioning of the human body

> Maybe it's something else. We don't have a mechanism. Maybe we should keep eating red mushrooms. Should we, or should we not? And how are we to make that decision? We have to eat something. We have to eat something today. We can't wait until the invention of gas chromatography and forensic toxicology, right?

If we have limited time to do more studies and limited knowledge about the situation, then it sounds like the best thing to do is to avoid eating the red mushrooms for now.

> "Eating red mushrooms might kill you" is a theory of reality. It is hard to vary: if we changed it to blue mushrooms, it wouldn't fit the facts. If we changed it to "toads", it wouldn't fit the facts: it wouldn't explain why only the red-mushroom-eaters died. It has "reach": the theory implies that if the tribe traveled a thousand miles away and encountered red mushrooms there, it still shouldn't eat them because they would still be poisonous. It might even imply that we shouldn't let our beloved pets eat them, etc. How does it differ in kind from explanations that a fallible ideas proponent would call "good"? Is it not just another theory, somewhere on the quality spectrum?

The mushrooms being poisonous is a good enough explanation for the problem faced by the villagers, given the circumstances (what’s at stake, the time frame, their level of knowledge and wealth, etc.). For one thing, no one in the village seems to have any good rival explanations.

[Blue Yogin]

Heritability is by definition whatever is left over after you've statistically accounted for "shared" and "non-shared" environment. All intervention strategies would fall into one of the environment categories. Check the caveats list on the heritability of IQ wiki to see what's being precisely asserted.

Second, a note on the meaning of intelligence. I don’t consider the ability to score highly on IQ tests to be equivalent to intelligence. IQ tests and the Turing test are both operational definitions that don’t explain the underlying phenomena. There’s already a convenient word for the ability to score highly on IQ tests, so, for clarity, let’s use it (“IQ”) rather than "intelligence", if that is indeed what you mean.

This is the crux of my trap, which I was waiting to spring until you agreed that we shouldn't eat the mushrooms: If we can say that mushrooms are poisonous without understanding how the poison they allegedly contain disrupts our normal functioning, why can't we say that high-IQ-scorers are more intelligent without understanding how intelligence makes us better at IQ tests? How is "you eat them and then you die" anything but an "operational definition"? If they're both operational definitions, why is it acceptable to base decisions off of an operational definition in one case but not the other?

Whenever I talk about this I try to switch to sports analogies as quickly as possible. I think we have better intuitions about athletic ability than about mental ability, and they have the same basic statistical properties, so it's easy to switch between them.

By shared statistical properties, I guess you mean that tests can be devised (such as a baseball batting average, golf score, or IQ test score) for which the results follow a normal distribution across some population.

That's one important property. Another is that, in terms of fame and lasting importance, almost no footballer or scientistic below +3 SD is ever remembered. This is important, because the shape of the ends of distributions is not like reasoning about the means of distributions.

And let me propose another thought experiment designed to evoke the same sort of question, but which makes, I feel, more sense absent the original context. The point of the question is to force you to make a decision without the benefit of a satisfying "explanatory theory" beyond a statistical likelihood calculation. 

To me, statistics is in the business of explanation - whatever that means.

This is an idiosyncratic usage of either statistics or explanation. (Luckily you qualified it with “to me”.) Statistics has a rich vocabulary, so how about if we use the specific terms from statistics?

I'm using statistic's rich vocabulary here. What the statisticians have done is achieved a reduction from a common-sense notion of "explanation" to a quantification. It's not an accident that they're using the same word.

A statistician is perfectly happy to run a regression analysis and then assert that IQ explains 65±5% of a US citizen's lifetime earning capacity, without having a fully baked idea of why that might be so.

Aren’t there are a lot of assumptions (theories) that go into a doing a regression analysis properly? (http://www.ma.utexas.edu/users/mks/statmistakes/regression.html) For example, think about elementary school students. As they get older, their shoe size increases, along with their reading scores. Would a statistician also be happy to assert that shoe size explains a large percentage of reading score among that population?

No, part of our analysis would involve controlling for age and then we'd find that shoe size is uncorrelated with reading score. Or, we'd find some tiny correlation. We'd find that shoe size is a bad explanation, and we'd be able to quantify how bad of an explanation it is on a scale of 0 to 100%.

It seems from your usage that you, with David Deutsch, assign a different meaning to the term explanation.

As far as I can tell, that statistical use of the term “explanation” is technical and has nothing to do with the common meaning of the word as used by Deutsch and the rest of the English-speaking world. I think this discussion would be easier if we reserve the word “explanation” for things like the axis-tilt explanation of seasons on Earth.

Common meanings of words lead us to fallible ideas all the time, don't they?

The point I am driving at is that neither of you seem to have given a bright line past which something can be said to be explained. It seems to me that there is a just a continuum of explanation quality, and contra Deutsch, statistics is the quantification that continuum.

No, we need theories/explanations in order to interpret our statistics correctly in the first place.

That's true, but statistics are how we quantify the goodness or badness of our theories. They aren't the theories themselves.

I need to understand how an explanation that you consider "good" differs in kind from a statistical explanation.

This is a bit hand-wavy, but an explanation is a model of the world which helps us understand cause and effect. A good explanation, like a good idea, is simply one that is un-refuted. In a better world, it would be common to criticize explanations that are easy to vary or have limited reach.

Deutsch has a little bit of this with "hard to vary" and "reach". Let me try to defeat these right now.

Let's say that twenty of us live in a village.  We find some red mushrooms and some blue mushrooms in the forest near our village. We've never seen either kind before. Ten of us eat the red mushrooms, and ten of us eat the blue mushrooms. A day later, we find that four people who ate the red mushrooms have died in their beds. Everyone who ate blue mushrooms feels fine.

We don't have the benefit of gas chromatography, and can't do an assay to find out the various exotic compounds in the mushrooms. We don't have the benefit of forensic medicine that would allow us to determine the cause of death for the red mushroom eaters. But we can run a p-test, so let's do that.

The null hypothesis is that neither red nor blue mushrooms affected anybody's physiology. The deaths occurred for some other reason. We assume for our test that the deaths were randomly distributed throughout the population of 20. 

This is ye old balls and urns. The number of distinguishable ways that all four deaths could have fallen into the red camp is 10 chose 4.

The number of distinguishable ways that the deaths could have been distributed is 20 chose 4.

Dividing, we find that the probability of only reds having fallen, assuming all configurations are equality likely, is 4.33e-2, significant to within the p > .05 level. 

So, we have statistical significance, but no real theory - right?

We have a theory that the mushrooms are poisonous. It’s a rough theory, in that we don’t understand the exact mechanism of the poison, but it falls into a class of physical/biological theories in which some substance disrupts the normal functioning of the human body

Maybe it's something else. We don't have a mechanism. Maybe we should keep eating red mushrooms. Should we, or should we not? And how are we to make that decision? We have to eat something. We have to eat something today. We can't wait until the invention of gas chromatography and forensic toxicology, right?

If we have limited time to do more studies and limited knowledge about the situation, then it sounds like the best thing to do is to avoid eating the red mushrooms for now.

"Eating red mushrooms might kill you" is a theory of reality. It is hard to vary: if we changed it to blue mushrooms, it wouldn't fit the facts. If we changed it to "toads", it wouldn't fit the facts: it wouldn't explain why only the red-mushroom-eaters died. It has "reach": the theory implies that if the tribe traveled a thousand miles away and encountered red mushrooms there, it still shouldn't eat them because they would still be poisonous. It might even imply that we shouldn't let our beloved pets eat them, etc. How does it differ in kind from explanations that a fallible ideas proponent would call "good"? Is it not just another theory, somewhere on the quality spectrum?

The mushrooms being poisonous is a good enough explanation for the problem faced by the villagers, given the circumstances (what’s at stake, the time frame, their level of knowledge and wealth, etc.). For one thing, no one in the village seems to have any good rival explanations.

[Alisa Zinov'yevna Rosenbaum]

On Dec 30, 2013, at 5:53 AM, Blue Yogin <blue....@gmail.com> wrote:

On Dec 29, 2013, at 7:59 PM, Blue Yogin <blue....@gmail.com> wrote:

Hi. I am the original asker of this question, and I'm new to the list. Happy to be here. Some of the context was lost here - we had been talking about the heritability upper and lower bounds for some common phenotypes, chief among them intelligence. (Or, if you prefer, "the ability to do well on IQ tests”).

First, a question about heritability: If a trait is 100% heritable, what does that say about the intervention strategies for changing that trait? How about if a trait is 0% heritable?

Heritability is by definition whatever is left over after you've statistically accounted for "shared" and "non-shared" environment.

You didn't answer my question. My question is, how does merely knowing that something is (roughly) 0% heritable or 100% bear on what interventions might be effective?

Also, what is the formula for heritability, and in that formula, what are the terms for shared and non-shared environment? Are researchers determine what falls into each category, without explanations of how the environment causes the things being measured?

All intervention strategies would fall into one of the environment categories. Check the caveats list on the heritability of IQ wiki to see what's being precisely asserted.

I looked at the caveats and it looks like there is a lot of background theories at work there. Interpreting heritability statistics involves a lot more than just pure extrapolation from numbers, right?

Would you agree that, among the US population of the 1950s:

- Wearing earings was highly heritable?

- Number of digits (fingers + toes) was roughly 0% heritable?

Suppose we come up with a quantitative measure for accent. (Along the lines of http://www.nytimes.com/interactive/2013/12/20/sunday-review/dialect-quiz-map.html?_r=0) Accent would be highly heritable, right?

Second, a note on the meaning of intelligence. I don’t consider the ability to score highly on IQ tests to be equivalent to intelligence. IQ tests and the Turing test are both operational definitions that don’t explain the underlying phenomena. There’s already a convenient word for the ability to score highly on IQ tests, so, for clarity, let’s use it (“IQ”) rather than "intelligence", if that is indeed what you mean.

This is the crux of my trap, which I was waiting to spring until you agreed that we shouldn't eat the mushrooms: If we can say that mushrooms are poisonous without understanding how the poison they allegedly contain disrupts our normal functioning, why can't we say that high-IQ-scorers are more intelligent without understanding how intelligence makes us better at IQ tests? How is "you eat them and then you die" anything but an "operational definition"? If they're both operational definitions, why is it acceptable to base decisions off of an operational definition in one case but not the other?

