That paragraph you quoted doesn’t sound smart to me. It seems like it’s argues against a strawman. Scientists who studies issues like this usually don’t publish raw correlations but try to control for various factors they can think of.
Of course you can still criticise that scientists failed to control for relevant factors but that means you actually have to read the papers.
You can also make general arguments against the usefulness of regression analysis but Scott doesn’t make those in that article.
I think Scott doesn’t argue against scientific papers in particular or in general. I think he is raising the sanity waterline. Increase the awareness for these things in general. The specialized scientists may be aware of thse—and probably joke about them. But I was surprised a bit. I could have come up with that—but I didn’t. Did you?
If you read “Married people live longer” what does that sentence mean?
The people on the street think it means (A):
“We measured the the lifespan of married people and we measured the lifespan of unmarried people. It turns out that the lifespan value we measured for the people who are married is higher.”
The problem is that’s not what it means. It rather means (B):
“We measured a bunch of factors among them lifespan and whether people are married. Then we did run a regression analysis and found that being marriages influences lifespan in a positive way.”
Knowing that the sentence means (B) is statistical literacy. Literacy that Scott isn’t showing when he assumes that a common factor like income isn’t factored out of the question of whether moderate drinking is healthy.
The specialized scientists may be aware of thse—and probably joke about them.
Why do you think they joke about them instead of fixing the issue by controlling for the factors they can think of?
I could have come up with that—but I didn’t. Did you?
I’m certainly able to not take the conclusions of observational studies as strong evidence.
The problem is that’s not what it means. It rather means (B): “We measured a bunch of factors among them lifespan and whether people are married. Then we did run a regression analysis and found that being marriages influences lifespan in a positive way.”
“Measuring a bunch of factors etc.” is an observational investigation; “being married influences lifespan” is a causal statement. The former absolutely does not mean the latter, although given additional causal information or assumptions you may be able to deduce it from the experiment. Merely controlling for common factors does not fix this.
Do you have suggestions for another verb to replace “influence” in that sentence?
“Is positively associated with.” “Tend to be found together with.” “Correlates with.”
Have statisticians who do not understand causation and philosophers who do not believe in it corrupted the language so much as to make “influence”, a purely causal concept in everyday language, be a synonym of “association”?
They do mean the right thing, though. And “tend to be found together with”? Everyday words, all of them, put together in an everyday way. Perhaps it is the concept that is not an everyday one. It needs to be.
That paragraph doesn’t sound smart to me. It seems like it’s argues against a strawman. Scientists who studies issues like this usually don’t publish raw correlations but try to control for various factors they can think of.
Of course you can still criticise that scientists failed to control for relevant factors but that means you actually have to read the papers.
You can also make general arguments against the usefulness of regression analysis but Scott doesn’t make those in that article.
[Link] The Health Advice Scott Adams Does’t Find Credible.
It’s a nice piece on correlation vs. causation e.g.
And so he goes on with dog owners, light drinkers and first of all exercise. Maybe before you go on to read it you wonder what the correlation may be.
Maybe also consider whether this applies to other Lifestyle interventions to increase longevity.
That paragraph you quoted doesn’t sound smart to me. It seems like it’s argues against a strawman. Scientists who studies issues like this usually don’t publish raw correlations but try to control for various factors they can think of.
Of course you can still criticise that scientists failed to control for relevant factors but that means you actually have to read the papers.
You can also make general arguments against the usefulness of regression analysis but Scott doesn’t make those in that article.
I think Scott doesn’t argue against scientific papers in particular or in general. I think he is raising the sanity waterline. Increase the awareness for these things in general. The specialized scientists may be aware of thse—and probably joke about them. But I was surprised a bit. I could have come up with that—but I didn’t. Did you?
If you read “Married people live longer” what does that sentence mean?
The people on the street think it means (A): “We measured the the lifespan of married people and we measured the lifespan of unmarried people. It turns out that the lifespan value we measured for the people who are married is higher.”
The problem is that’s not what it means. It rather means (B): “We measured a bunch of factors among them lifespan and whether people are married. Then we did run a regression analysis and found that being marriages influences lifespan in a positive way.”
Knowing that the sentence means (B) is statistical literacy. Literacy that Scott isn’t showing when he assumes that a common factor like income isn’t factored out of the question of whether moderate drinking is healthy.
Why do you think they joke about them instead of fixing the issue by controlling for the factors they can think of?
I’m certainly able to not take the conclusions of observational studies as strong evidence.
“Measuring a bunch of factors etc.” is an observational investigation; “being married influences lifespan” is a causal statement. The former absolutely does not mean the latter, although given additional causal information or assumptions you may be able to deduce it from the experiment. Merely controlling for common factors does not fix this.
I didn’t want to imply that there’s a causal link. Do you have suggestions for another verb to replace “influence” in that sentence?
“Is positively associated with.” “Tend to be found together with.” “Correlates with.”
Have statisticians who do not understand causation and philosophers who do not believe in it corrupted the language so much as to make “influence”, a purely causal concept in everyday language, be a synonym of “association”?
To me those options don’t feel like they are everyday language.
They do mean the right thing, though. And “tend to be found together with”? Everyday words, all of them, put together in an everyday way. Perhaps it is the concept that is not an everyday one. It needs to be.
That paragraph doesn’t sound smart to me. It seems like it’s argues against a strawman. Scientists who studies issues like this usually don’t publish raw correlations but try to control for various factors they can think of.
Of course you can still criticise that scientists failed to control for relevant factors but that means you actually have to read the papers.
You can also make general arguments against the usefulness of regression analysis but Scott doesn’t make those in that article.