In the berry scenario, the villagers didn't think, "you eat them and you die", they must have thought, "something in them poisons you".  Because maybe if they counted the number of syllables said by each of the people in the tribe who died that day, it would turn out to be a prime number between 3000 and 4000, while everyone else spoke more or less. So should everyone be careful to speak less than 3000 syllables and more than 4000 syllables? My point is that in hindsight the villagers could have found an infinite number of ways to distinguish those who died from those who didn't. How did they even know to focus on the berries in the first place? Because they had theories about poisonous foods. 

Also, "you eat them and then you die" is too broad for what you described - it doesn't distinguish between things like:

- if you eat Farmer Magee's red berries, he shoots you and you die

- everyone who is sick and about to die is given a special final meal of red berries that, for traditional reasons, is eaten by no one else

Same with intelligence - there must be theories (explanations) linking IQ scores with intelligence, and these can be analyzed once they are stated explicitly.

Whenever I talk about this I try to switch to sports analogies as quickly as possible. I think we have better intuitions about athletic ability than about mental ability, and they have the same basic statistical properties, so it's easy to switch between them.

By shared statistical properties, I guess you mean that tests can be devised (such as a baseball batting average, golf score, or IQ test score) for which the results follow a normal distribution across some population.

That's one important property. Another is that, in terms of fame and lasting importance, almost no footballer or scientistic below +3 SD is ever remembered.

Below +3 SD on what test(s) for footballers and scientists? And what makes us think that particular test is relevant, in each case?

This is important, because the shape of the ends of distributions is not like reasoning about the means of distributions.

Can you elaborate? Are you saying the distribution isn't normal? Or that lots of people are outliers?

And let me propose another thought experiment designed to evoke the same sort of question, but which makes, I feel, more sense absent the original context. The point of the question is to force you to make a decision without the benefit of a satisfying "explanatory theory" beyond a statistical likelihood calculation. 

To me, statistics is in the business of explanation - whatever that means.

This is an idiosyncratic usage of either statistics or explanation. (Luckily you qualified it with “to me”.) Statistics has a rich vocabulary, so how about if we use the specific terms from statistics?

I'm using statistic's rich vocabulary here. What the statisticians have done is achieved a reduction from a common-sense notion of "explanation" to a quantification. It's not an accident that they're using the same word.

Can you provide a citation on the web for this usage? http://en.wikipedia.org/wiki/Statistics only uses the word "explanation" once, and it's as part of a larger sentence listing all the things statistics is about.

Also, is explanation the ONLY term used in statistics for this concept, or do they have an alternate term we could use?

A statistician is perfectly happy to run a regression analysis and then assert that IQ explains 65±5% of a US citizen's lifetime earning capacity, without having a fully baked idea of why that might be so.

Aren’t there are a lot of assumptions (theories) that go into a doing a regression analysis properly? (http://www.ma.utexas.edu/users/mks/statmistakes/regression.html) For example, think about elementary school students. As they get older, their shoe size increases, along with their reading scores. Would a statistician also be happy to assert that shoe size explains a large percentage of reading score among that population?

No, part of our analysis would involve controlling for age and then we'd find that shoe size is uncorrelated with reading score.

Without explanations of what things could be relevant to the experiment and how they would affect the outcome, how would you know what to control for?

Or, we'd find some tiny correlation. We'd find that shoe size is a bad explanation, and we'd be able to quantify how bad of an explanation it is on a scale of 0 to 100%.

It seems from your usage that you, with David Deutsch, assign a different meaning to the term explanation.

As far as I can tell, that statistical use of the term “explanation” is technical and has nothing to do with the common meaning of the word as used by Deutsch and the rest of the English-speaking world. I think this discussion would be easier if we reserve the word “explanation” for things like the axis-tilt explanation of seasons on Earth.

Common meanings of words lead us to fallible ideas all the time, don't they?

Sure, but for communicating we start with the common meanings. Here you don't seem to think that's a good approach, so I'll happily adopt whatever terminology you use to distinguish between the two concepts.  I think the issue with that, though, is that you don't recognize a difference between them. Is that right?

The point I am driving at is that neither of you seem to have given a bright line past which something can be said to be explained. It seems to me that there is a just a continuum of explanation quality, and contra Deutsch, statistics is the quantification that continuum.

No, we need theories/explanations in order to interpret our statistics correctly in the first place.

That's true, but statistics are how we quantify the goodness or badness of our theories. They aren't the theories themselves.

So statistics, together with theories on how to interpret them, constitute a quantitative way to evaluate other theories? If so, I agree.

[Mason Kramer]

Second, a note on the meaning of intelligence. I don’t consider the ability to score highly on IQ tests to be equivalent to intelligence. IQ tests and the Turing test are both operational definitions that don’t explain the underlying phenomena. There’s already a convenient word for the ability to score highly on IQ tests, so, for clarity, let’s use it (“IQ”) rather than "intelligence", if that is indeed what you mean. This is the crux of my trap, which I was waiting to spring until you agreed that we shouldn't eat the mushrooms: If we can say that mushrooms are poisonous without understanding how the poison they allegedly contain disrupts our normal functioning, why can't we say that high-IQ-scorers are more intelligent without understanding how intelligence makes us better at IQ tests? How is "you eat them and then you die" anything but an "operational definition"? If they're both operational definitions, why is it acceptable to base decisions off of an operational definition in one case but not the other? In the berry scenario, the villagers didn't think, "you eat them and you die", they must have thought, "something in them poisons you".  Because maybe if they counted the number of syllables said by each of the people in the tribe who died that day, it would turn out to be a prime number between 3000 and 4000, while everyone else spoke more or less. So should everyone be careful to speak less than 3000 syllables and more than 4000 syllables? My point is that in hindsight the villagers could have found an infinite number of ways to distinguish those who died from those who didn't. How did they even know to focus on the berries in the first place? Because they had theories about poisonous foods.  Also, "you eat them and then you die" is too broad for what you described - it doesn't distinguish between things like: - if you eat Farmer Magee's red berries, he shoots you and you die - everyone who is sick and about to die is given a special final meal of red berries that, for traditional reasons, is eaten by no one else Same with intelligence - there must be theories (explanations) linking IQ scores with intelligence, and these can be analyzed once they are stated explicitly.

It is the tribe's definition of "poisonous" that is "operational" - to use your terminology. The definition is "something is poisonous if eating it causes you to die." Analogously, the definition of intelligence is operational. The definition is, "how well you do on g-loaded tests".  There's as much or more theory behind why g-loaded tests predict or explain life success as there surrounding the tribe's new red mushroom taboo. The g theory of intelligence is much more well-developed, but you're willing to accept the red mushroom taboo and not the g theory. This is "a problem in need of an explanation", as Deutsch might say.

You didn't answer my question. My question is, how does merely knowing that something is (roughly) 0% heritable or 100% bear on what interventions might be effective?

Yes, I did, actually. I said:

All intervention strategies would fall into one of the environment categories. 

In other words, 100% heritability of a trait means no intervention will make any difference.

Can you provide a citation on the web for this usage? http://en.wikipedia.org/wiki/Statistics only uses the word "explanation" once, and it's as part of a larger sentence listing all the things statistics is about.
Also, is explanation the ONLY term used in statistics for this concept, or do they have an alternate term we could use?

See also http://en.wikipedia.org/wiki/Dependent_and_independent_variables, where the independent variables are also known as the explanatory variables. Another term would be "predictor variables". Anyway, "X explains/accounts for 60% of the observed variability" is a common idiom in regression analysis. I'm not making it up.

Without explanations of what things could be relevant to the experiment and how they would affect the outcome, how would you know what to control for?

Yes, you need a theory - also called a model. As I asserted in my original question:

there is a just a continuum of explanation quality, and contra Deutsch, statistics is the quantification [of] that continuum.

 

To summarize my original position: all theories exist on a continuum of good and bad and do not differ in kind from each other. Statistics is the way that we quantify the relative goodness or badness of a theory. "Easy to vary" and "reach" are mushy and useless compared to statistical interpretation. The red mushroom example is designed to show that every event in our entire lives is cloaked in various degrees of ignorance about root causes, and we use statistics to distinguish between these relative levels of ignorance. The key of the story is that we have a theory that things are poisonous but an operational definition of poison, and that's good enough.

[Alisa Zinov'yevna Rosenbaum]

In your reply here, you ignored - and simply cut out of your reply - a lot of
my questions. I asked them because I thought they would be helpful in this
discussion. Did you ignore them because they were irrelevant, would take too
long to answer, misleading, or what?

A few of my questions should require only a yes or no (I think). For your
convenience I repeat those ones here:

Would you agree that, among the US population of the 1950s:
- Wearing earings was highly heritable?
- Number of digits (fingers + toes) was roughly 0% heritable?

Suppose we come up with a quantitative measure for accent. (Along the lines of
http://www.nytimes.com/interactive/2013/12/20/sunday-review/dialect-quiz-map.html?_r=0)

Accent would be highly heritable, right?

> It is the tribe's definition of "poisonous" that is "operational" - to use
> your terminology. The definition is "something is poisonous if eating it
> causes you to die."

There’s a lot hidden behind the word “causes”. All statistics can say is that
eating the berries is correlated with death, right? It’s up to our
interpretations/ideas/theories to make sure the causation is as we think it is.

> Analogously, the definition of intelligence is operational. The definition
> is, "how well you do on g-loaded tests".

That's *a* definition of intelligence. It's not the dictionary definition...

> There's as much or more theory behind why g-loaded tests predict or explain
> life success as there surrounding the tribe's new red mushroom taboo.

Great! Those theories are some of the things we should be talking about.

> The g theory of intelligence is much more well-developed, but you're willing
> to accept the red mushroom taboo and not the g theory.

I only said that if given their limited time and limited knowledge, maybe not
eating the mushrooms is the right thing. If they really need food and mushrooms
seem to be the only food around, then maybe should try to understand what
causes the deaths of people who eat those mushrooms. If it's a farmer with a
gun, maybe we try to find a way to sneak over to the mushrooms at night. If
it's poison, maybe they look for an antidote, or maybe they look for some
different food. Does that count as “accepting the red mushroom taboo?"

>> You didn't answer my question. My question is, how does merely knowing that
>> something is (roughly) 0% heritable or 100% bear on what interventions might
>> be effective?
>
>
> Yes, I did, actually. I said:
>
>> All intervention strategies would fall into one of the environment
>> categories.

That didn't explain to me the difference between the kinds of interventions
that are warranted when all we know is that the trait is roughly 0% vs 100%
heritable. Did you mean that there is no difference between the kinds of
interventions in those two cases?

> In other words, 100% heritability of a trait means no intervention will make
> any difference.

I didn't say exactly 100%, I said roughly 100%, like the heritability for total
number of digits. Does that number have any bearing on what interventions might
be effective?

>> Can you provide a citation on the web for this usage?
>> http://en.wikipedia.org/wiki/Statistics only uses the word "explanation"
>> once, and it's as part of a larger sentence listing all the things
>> statistics is about. Also, is explanation the ONLY term used in statistics
>> for this concept, or do they have an alternate term we could use?
>
>
> See also http://en.wikipedia.org/wiki/Dependent_and_independent_variables,
> where the independent variables are also known as the explanatory variables.

I'm happy to use "explanatory variable", but note that "explanatory variable"
is a different term than "explanation".

> Another term would be "predictor variables". Anyway, "X explains/accounts for
> 60% of the observed variability" is a common idiom in regression analysis.
> I'm not making it up.

I agree with the verb form ("explains"), but I was asking about the noun form
("explanation"). I suggest that in this discussion we use "predictor
variables" for what statistics is talking about, and "explanations" for
explanations in the usual sense. Thoughts?

>> Without explanations of what things could be relevant to the experiment and
>> how they would affect the outcome, how would you know what to control for?
>
>
> Yes, you need a theory - also called a model. As I asserted in my original
> question:
>
>> there is a just a continuum of explanation quality, and contra Deutsch,
>> statistics is the quantification [of] that continuum.

My point is that statistics alone don't tell you anything about how good a
theory is. In to interpret the statistics, you need other relevant ideas about
how the world works.

> To summarize my original position: all theories exist on a continuum of good
> and bad and do not differ in kind from each other.

I don't think there's just one continuum from good to bad. Different theories
have different flaws and there's no single objective way of ranking them all.
In certain specific cases there may be a way of ranking theories
quantitatively, but that doesn't seem to be what you're saying here.

> Statistics is the way that we quantify the relative goodness or badness of a
> theory.

I would say, "Statistics, plus the ways of interpreting them, are *a*
way of quantifying the goodness or badness of a
theory along one particular axis."

Also, there are non-statistical ways to evaluate the relative goodness
or badness
of a theory. For example, the theory that "there exist an infinite
number of primes"
can be evaluated without reference to statistics.

> "Easy to vary" and "reach" are mushy and useless compared to statistical
> interpretation.

They're not quantitative, but they're far from useless...

> The red mushroom example is designed to show that every event in our entire
> lives is cloaked in various degrees of ignorance about root causes,

Yes.

> and we use statistics to distinguish between these relative levels of
> ignorance.

No, we use explanatory theories of how the world works, perhaps together with
some statistics.

> The key of the story is that we have a theory that things are poisonous but
> an operational definition of poison, and that's good enough.

The definition of poison as “eating it causes you to die” isn’t operational
because of the word “cause”. How do we know if x causes y? And if you leave out
the word “cause” and say that poison means “you eat it and then you die”, then
it applies to many of scenarios that don't involve poison such as the examples
I gave above. I all cases, we need an explanation, a theory of how the world
works, to know which interpretations makes sense.

[Blue Yogin]

I should have mentioned that I was preparing a different reply that answers some questions about the theory of heritability analysis directly, since that seems to be a matter of some interest here (for a reason that isn't clear to me just yet).

A few of my questions should require only a yes or no (I think). For your
convenience I repeat those ones here:

Would you agree that, among the US population of the 1950s:
- Wearing earings was highly heritable?

Technically this is also a matter for experiment. In order to find out whether it's heritable you need to isolate the genes from environment. You do this chiefly through studying identical twins who have been adopted into different families - these are called twin adoption studies. Twins who have been adopted have the same genes but different environment. This allows us to isolate the variables and find out how much of a phenotype can be explained by genetics. These twin studies have been done for a few hundred pairs of twins reared apart, the most famous study of which is the so-called Jim Twins study. My guess is that wearing earrings is surprisingly heritable, because most things are surprisingly heritable, but again, that's a matter for experiment.

- Number of digits (fingers + toes) was roughly 0% heritable?

I don't know if dactyly variations are heritable. My guess is that some syndromes are and some aren't. Similar to above, you find this kind of information by finding families with histories of deviant dactyly, or if you get really lucky you find some identical twins with either one or both exhibiting the phenotype. If both twins have the phenotype, they are called concordant and it's likely genetic. If just one twin has the pheno, they are called discordant for that pheno and it's strong evidence against a genetic cause.  (As an interesting aside, identical twins tend to be discordant for homosexuality, (75% of the time), which provides strong evidence that the root cause of homosexuality is not genetic.)  In both of these cases, the question of heritability is a matter for experiment, not philosophy.

Suppose we come up with a quantitative measure for accent. (Along the lines of
http://www.nytimes.com/interactive/2013/12/20/sunday-review/dialect-quiz-map.html?_r=0)

Accent would be highly heritable, right?

Probably not. You would need to do twin studies to find out, though. What you're doing here is failing to appreciate the sophistication which modern statisticians bring to bear on the problem of separating correlation from causation. Yes there are a lot of bullshit "Eating bananas is correlated with diabetes!" type studies published in the science sections of mainstream dailies, but you can't hold that against the researchers. They have to make a living somehow!

It is the tribe's definition of "poisonous" that is "operational" - to use
your terminology. The definition is "something is poisonous if eating it
causes you to die."

There’s a lot hidden behind the word “causes”. All statistics can say is that
eating the berries is correlated with death, right? It’s up to our
interpretations/ideas/theories to make sure the causation is as we think it is.

In this case, this is all that statistics can say. Statistics can also be used - in conjunction with sound experimental design - to tease apart correlation and causation.

Analogously, the definition of intelligence is operational. The definition
is, "how well you do on g-loaded tests".

That's *a* definition of intelligence. It's not the dictionary definition...

There's as much or more theory behind why g-loaded tests predict or explain
life success as there surrounding the tribe's new red mushroom taboo.

Great! Those theories are some of the things we should be talking about.

Well, let's definitely do that, but let's maybe do it in a separate post.  I need some more time to put together a basic case.

The g theory of intelligence is much more well-developed, but you're willing
to accept the red mushroom taboo and not the g theory.

I only said that if given their limited time and limited knowledge, maybe not
eating the mushrooms is the right thing. If they really need food and mushrooms
seem to be the only food around, then maybe should try to understand what
causes the deaths of people who eat those mushrooms. If it's a farmer with a
gun, maybe we try to find a way to sneak over to the mushrooms at night. If
it's poison, maybe they look for an antidote, or maybe they look for some
different food. Does that count as “accepting the red mushroom taboo?"

Why is not eating the mushrooms the right thing? It must be because we've accepted the theory that mushrooms are poisonous as the best we've got right now. 

You didn't answer my question. My question is, how does merely knowing that
something is (roughly) 0% heritable or 100% bear on what interventions might
be effective?

Yes, I did, actually. I said:

All intervention strategies would fall into one of the environment
categories.

That didn't explain to me the difference between the kinds of interventions
that are warranted when all we know is that the trait is roughly 0% vs 100%
heritable. Did you mean that there is no difference between the kinds of
interventions in those two cases?

In other words, 100% heritability of a trait means no intervention will make
any difference.

I didn't say exactly 100%, I said roughly 100%, like the heritability for total
number of digits. Does that number have any bearing on what interventions might
be effective?

The heritability figure gives upper limits on the size of the effect that an intervention can achieve. We say a "heritability of 70%" and mean "70% of the variance in this population can be explained on the basis of genes alone". We also are implying that genes caused 70% of the variance. The point of making a distinction between heritability and environment is to tease out the causal mechanisms of the pheno. This means that if we say 70% heritable - and we're correct -  only 30% of the variance, at best, is amenable to intervention. Roughly 100% heritable would mean roughly 0% of the variance is amenable to intervention.

Can you provide a citation on the web for this usage?
http://en.wikipedia.org/wiki/Statistics only uses the word "explanation"
once, and it's as part of a larger sentence listing all the things
statistics is about.  Also, is explanation the ONLY term used in statistics
for this concept, or do they have an alternate term we could use?

See also http://en.wikipedia.org/wiki/Dependent_and_independent_variables,
where the independent variables are also known as the explanatory variables.

I'm happy to use "explanatory variable", but note that "explanatory variable"
is a different term than "explanation".

Ok, but my original usage was

A statistician is perfectly happy to run a regression analysis and then assert that IQ explains 65±5% of a US citizen's lifetime earning capacity, without having a fully baked idea of why that might be so.

And you seemed to take exception with that as not a normal usage.

Another term would be "predictor variables". Anyway, "X explains/accounts for
60% of the observed variability" is a common idiom in regression analysis.
I'm not making it up.

I agree with the verb form ("explains"), but I was asking about the noun form
("explanation").  I suggest that in this discussion we use "predictor
variables" for what statistics is talking about, and "explanations" for
explanations in the usual sense. Thoughts?

Yes, that will work for me. Can you unfold the difference between explanation per se and prediction? I think that might be fruitful to help us agree on terminology. In statistics they are conflated, but here they seem distinct. This might be a key point.

Math is special. I don't even think you can rightly call them theories. At any event, the epistemology of math is different the epistemology of pretty much everything else - wouldn't you agree?

"Easy to vary" and "reach" are mushy and useless compared to statistical
interpretation.

They're not quantitative, but they're far from useless...

The red mushroom example is designed to show that every event in our entire
lives is cloaked in various degrees of ignorance about root causes,

Yes.

and we use statistics to distinguish between these relative levels of
ignorance.

No, we use explanatory theories of how the world works, perhaps together with
some statistics.

The key of the story is that we have a theory that things are poisonous but
an operational definition of poison, and that's good enough.

The definition of poison as “eating it causes you to die” isn’t operational
because of the word “cause”. How do we know if x causes y? And if you leave out
the word “cause” and say that poison means “you eat it and then you die”, then
it applies to many of scenarios that don't involve poison such as the examples
I gave above. I all cases, we need an explanation, a theory of how the world
works, to know which interpretations makes sense.

I see. This is very helpful to me. I need to think more about this. 

[Alisa Zinov'yevna Rosenbaum]

On Mon, Dec 30, 2013 at 3:57 PM, Blue Yogin <blue....@gmail.com> wrote:
> On Mon, Dec 30, 2013, Alisa Zinov'yevna Rosenbaum <petrogradp...@gmail.com> wrote:

>> On Mon, Dec 30, 2013 at 3:57 PM, Blue Yogin <blue....@gmail.com> wrote:
>> Would you agree that, among the US population of the 1950s:
>> - Wearing earings was highly heritable?
>
> Technically this is also a matter for experiment. In order to find out
> whether it's heritable you need to isolate the genes from environment. You
> do this chiefly through studying identical twins who have been adopted into
> different families - these are called twin adoption studies. Twins who have
> been adopted have the same genes but different environment. This allows us
> to isolate the variables and find out how much of a phenotype can be
> explained by genetics. These twin studies have been done for a few hundred
> pairs of twins reared apart, the most famous study of which is the so-called
> Jim Twins study.

Yes. The ways in which twin studies isolate genes from the environment
are not straightforward or obvious; they involve complicated ideas,
and those ideas may be wrong. Studying the heritability of traits is
humans much more complicated than studying the heritability of traits
in varieties of corn.

> My guess is that wearing earrings is surprisingly
> heritable, because most things are surprisingly heritable, but again, that's
> a matter for experiment.

In the US in the 1950s, almost all women wore earrings, and almost no
men did. So, in that environment and among that population, variations
in the XX/XY chromosome would almost entirely account for the
variation in number of earrings worn, which would result in the trait
of wearing earrings being highly heritable among that population. In
the America of 2013, it's somewhat more common for men to wear
earrings, so I guess that trait would be less heritable for that
population/environment.

> - Number of digits (fingers + toes) was roughly 0% heritable?
>
> I don't know if dactyly variations are heritable. My guess is that some
> syndromes are and some aren't. Similar to above, you find this kind of
> information by finding families with histories of deviant dactyly, or if you
> get really lucky you find some identical twins with either one or both
> exhibiting the phenotype. If both twins have the phenotype, they are called
> concordant and it's likely genetic. If just one twin has the pheno, they are
> called discordant for that pheno and it's strong evidence against a genetic
> cause. (As an interesting aside, identical twins tend to be discordant for
> homosexuality, (75% of the time), which provides strong evidence that the
> root cause of homosexuality is not genetic.)

There are good explanations for why most of the variation in numbers
of fingers and toes are due not to genes but to the environment -
industrial accidents, mothers taking thalidomide during pregnancy,
etc. In light of this, wouldn't it be rather surprising if the number
of digits turned out to be highly heritable?

> In both of these cases, the question of heritability is a matter for experiment, not philosophy.

Doing experiments is a good way of evaluating some theories, but
experiments can be done in the wrong way, or give the wrong result the
first time and need to be repeated, etc. Philosophy helps us choose
which experiments to carry out, design them properly, and interpret
the results.

> Suppose we come up with a quantitative measure for accent. (Along the lines
> of
> http://www.nytimes.com/interactive/2013/12/20/sunday-review/dialect-quiz-map.html?_r=0)
>
> Accent would be highly heritable, right?
>
> Probably not. You would need to do twin studies to find out, though.

Identical twins would have been born at the same time, so they would
learn to talk in a roughly similar manner to the way everyone else in
the country did at that time. They were also born in the same place,
so unless they were reared apart, they would learn to talk from people
in the same general area. If they were reared apart, we would have to
consider that the families that make adoptions are not randomly
distributed - the fathers tend to have higher IQ, for one thing. Also,
maybe white children are more likely to be adopted by white parents,
and this would affect they way those children learned to talk. So it
there are a variety of explanations for why twins, even those raised
apart, would have more similar accents than two randomly selected
members of the population. We can try to control for all the relevant
factors in our experiments, but in order to decide what's relevant in
the first place, we need an explanation for the phenomena we are
studying.

> What
> you're doing here is failing to appreciate the sophistication which modern
> statisticians bring to bear on the problem of separating correlation from
> causation. Yes there are a lot of bullshit "Eating bananas is correlated
> with diabetes!" type studies published in the science sections of mainstream
> dailies, but you can't hold that against the researchers. They have to make
> a living somehow!

We can't control everything that could possibly interfere with our
experiments, and so our experiments are necessarily inadequate.
Philosophy and our theories about the world help us decide which
things to control for and how to interpret the results of our
inadequate experiments.

>>> It is the tribe's definition of "poisonous" that is "operational" - to use
>>> your terminology. The definition is "something is poisonous if eating it
>>> causes you to die."
>
>
>> There’s a lot hidden behind the word “causes”. All statistics can say is
>> that eating the berries is correlated with death, right? It’s up to our
>> interpretations/ideas/theories to make sure the causation is as we think it
>> is.
>
> In this case, this is all that statistics can say. Statistics can also be
> used - in conjunction with sound experimental design - to tease apart
> correlation and causation.

Yes. My point is that explanatory theories are exactly what we need in
order to make a sound experimental design in the first place.

>>> The g theory of intelligence is much more well-developed, but you're willing
>>> to accept the red mushroom taboo and not the g theory.

>> I only said that if given their limited time and limited knowledge, maybe
>> not eating the mushrooms is the right thing. If they really need food and
>> mushrooms
>> seem to be the only food around, then maybe should try to understand what
>> causes the deaths of people who eat those mushrooms. If it's a farmer with a
>> gun, maybe we try to find a way to sneak over to the mushrooms at night. If
>> it's poison, maybe they look for an antidote, or maybe they look for some
>> different food. Does that count as “accepting the red mushroom taboo?"
>
>
> Why is not eating the mushrooms the right thing?

There's a philosophical field that deals with making choices about
what to do. I've heard this field referred to as "morality". One
principle of morality that's relevant here is: if no one has a good
reason why we shouldn't act on a theory, then it's OK to act on it.

> It must be because we've
> accepted the theory that mushrooms are poisonous as the best we've got right
> now.

Although the poisonous mushroom theory is the villagers' best theory,
it's also the ONLY viable theory they have for what killed their
fellows the other day. Coming to this conclusion requires all sorts of
knowledge about how the world works.

As an aside, it's usually the case that we have only one good theory
at a time for any particular phenomenon. Rarely do we have multiple,
competing good theories. Offhand, the only potential exception I can
think of is the case of quantum theory and general relativity. They
are both good theories and yet I am given to understand that they make
different predictions about the world at some scales. However, this
not a great example, because for any particular problem we care about,
we usually know which of the two theories we should use, so even then,
there's only one good theory in any given situation.

>> That didn't explain to me the difference between the kinds of interventions
>> that are warranted when all we know is that the trait is roughly 0% vs 100%
>> heritable. Did you mean that there is no difference between the kinds of
>> interventions in those two cases?
>
>
> In other words, 100% heritability of a trait means no intervention will make
> any difference.
>
>
>> I didn't say exactly 100%, I said roughly 100%, like the heritability for
>> total
>> number of digits. Does that number have any bearing on what interventions
>> might
>> be effective?
>
>
> The heritability figure gives upper limits on the size of the effect that an
> intervention can achieve.

Suppose we want to increase the number of people wearing earrings, and
let's stipulate that among the population of 1950s adult americans,
almost all women wear them and almost no men do, making the wearing of
earrings about 95% heritable. What can we conclude from this about the
upper limit on the size of effect that an intervention could achieve?

> We say a "heritability of 70%" and mean "70% of
> the variance in this population can be explained on the basis of genes
> alone".

I agree that this is a common phrasing, but I quibble with the use of
"explained" there. I would prefer to say that "heritability of 70%"
means "For some trait within a particular fixed environment and
population, the ratio of variance in the relevant genes to variance in
the trait itself was 70%."

> We also are implying that genes caused 70% of the variance.

Isn't it a rather big leap to go from heritability to genetic causes?
Wearing earrings was highly heritable in the 50s, but that doesn't
mean the genes themselves CAUSED the variation. A better explanation
is that the variation was caused by cultural traditions and gender
expectations.

> The point of making a distinction between heritability and environment is to
> tease out the causal mechanisms of the pheno. This means that if we say 70%
> heritable - and we're correct - only 30% of the variance, at best, is
> amenable to intervention. Roughly 100% heritable would mean roughly 0% of
> the variance is amenable to intervention.

I think it is a misconception to say that "roughly 100% heritable

would mean roughly 0% of the variance is amenable to intervention".

For example, the trait of being a slave in the U.S. south was highly
heritable in the mid 1800s, and yet that variance was eminently
amenable to intervention by the North.

To go back to the other examples from above, wearing earrings is
highly heritable, while number of digits should be almost totally
non-heritable. Yet I don't think this tells us anything about how
amenable to intervention each trait is.

> Ok, but my original usage was
>
> A statistician is perfectly happy to run a regression analysis and then
> assert that IQ explains 65±5% of a US citizen's lifetime earning capacity,
> without having a fully baked idea of why that might be so.
>
>
> And you seemed to take exception with that as not a normal usage.

Yes, good point. I didn't realize how common the "X explains Y"
terminology was among statisticians when I complained about it
originally.

>> I suggest that in this discussion we use "predictor
>> variables" for what statistics is talking about, and "explanations" for
>> explanations in the usual sense. Thoughts?
>
>
> Yes, that will work for me. Can you unfold the difference between
> explanation per se and prediction? I think that might be fruitful to help us
> agree on terminology. In statistics they are conflated, but here they seem
> distinct. This might be a key point.

An explanation is a model of the world, an idea about *why* things are
the way the are. It is an idea about what is actually happening in the
world, about the mechanisms that cause things to happen. An example
from The Beginning of Infinity is the axis-tilt explanation of seasons
(http://en.wikipedia.org/wiki/Season#Axis_tilt).

A prediction, on the other hand, is simply an assertion about the
future. For example, one may predict that in the Northern hemisphere
next year, the average temperature will be lower in December than in
July.

>> I would say, "Statistics, plus the ways of interpreting them, are *a*
>> way of quantifying the goodness or badness of a
>> theory along one particular axis."
>>
>> Also, there are non-statistical ways to evaluate the relative goodness
>> or badness
>> of a theory. For example, the theory that "there exist an infinite
>> number of primes"
>> can be evaluated without reference to statistics.
>
>
> Math is special. I don't even think you can rightly call them theories.

When you say that we can't rightly call them theories, by "them" I
guess you mean "mathematical theorems we have proved." I think they
are still theories. First of all, we can be mistaken about them. There
are many examples of incorrect proofs. Also, mathematics has
conjectures, things we think are true but haven't yet proved. For
instance, there are a lot of explanations for why the
http://en.wikipedia.org/wiki/Twin_prime is true, even though it hasn't
been proven. Same for the
http://en.wikipedia.org/wiki/P_versus_NP_problem.

I would like to point out that in neither case would we use statistics
to evaluate how good the theory is.

> At any event, the epistemology of math is different the epistemology of pretty
> much everything else - wouldn't you agree?

Epistemology consists of general ideas about what knowledge is, how it
is created, how to evaluate it, etc. I think these general ideas
apply to all fields. There aren't any exceptions for math. Also, I'm
not sure if there is a good way to separate knowledge of math from
knowledge of physics. (Maybe there is, do you know of one?) For
instance, I can say that it's impossible to arrange 79 pebbles into a
rectangular grid, because 79 is prime. This is a falsifiable theory,
ostensibly about the real world, and yet we wouldn't use statistics to
evaluate it.

[Alisa Zinov'yevna Rosenbaum]

On Tue, Dec 31, 2013 at 5:01 AM, Anon Too <ano...@yahoo.com> wrote:

> On 30/12/2013, at 03:09, Alisa Zinov'yevna Rosenbaum <petrogradp...@gmail.com> wrote:
> Someone asked me:
> > Let's say your goal was to push your country to be better at football?
>

> Is this a good goal for an individual to have?

Why wouldn't it be a good goal? Anyway, the question was a
hypothetical. Blue Yogin wrote:
> We had been talking about the heritability upper and lower bounds for some common phenotypes, chief among them intelligence. [...] Whenever I talk about this I try to switch to sports analogies as quickly as possible. I think we have better intuitions about athletic ability than about mental ability, and they have the same basic statistical properties, so it's easy to switch between them.

[Blue Yogin]

On Dec 30, 2013, at 8:29 PM, Alisa Zinov'yevna Rosenbaum <petrogradp...@gmail.com> wrote:

On Mon, Dec 30, 2013 at 3:57 PM, Blue Yogin <blue....@gmail.com> wrote:

On Mon, Dec 30, 2013, Alisa Zinov'yevna Rosenbaum <petrogradp...@gmail.com> wrote:

On Mon, Dec 30, 2013 at 3:57 PM, Blue Yogin <blue....@gmail.com> wrote:
Would you agree that, among the US population of the 1950s:
- Wearing earings was highly heritable?

Technically this is also a matter for experiment. In order to find out
whether it's heritable you need to isolate the genes from environment. You
do this chiefly through studying identical twins who have been adopted into
different families - these are called twin adoption studies. Twins who have
been adopted have the same genes but different environment. This allows us
to isolate the variables and find out how much of a phenotype can be
explained by genetics. These twin studies have been done for a few hundred
pairs of twins reared apart, the most famous study of which is the so-called
Jim Twins study.

Yes. The ways in which twin studies isolate genes from the environment
are not straightforward or obvious; they involve complicated ideas,
and those ideas may be wrong. Studying the heritability of traits is
humans much more complicated than studying the heritability of traits
in varieties of corn.

It's really the same process. The only complicating factor is that we can't breed humans to satisfy our experimental designs, so we have to spend a lot of money to find these natural experiments.

My guess is that wearing earrings is surprisingly
heritable, because most things are surprisingly heritable, but again, that's
a matter for experiment.

In the US in the 1950s, almost all women wore earrings, and almost no
men did. So, in that environment and among that population, variations
in the XX/XY chromosome would almost entirely account for the
variation in number of earrings worn, which would result in the trait
of wearing earrings being highly heritable among that population. In
the America of 2013, it's somewhat more common for men to wear
earrings, so I guess that trait would be less heritable for that
population/environment.

Heritability, shared environment, and unshared environment are terminology for a theory which relates the differences in observed variation to three possible categories of causal effects. The observed variation of earring wearing habits in the 1950s is different from the observed variation of earring habits in the 2010's. But that doesn't necessarily mean that the observed variation in the 50's is more attributable to genetic effects than the observed variation we see in the 2010's.

- Number of digits (fingers + toes) was roughly 0% heritable?

I don't know if dactyly variations are heritable. My guess is that some
syndromes are and some aren't. Similar to above, you find this kind of
information by finding families with histories of deviant dactyly, or if you
get really lucky you find some identical twins with either one or both
exhibiting the phenotype. If both twins have the phenotype, they are called
concordant and it's likely genetic. If just one twin has the pheno, they are
called discordant for that pheno and it's strong evidence against a genetic
cause.  (As an interesting aside, identical twins tend to be discordant for
homosexuality, (75% of the time), which provides strong evidence that the
root cause of homosexuality is not genetic.)

There are good explanations for why most of the variation in numbers
of fingers and toes are due not to genes but to the environment -
industrial accidents, mothers taking thalidomide during pregnancy,
etc. In light of this, wouldn't it be rather surprising if the number
of digits turned out to be highly heritable?

I'm not following you here. If the root cause of missing your left big toe is a genetic syndrome that runs in your family, then it's going to show up as heritable when you do family histories and twin studies. If the root cause is that a saw chopped your left toe off, then it's not going to show up as heritable. No one is going to propound a theory that lumps all missing big left toes under one root cause. 

I think what you're driving at here is that statistics is not the data equivalent of a meat grinder. You can't just drop data into the top, crank the stats wheel, and get some sausage out of the bottom. Yes, you need a theory to organize the data. I've never intended to claim otherwise, nor do I think I have claimed otherwise. I'm going to keep circling back to my original position: theories exist on a continuum of quality and statistics are how we quantify the goodness of theories. I'm willing to concede that there are other ways of evaluating the goodness of theories, but I am not yet willing to concede that any of the other ways even approach the usefulness of the statistical, quantitative methods of modern science.

In both of these cases, the question of heritability is a matter for experiment, not philosophy.

Doing experiments is a good way of evaluating some theories, but
experiments can be done in the wrong way, or give the wrong result the
first time and need to be repeated, etc. Philosophy helps us choose
which experiments to carry out, design them properly, and interpret
the results.

If "the scientific method" is part of "philosophy" then I agree with this statement.

Suppose we come up with a quantitative measure for accent. (Along the lines
of
http://www.nytimes.com/interactive/2013/12/20/sunday-review/dialect-quiz-map.html?_r=0)

Accent would be highly heritable, right?

Probably not. You would need to do twin studies to find out, though.

Identical twins would have been born at the same time, so they would
learn to talk in a roughly similar manner to the way everyone else in
the country did at that time. They were also born in the same place,
so unless they were reared apart, they would learn to talk from people
in the same general area. If they were reared apart, we would have to
consider that the families that make adoptions are not randomly
distributed - the fathers tend to have higher IQ, for one thing. Also,
maybe white children are more likely to be adopted by white parents,
and this would affect they way those children learned to talk. So it
there are a variety of explanations for why twins, even those raised
apart, would have more similar accents than two randomly selected
members of the population.  We can try to control for all the relevant
factors in our experiments, but in order to decide what's relevant in
the first place, we need an explanation for the phenomena we are
studying.

There's quite a bit you can do to tease out the causation from the correlation.

You can find monozygotic twins raised together and see if they both have the same level of accent. If you get a lot of discordances then this is strong evidence that accents aren't very heritable. You would have to attribute accents to "non-shared environment" in that case - nothing else would match the data. If you get a lot of concordance you admittedly haven't learned much; essentially you've only ruled out "non-shared" environment and are still confounded by the nature/nurture problem. Then you could go to twins reared apart to separate those two confounding factorsBut most of the systematic biases that are introduced by the fact that the kids are all adopted won't matter. 

Maybe the concession that you're attempting to extract is that our theories, which we must develop prior to testing them statistically, give us some clues about which systematic biases are likely to matter. If that's the case, I agree: Yes, you need a theory in order to anticipate likely confounding factors when planning the experiment and crunching the numbers. 

What
you're doing here is failing to appreciate the sophistication which modern
statisticians bring to bear on the problem of separating correlation from
causation. Yes there are a lot of bullshit "Eating bananas is correlated
with diabetes!" type studies published in the science sections of mainstream
dailies, but you can't hold that against the researchers. They have to make
a living somehow!

We can't control everything that could possibly interfere with our
experiments, and so our experiments are necessarily inadequate.
Philosophy and our theories about the world help us decide which
things to control for and how to interpret the results of our
inadequate experiments.

Hmm, yes. Looks like it _was_ the concession you were trying to extract.

It is the tribe's definition of "poisonous" that is "operational" - to use
your terminology. The definition is "something is poisonous if eating it
causes you to die."

There’s a lot hidden behind the word “causes”. All statistics can say is
that eating the berries is correlated with death, right? It’s up to our
interpretations/ideas/theories to make sure the causation is as we think it
is.

In this case, this is all that statistics can say. Statistics can also be
used - in conjunction with sound experimental design - to tease apart
correlation and causation.

Yes. My point is that explanatory theories are exactly what we need in
order to make a sound experimental design in the first place.

Agreed.

Again, this isn't what heritable means. I will agree to your definition of "explanation", but the definition of heritable is well-defined in a formal sense and I'd prefer not to equivocate.

We say a "heritability of 70%" and mean "70% of
the variance in this population can be explained on the basis of genes
alone".

I agree that this is a common phrasing, but I quibble with the use of
"explained" there. I would prefer to say that "heritability of 70%"
means "For some trait within a particular fixed environment and
population, the ratio of variance in the relevant genes to variance in
the trait itself was 70%."

We also are implying that genes caused 70% of the variance.

Isn't it a rather big leap to go from heritability to genetic causes?
Wearing earrings was highly heritable in the 50s, but that doesn't
mean the genes themselves CAUSED the variation. A better explanation
is that the variation was caused by cultural traditions and gender
expectations.

Well, the stats don't tell us that genes caused 70% of the variance. The intertwined theories of evolution and molecular cell biology tell us that genes are the cause of many of the differences we observe in individual specimens.

The point of making a distinction between heritability and environment is to
tease out the causal mechanisms of the pheno. This means that if we say 70%
heritable - and we're correct -  only 30% of the variance, at best, is
amenable to intervention. Roughly 100% heritable would mean roughly 0% of
the variance is amenable to intervention.

I think it is a misconception to say that "roughly 100% heritable
would mean roughly 0% of the variance is amenable to intervention".
For example, the trait of being a slave in the U.S. south was highly
heritable in the mid 1800s, and yet that variance was eminently
amenable to intervention by the North.

Here are two answers: 

1. There's no observed variation in the children of the population of black slaves in the south vis a vis their legal status. Heritability, shared environment, and non-shared environment are how we account for variation within a population. So, the term doesn't really apply here. "Having a body" is not heritable either. The question of nature vs nurture doesn't apply to having a body, since there is no variation. Everyone just has one. We couldn't devise any sort of experiment to determine to what extent having a body is caused by nature or by nurture, since any such experiment has to yield results showing a differential variation when we alter something in one of the categories of explanation while holding the other constant.

2. Being a slave isn't a phenotype. Phenotypes are intrinsic characteristics of an organism, but the legal status of being a slave is an extrinsic status. Extrinsic statuses are not in the domain of things that heritability studies are designed to shed light on. Because of it extrinsic nature, and our understanding of molecular biology, we can rule out the theory that some chemical reaction in our biology causes the condition of slavery. Thus theory of genetics tell us that "being a slave" isn't eligible for analysis qua phenotype.

To go back to the other examples from above, wearing earrings is
highly heritable, while number of digits should be almost totally
non-heritable. Yet I don't think this tells us anything about how
amenable to intervention each trait is

My understanding of the word epistemology is that it is the study of how we know things. The way we know things are true in math is by concluding that they can't possibly false. The way we know things about the physical world is the topic under discussion right now, but I think we'll agree that it isn't by concluding that all alternatives are certainly false. Thus the way we know things about the physical world is not the same as the way we know things about mathematics.

The pebbles example is actually great to demonstrate the difference. If you take 79 pebbles and try to test the theory, you will never arrive at certainty, because the kinds of tests you can do with pebbles never lead to certainty. Meanwhile you could prove it mathematically to a certainty. 

[Elliot Temple]

On Dec 29, 2013, at 7:09 PM, Alisa Zinov'yevna Rosenbaum <petrogradp...@gmail.com> wrote:

> Someone asked me:
>> Let's say your goal was to push your country to be better at football? Would you try to encourage large people or small people to reproduce? Black people or white people? Why?

Why assume the way to approach football improvement is through eugenics or heredity or whatever?

That's not where I would have started at all.

And if you were to get certain groups of people to reproduce more, I'd be more interested in how you do it than which groups they are. Like I'd be wondering how you'd manage that without extreme evil? And how might you find out whether you were making a mistake, and if so how would you undo it?


I also don't see the point. There's plenty of large people to play football. There's way more large people than football players. And there's plenty of black people too. What would we need more of whatever category for? It's not like we needed 2000 football players but only had 500 large black males and ran out and had to let some small asian women join teams.

I also don't even know what the *country* being better at football means. We haven't got a national team, there's various individual teams.

If the goal is to make all of them better at football, maybe inventing the iPad would be a good place to start so they can choose better plays using iPads instead of giant binders of inconvenient paper.

Or you could just go invent some better strategies, then get a job coaching one team, win the Superbowl 5x in a row, then publish your ideas. If you invented substantially better football strategies, you could accomplish that and people would listen and learn.

Or, more indirectly, you could spread TCS. If football is any good, then if there's more rational people solving problems better, some of them will work on football and it will make progress like everything else. this only wouldn't work if football improvement wasn't actually a good goal.

-- Elliot Temple
http://beginningofinfinity.com/

[Elliot Temple]

On Dec 29, 2013, at 7:59 PM, Blue Yogin <blue....@gmail.com> wrote:

> Hi. I am the original asker of this question, and I'm new to the list. Happy to be here. Some of the context was lost here - we had been talking about the heritability upper and lower bounds for some common phenotypes, chief among them intelligence. (Or, if you prefer, "the ability to do well on IQ tests"). Whenever I talk about this I try to switch to sports analogies as quickly as possible. I think we have better intuitions about athletic ability than about mental ability, and they have the same basic statistical properties, so it's easy to switch between them.

So your approach to the subject is to stop considering it, and start arguing by analogy, as quickly as possible?

> So, that's the context.
>
> To make this question harder and to avoid sneaking away from the pain I'm trying to inflict, allow me to add another proviso. Suppose that any good idea in football can be quickly copied by other teams.

How will supposing something vague, and seemingly false, get us anywhere? And doing it regarding an analogy rather than the actual topic? When it sounds way more false on the actual topic? (Good ideas often are not quickly copied by other people.)

> Thus, though individual coaches might have some good ideas about how to train better or play better, whenever a good idea like that is shown to be successful, it quickly sweeps the entire league, and the game returns to an equilibrium state in which every team employs the best training and playing strategies, so that the only difference between teams are physical differences that can't be easily copied.

So you only want to approach the subject via *unrealistic* analogy?

>
> And let me propose another thought experiment designed to evoke the same sort of question, but which makes, I feel, more sense absent the original context. The point of the question is to force you to make a decision without the benefit of a satisfying "explanatory theory" beyond a statistical likelihood calculation.

So now the thought experiment involves attacking epistemology, too?

>
> To me, statistics is in the business of explanation - whatever that means.

Might learning what explanation means shed light on intelligence? And help differentiate good and bad explanations of how intelligence works? Might that be a better approach than telling stories about football as a way of telling stories about intelligence?

> A statistician is perfectly happy to run a regression analysis and then assert that IQ explains 65±5% of a US citizen's lifetime earning capacity, without having a fully baked idea of why that might be so.

Not a good one. For example, you haven't mentioned the possibility that X causes both IQ and high earnings. Correlation isn't causation. All good statisticians would know that.

The topic is supposed to be intelligence but so far you've wanted to make mistakes about football and statistics, while dismissing explanation without understanding it (even though it is a key aspect of intelligence, or at least some people think so, and I don't see how you could reasonably reject that without knowing about it).

> It seems from your usage that you, with David Deutsch, assign a different meaning to the term explanation. The point I am driving at is that neither of you seem to have given a bright line past which something can be said to be explained.

Right, there's always improvement possible. There's never a time that we're done thinking forever. It's never over. We're always at the beginning of infinity.

> It seems to me that there is a just a continuum of explanation quality, and contra Deutsch, statistics is the quantification that continuum. I need to understand how an explanation that you consider "good" differs in kind from a statistical explanation. Deutsch has a little bit of this with "hard to vary" and "reach". Let me try to defeat these right now.

Explanations tell you about *whys*. Even if some statistician is correct that X causes Y, he hasn't yet explained *how or why* it does so.

>
> Let's say that twenty of us live in a village. We find some red mushrooms and some blue mushrooms in the forest near our village. We've never seen either kind before. Ten of us eat the red mushrooms, and ten of us eat the blue mushrooms. A day later, we find that four people who ate the red mushrooms have died in their beds. Everyone who ate blue mushrooms feels fine.
>
> We don't have the benefit of gas chromatography, and can't do an assay to find out the various exotic compounds in the mushrooms. We don't have the benefit of forensic medicine that would allow us to determine the cause of death for the red mushroom eaters. But we can run a p-test, so let's do that.
>
> The null hypothesis is that neither red nor blue mushrooms affected anybody's physiology. The deaths occurred for some other reason. We assume for our test that the deaths were randomly distributed throughout the population of 20.
>
> This is ye old balls and urns. The number of distinguishable ways that all four deaths could have fallen into the red camp is 10 chose 4.
>
> The number of distinguishable ways that the deaths could have been distributed is 20 chose 4.
>
> Dividing, we find that the probability of only reds having fallen, assuming all configurations are equality likely, is 4.33e-2, significant to within the p > .05 level. (And someone please chip in if I've got this math wrong, but even if I have, I think the point stands. You can change the numbers to make this scenario work).
>
> So, we have statistical significance, but no real theory - right?

Wrong. There are theories involved such as that foods can cause overnight death in bed hours later. Rather than consider the theories involved you've assumed them. That's a bad approach in general because some of your assumptions might be wrong sometimes. It's better to understand what theories you're using instead of use them unconsciously and uncritically.

> Maybe it's something else. We don't have a mechanism. Maybe we should keep eating red mushrooms. Should we, or should we not? And how are we to make that decision? We have to eat something. We have to eat something today. We can't wait until the invention of gas chromatography and forensic toxicology, right?

This issue, like all issues, requires knowledge to make a good decision about. Knowledge is created by conjectures and refutations. No other approach is known. It's reasonably straightforward how the method of conjectures and refutations could be used here. When people make decisions seemingly without it, they are using it inexplicitly which makes it harder to catch mistakes or organize the process.

You can consider several conjectures like:

- we should eat only the blue mushrooms
- we should eat all the mushrooms

and you can consider criticisms of each one. and you can evaluate the issue until you reach a tentative conclusion – meaning you have exactly one non-refuted conjecture. then use that.

>
> "Eating red mushrooms might kill you" is a theory of reality. It is hard to vary: if we changed it to blue mushrooms, it wouldn't fit the facts.

sure it would fit the facts. it could easily be true that blue mushrooms might kill you but no one who ate them died yet. there are millions of possible scenarios in which the blue mushrooms are deadly and the story begins as told above. if there is anything wrong with those scenarios, it cannot be the facts because they don't contradict any of the facts under consideration.

this is a good example of how people with bad epistemologies routinely say silly things. perhaps you can find that motivating to learn a different epistemology.

> If we changed it to "toads", it wouldn't fit the facts: it wouldn't explain why only the red-mushroom-eaters died.

Fitting the facts means not contradicting the facts, right? But you aren't using it that way.

If I said the house over there is green, and it is green, you wouldn't tell me my statement didn't fit the facts because it didn't explain why red-mushroom-eaters died.


-- Elliot Temple
http://curi.us/

[Elliot Temple]

In other words, "heritability" doesn't mean people inherit it from their parents. But the word is used anyway, in order to mislead people. This is dishonest. Stop doing it if you want to make progress.

Also a lot is already known about these topics, like this:

http://vserver1.cscs.lsa.umich.edu/~crshalizi/weblog/520.html

Do you have any criticisms of existing knowledge in the field like this? Any suggested improvements? What problem(s) are you trying to solve by entering the field? What are you trying to add or subtract from the field? It's important to relate your contribution to the existing field instead of just start from scratch and tell some stories.

>
>>
>> Second, a note on the meaning of intelligence. I don’t consider the ability to score highly on IQ tests to be equivalent to intelligence. IQ tests and the Turing test are both operational definitions that don’t explain the underlying phenomena. There’s already a convenient word for the ability to score highly on IQ tests, so, for clarity, let’s use it (“IQ”) rather than "intelligence", if that is indeed what you mean.
>
> This is the crux of my trap, which I was waiting to spring

If you want to trap people you should go to a different list. This one is for cooperative learning, not adversarial trapping.

> until you agreed that we shouldn't eat the mushrooms: If we can say that mushrooms are poisonous without understanding how the poison they allegedly contain disrupts our normal functioning, why can't we say that high-IQ-scorers are more intelligent without understanding how intelligence makes us better at IQ tests?

you could say whatever you want, but what would be your point? what problem(s) are you trying to address and how does this address them?


-- Elliot Temple
http://elliottemple.com/

[Elliot Temple]

On Dec 30, 2013, at 1:36 PM, Mason Kramer <blue....@gmail.com> wrote:

> 100% heritability of a trait means no intervention will make any difference.

But:

http://vserver1.cscs.lsa.umich.edu/~crshalizi/weblog/520.html

> To summarize: Heritability is a technical measure of how much of the variance in a quantitative trait (such as IQ) is associated with genetic differences, in a population with a certain distribution of genotypes and environments. Under some very strong simplifying assumptions, quantitative geneticists use it to calculate the changes to be expected from artificial or natural selection in a statistically steady environment. It says nothing about how much the over-all level of the trait is under genetic control, and it says nothing about how much the trait can change under environmental interventions.

[Elliot Temple]

On Dec 30, 2013, at 3:57 PM, Blue Yogin <blue....@gmail.com> wrote:

> What you're doing here is failing to appreciate the sophistication which modern statisticians bring to bear on the problem of separating correlation from causation.

You mean the authorities on your side have bigger dicks? Or what?

[Elliot Temple]

On Dec 30, 2013, at 3:57 PM, Blue Yogin <blue....@gmail.com> wrote:

> The heritability figure gives upper limits on the size of the effect that an intervention can achieve. We say a "heritability of 70%" and mean "70% of the variance in this population can be explained on the basis of genes alone". We also are implying that genes caused 70% of the variance. The point of making a distinction between heritability and environment is to tease out the causal mechanisms of the pheno. This means that if we say 70% heritable - and we're correct - only 30% of the variance, at best, is amenable to intervention. Roughly 100% heritable would mean roughly 0% of the variance is amenable to intervention.

Why should I accept these claims?

-- Elliot Temple
http://fallibleideas.com/

[Elliot Temple]

On Dec 30, 2013, at 3:57 PM, Blue Yogin <blue....@gmail.com> wrote:

> Math is special. I don't even think you can rightly call them theories. At any event, the epistemology of math is different the epistemology of pretty much everything else - wouldn't you agree?

No. The Fabric of Reality has a chapter about this.

[Elliot Temple]

On Dec 31, 2013, at 12:42 PM, Alisa Zinov'yevna Rosenbaum <petrogradp...@gmail.com> wrote:

> On Tue, Dec 31, 2013 at 5:01 AM, Anon Too <ano...@yahoo.com> wrote:
>> On 30/12/2013, at 03:09, Alisa Zinov'yevna Rosenbaum <petrogradp...@gmail.com> wrote:
>> Someone asked me:
>>> Let's say your goal was to push your country to be better at football?
>>
>> Is this a good goal for an individual to have?
>
> Why wouldn't it be a good goal?

It's nationalistic. Can you think of any other issues with it?

[Elliot Temple]

On Jan 1, 2014, at 9:03 AM, Kermit Rose <ker...@polaris.net> wrote:

> Intelligence is the ability to use information to solve problems.
>
> Wisdom is knowing which problems it is useful to solve.

What problem(s) are you trying to solve with these statements?

What problem(s) do you think the other people in the discussion were trying to solve?

Are your answers to my two questions the same? Do they overlap? Are they different? Does it matter?

[Blue Yogin]

On Jan 17, 2014, at 7:47 AM, Elliot Temple <cu...@curi.us> wrote:

> On Dec 29, 2013, at 7:59 PM, Blue Yogin <blue....@gmail.com> wrote:
>
>> Hi. I am the original asker of this question, and I'm new to the list. Happy to be here. Some of the context was lost here - we had been talking about the heritability upper and lower bounds for some common phenotypes, chief among them intelligence. (Or, if you prefer, "the ability to do well on IQ tests"). Whenever I talk about this I try to switch to sports analogies as quickly as possible. I think we have better intuitions about athletic ability than about mental ability, and they have the same basic statistical properties, so it's easy to switch between them.
>
> So your approach to the subject is to stop considering it, and start arguing by analogy, as quickly as possible?

Yes.

>> To make this question harder and to avoid sneaking away from the pain I'm trying to inflict, allow me to add another proviso. Suppose that any good idea in football can be quickly copied by other teams.
>
> How will supposing something vague, and seemingly false, get us anywhere?
>

>> Thus, though individual coaches might have some good ideas about how to train better or play better, whenever a good idea like that is shown to be successful, it quickly sweeps the entire league, and the game returns to an equilibrium state in which every team employs the best training and playing strategies, so that the only difference between teams are physical differences that can't be easily copied.
>
> So you only want to approach the subject via *unrealistic* analogy?

It's not unrealistic. Strategies in football are in actual fact routinely copied whenever they work. Major league teams employ scouts whose full-time job is to figure out what training routines and play strategies the other teams are practicing. Trainers and coaches routinely switch from team to team and cross-pollinate their institutional knowledge. Whenever a team loses to a significantly new strategy, everyone in the entire league studies that play on their iPads in the week after the game and either adopts or develops a counter strategy.

>>
>> And let me propose another thought experiment designed to evoke the same sort of question, but which makes, I feel, more sense absent the original context. The point of the question is to force you to make a decision without the benefit of a satisfying "explanatory theory" beyond a statistical likelihood calculation.
>
> So now the thought experiment involves attacking epistemology, too?

Let's say that I'm trying to elucidate, through dialog, where my ideas of how we know things differ from your ideas of how we know things. I am always happy to improve my model.

>> To me, statistics is in the business of explanation - whatever that means.
>
> Might learning what explanation means shed light on intelligence? And help differentiate good and bad explanations of how intelligence works? Might that be a better approach than telling stories about football as a way of telling stories about intelligence?

I have a well developed sense of the word explanation. My "whatever that means" here was only to call attention to the fact that, in this dialog, we haven't yet agreed on the meaning of the term. I should have been more clear. My intention is to figure out what is insufficient about the theory of intelligence - what objects are made to it on epistemological grounds - such that FI thinkers feel justified in discarding it. After that's been determined I hope to either join you in discarding the theory, or else to amend the presentation so that you'll be willing to take it seriously.

>> A statistician is perfectly happy to run a regression analysis and then assert that IQ explains 65±5% of a US citizen's lifetime earning capacity, without having a fully baked idea of why that might be so.
>
> Not a good one. For example, you haven't mentioned the possibility that X causes both IQ and high earnings. Correlation isn't causation. All good statisticians would know that.
>
> The topic is supposed to be intelligence but so far you've wanted to make mistakes about football and statistics,

Epistemology is the topic.

> while dismissing explanation without understanding it (even though it is a key aspect of intelligence, or at least some people think so, and I don't see how you could reasonably reject that without knowing about it).

Since my entire letter was concerned with the meaning of explanation, I think it's uncharitable to claim that I've dismissed explanation without understanding it.

>
>> It seems from your usage that you, with David Deutsch, assign a different meaning to the term explanation. The point I am driving at is that neither of you seem to have given a bright line past which something can be said to be explained.
>
> Right, there's always improvement possible. There's never a time that we're done thinking forever. It's never over. We're always at the beginning of infinity.
>
>> It seems to me that there is a just a continuum of explanation quality, and contra Deutsch, statistics is the quantification that continuum. I need to understand how an explanation that you consider "good" differs in kind from a statistical explanation. Deutsch has a little bit of this with "hard to vary" and "reach". Let me try to defeat these right now.
>
> Explanations tell you about *whys*. Even if some statistician is correct that X causes Y, he hasn't yet explained *how or why* it does so.

If X causes Y, then the answer to the question, "why did Y happen?" is "X caused Y", or, propositionally, "X causes Y, and X happened".

>> Maybe it's something else. We don't have a mechanism. Maybe we should keep eating red mushrooms. Should we, or should we not? And how are we to make that decision? We have to eat something. We have to eat something today. We can't wait until the invention of gas chromatography and forensic toxicology, right?
>
> This issue, like all issues, requires knowledge to make a good decision about. Knowledge is created by conjectures and refutations. No other approach is known. It's reasonably straightforward how the method of conjectures and refutations could be used here. When people make decisions seemingly without it, they are using it inexplicitly which makes it harder to catch mistakes or organize the process.
>
> You can consider several conjectures like:
>
> - we should eat only the blue mushrooms
> - we should eat all the mushrooms
>
> and you can consider criticisms of each one. and you can evaluate the issue until you reach a tentative conclusion – meaning you have exactly one non-refuted conjecture. then use that.
>

How about this theory: the red mushrooms are poisonous, and the blue mushrooms aren't.

The main point of the story though is "poisonous". What is poisonous? The villagers don't really know. They don't have a mechanism of action. Poison for them could be "bad juju". But "bad juju" is enough of an explanation for them to reach a state of actionable knowledge. The concession I'm trying to extract is that "theories which suggest that they shouldn't eat the red mushrooms - because

>>
>> "Eating red mushrooms might kill you" is a theory of reality. It is hard to vary: if we changed it to blue mushrooms, it wouldn't fit the facts.
>
> sure it would fit the facts. it could easily be true that blue mushrooms might kill you but no one who ate them died yet. there are millions of possible scenarios in which the blue mushrooms are deadly and the story begins as told above. if there is anything wrong with those scenarios, it cannot be the facts because they don't contradict any of the facts under consideration.
>
> this is a good example of how people with bad epistemologies routinely say silly things. perhaps you can find that motivating to learn a different epistemology.

If I apply the amended theory, this works: "eating red mushrooms might kill you, but eating blue mushrooms is safe" cannot be switched to "eating blue mushrooms might kill you, but eating red mushrooms is safe".

[Elliot Temple]

On Jan 17, 2014, at 7:07 AM, Blue Yogin <blue....@gmail.com> wrote:

> On Jan 17, 2014, at 7:47 AM, Elliot Temple <cu...@curi.us> wrote:
>
>> On Dec 29, 2013, at 7:59 PM, Blue Yogin <blue....@gmail.com> wrote:
>>
>>> Hi. I am the original asker of this question, and I'm new to the list. Happy to be here. Some of the context was lost here - we had been talking about the heritability upper and lower bounds for some common phenotypes, chief among them intelligence. (Or, if you prefer, "the ability to do well on IQ tests"). Whenever I talk about this I try to switch to sports analogies as quickly as possible. I think we have better intuitions about athletic ability than about mental ability, and they have the same basic statistical properties, so it's easy to switch between them.
>>
>> So your approach to the subject is to stop considering it, and start arguing by analogy, as quickly as possible?
>
> Yes.

Isn't that really bad? Arguing about something by analogy is indirect and adds a lot of potential sources of errors compared with the direct approach.

>
>>> To make this question harder and to avoid sneaking away from the pain I'm trying to inflict, allow me to add another proviso. Suppose that any good idea in football can be quickly copied by other teams.
>>
>> How will supposing something vague, and seemingly false, get us anywhere?
>>
>>> Thus, though individual coaches might have some good ideas about how to train better or play better, whenever a good idea like that is shown to be successful, it quickly sweeps the entire league, and the game returns to an equilibrium state in which every team employs the best training and playing strategies, so that the only difference between teams are physical differences that can't be easily copied.
>>
>> So you only want to approach the subject via *unrealistic* analogy?
>
> It's not unrealistic. Strategies in football are in actual fact routinely copied whenever they work. Major league teams employ scouts whose full-time job is to figure out what training routines and play strategies the other teams are practicing. Trainers and coaches routinely switch from team to team and cross-pollinate their institutional knowledge. Whenever a team loses to a significantly new strategy, everyone in the entire league studies that play on their iPads in the week after the game and either adopts or develops a counter strategy.

That people put a lot of effort into trying to do something is no argument that they are always fully successful.

>
>>>
>>> And let me propose another thought experiment designed to evoke the same sort of question, but which makes, I feel, more sense absent the original context. The point of the question is to force you to make a decision without the benefit of a satisfying "explanatory theory" beyond a statistical likelihood calculation.
>>
>> So now the thought experiment involves attacking epistemology, too?
>
> Let's say that I'm trying to elucidate, through dialog, where my ideas of how we know things differ from your ideas of how we know things. I am always happy to improve my model.

Is that a yes, a no, or what?

If you wanted to communicate about your epistemology views you could just post criticism of some of our epistemology literature, or ask for criticism of some of yours that you don't know any criticism of. I think that'd be more effective.

>
>>> To me, statistics is in the business of explanation - whatever that means.
>>
>> Might learning what explanation means shed light on intelligence? And help differentiate good and bad explanations of how intelligence works? Might that be a better approach than telling stories about football as a way of telling stories about intelligence?
>
> I have a well developed sense of the word explanation. My "whatever that means" here was only to call attention to the fact that, in this dialog, we haven't yet agreed on the meaning of the term. I should have been more clear. My intention is to figure out what is insufficient about the theory of intelligence - what objects are made to it on epistemological grounds - such that FI thinkers feel justified in discarding it. After that's been determined I hope to either join you in discarding the theory, or else to amend the presentation so that you'll be willing to take it seriously.

to start with, we don't feel justified. justification is part of your epistemology, not ours. it might help if you read some FI literature and got a rough understanding of our epistemology.

>
>>> A statistician is perfectly happy to run a regression analysis and then assert that IQ explains 65±5% of a US citizen's lifetime earning capacity, without having a fully baked idea of why that might be so.
>>
>> Not a good one. For example, you haven't mentioned the possibility that X causes both IQ and high earnings. Correlation isn't causation. All good statisticians would know that.
>>
>> The topic is supposed to be intelligence but so far you've wanted to make mistakes about football and statistics,
>
> Epistemology is the topic.
>
>> while dismissing explanation without understanding it (even though it is a key aspect of intelligence, or at least some people think so, and I don't see how you could reasonably reject that without knowing about it).
>
> Since my entire letter was concerned with the meaning of explanation, I think it's uncharitable to claim that I've dismissed explanation without understanding it.
>
>>
>>> It seems from your usage that you, with David Deutsch, assign a different meaning to the term explanation. The point I am driving at is that neither of you seem to have given a bright line past which something can be said to be explained.
>>
>> Right, there's always improvement possible. There's never a time that we're done thinking forever. It's never over. We're always at the beginning of infinity.
>>
>>> It seems to me that there is a just a continuum of explanation quality, and contra Deutsch, statistics is the quantification that continuum. I need to understand how an explanation that you consider "good" differs in kind from a statistical explanation. Deutsch has a little bit of this with "hard to vary" and "reach". Let me try to defeat these right now.
>>
>> Explanations tell you about *whys*. Even if some statistician is correct that X causes Y, he hasn't yet explained *how or why* it does so.
>
> If X causes Y, then the answer to the question, "why did Y happen?" is "X caused Y", or, propositionally, "X causes Y, and X happened".

but there's still the issue of how or why X causes Y.

just saying that the earth's axis' tilt causes the season summer doesn't explain how or why. there are actual good explanations of that which are valuable.

>
>>> Maybe it's something else. We don't have a mechanism. Maybe we should keep eating red mushrooms. Should we, or should we not? And how are we to make that decision? We have to eat something. We have to eat something today. We can't wait until the invention of gas chromatography and forensic toxicology, right?
>>
>> This issue, like all issues, requires knowledge to make a good decision about. Knowledge is created by conjectures and refutations. No other approach is known. It's reasonably straightforward how the method of conjectures and refutations could be used here. When people make decisions seemingly without it, they are using it inexplicitly which makes it harder to catch mistakes or organize the process.
>>
>> You can consider several conjectures like:
>>
>> - we should eat only the blue mushrooms
>> - we should eat all the mushrooms
>>
>> and you can consider criticisms of each one. and you can evaluate the issue until you reach a tentative conclusion – meaning you have exactly one non-refuted conjecture. then use that.
>>
>
> How about this theory: the red mushrooms are poisonous, and the blue mushrooms aren't.
>
> The main point of the story though is "poisonous". What is poisonous? The villagers don't really know. They don't have a mechanism of action. Poison for them could be "bad juju". But "bad juju" is enough of an explanation for them to reach a state of actionable knowledge. The concession I'm trying to extract is that "theories which suggest that they shouldn't eat the red mushrooms - because

your paragraph ends abruptly

>
>>>
>>> "Eating red mushrooms might kill you" is a theory of reality. It is hard to vary: if we changed it to blue mushrooms, it wouldn't fit the facts.
>>
>> sure it would fit the facts. it could easily be true that blue mushrooms might kill you but no one who ate them died yet. there are millions of possible scenarios in which the blue mushrooms are deadly and the story begins as told above. if there is anything wrong with those scenarios, it cannot be the facts because they don't contradict any of the facts under consideration.
>>
>> this is a good example of how people with bad epistemologies routinely say silly things. perhaps you can find that motivating to learn a different epistemology.
>
> If I apply the amended theory, this works: "eating red mushrooms might kill you, but eating blue mushrooms is safe" cannot be switched to "eating blue mushrooms might kill you, but eating red mushrooms is safe".

sure it can be switched without contradicting the facts. it could be that everyone who ate blue mushrooms will be dead within a week, but the people who ate red mushrooms and died also ate something else and weren't killed by the red mushrooms. if there's anything wrong with this view, it's not how well it fits the facts, because it agrees with all the facts we stipulated.

[Alisa Zinov'yevna Rosenbaum]

What's wrong with having a nationalistic goal? I am nationalistic towards the USA: I think it is better than all other countries.

[Elliot Temple]

if the concept was the push the best country to be better at football, whichever the country is, the sentence would have said that. instead it was differentiating by which country is "your[s]".

https://www.google.com/search?client=safari&rls=en&q=what's+wrong+with+nationalism&ie=UTF-8&oe=UTF-8