3 ms·

The only problem is that his methodology is no better than lining up time-series data and finding spurious correlations[1]. If we did this with any other proble

The only problem is that his methodology is no better than lining up time-series data and finding spurious correlations[1]. If we did this with any other problem, we would be happy to point out the problem with this approach. Nutritional studies have more scientific rigor than what Haidt is hocking and we're happy to point out standards of evidence. All I'm asking is that we don't alter our standard of evidence simply because we agree with the conclusion. It's bad science and a bad way of making decisions. RCT advocates should have easy pickings here. 1. http://www.tylervigen.com/spurious-correlations http://www.tylervigen.com/spurious-correlations

navane 3y ago
It's risk times consequences. The risk assessment might not be up to scientific standards, but the consequences are big. Also, it's not just about statistics, in this case we have a pretty intuitive grasp on the mechanics at play, they are not far fetched.
- kelseyfrog 3y ago
  Risks have to be based in reality. Our greatest reality check is the scientific process. We can make up risks endlessly, see the AI-as-an-existential-threat debate. It simply is not grounded in reality. All I'm asking is transparency. People make non-scientific claims and conclusions all the time. Folks just need to be honest that this is what they're doing so they aren't fooling themselves and others.
hackerlight 3y ago
Haidt's rebuttal to your argument of spurious correlation: https://www.afterbabel.com/p/phone-based-childhood-cause-epidemic https://www.afterbabel.com/p/phone-based-childhood-cause-epi...
- kelseyfrog 3y ago
  > Alcott and colleagues (2020) randomly assigned 2743 adults. I stopped there. We're talking specifically about teenage depression and the cited study is adults. If Haidt wants us to take results from one subgroup and apply it to another, he has to provide evidence. That's how this process works.
  - neglesaks 3y ago
    You conveniently stopped reading at the first cite, before he provided more experimental evidence.
    - kranke155 3y ago
      Because he has no intention of being convinced by data that goes against his beliefs, which is an ego based anti scientific stance.
      - kelseyfrog 3y ago
        I'm much more interested in the rhetorical device, "I stopped reading at X and couldn't take anything after that seriously." It looks like I have some more learning to do before employing it successfully.
      - cap1434 3y ago
        Project much?
        kranke155 3y ago
        how did I project? Where do you see me doing the same thing?
  - hackerlight 3y ago
    This is weirdly uncharitable. Why would you stop there when there's 22 experimental studies[1] and the majority of the article left to read? [1] https://docs.google.com/document/d/1w-HOfseF2wF9YIpXwUUtP65-olnkPyWcgF5BiAtBEy0/edit https://docs.google.com/document/d/1w-HOfseF2wF9YIpXwUUtP65-...
  - tomrod 3y ago
    > If Haidt wants us to take results from one subgroup and apply it to another, he has to provide evidence. That's how this process works. Perhaps considering where he does so, further in the article which you stopped reading.
nabla9 3y ago
That's not true. 1. Haid has evidence from randomized experiments with kids. 2. There are natural experiments caused by different regional adoption rates of high-speed internet. 3. When the effect size of harm correlates with the frequency of use in population, that's more than time series and spurious-correlations. Social Media and Mental Health: A Collaborative Review https://docs.google.com/document/d/1w-HOfseF2wF9YIpXwUUtP65-olnkPyWcgF5BiAtBEy0/edit https://docs.google.com/document/d/1w-HOfseF2wF9YIpXwUUtP65-...
- kelseyfrog 3y ago
  It's not possible to blind the experiment's test subjects. Again, all I'm asking for is the level of critique to be comparable to the strength of the claims. We are eager to critique lower stakes claims, but for some reason Haidt typically gets a pass here. All I'm asking for is consistency. I'm not sure how to put this, but effect size and causation are different things. It's a category mistake to compare.
  - nabla9 3y ago
    > It's not possible to blind the experiment's test subjects. It's not necessary. Randomized experiment is good enough. So are natural experiments.
- timmytokyo 3y ago
  Mike Masnick has written a review of Haidt's book, and it's a nice rebuttal to each of these arguments. https://www.thedailybeast.com/the-coddling-of-the-american-parent https://www.thedailybeast.com/the-coddling-of-the-american-p... One of the points he makes -- and that I don't see much discussed when it comes to Haidt's work -- is that Haidt's prescriptions for a solution are dangerous. If Haidt is wrong, one of the most unsettling consequences of relying on his solutions is that we stop looking for the real causes of any degradation in childhood mental health. Even if he's right, his solutions will have all sorts of unintended consequences, including loss of privacy on the internet, lack of access by children to resources that are in fact helpful to their mental health, and violations of various legal and constitutional human rights frameworks.

navane 3y ago

It's risk times consequences. The risk assessment might not be up to scientific standards, but the consequences are big. Also, it's not just about statistics, in this case we have a pretty intuitive grasp on the mechanics at play, they are not far fetched.

kelseyfrog 3y ago
Risks have to be based in reality. Our greatest reality check is the scientific process. We can make up risks endlessly, see the AI-as-an-existential-threat debate. It simply is not grounded in reality. All I'm asking is transparency. People make non-scientific claims and conclusions all the time. Folks just need to be honest that this is what they're doing so they aren't fooling themselves and others.

hackerlight 3y ago

Haidt's rebuttal to your argument of spurious correlation: https://www.afterbabel.com/p/phone-based-childhood-cause-epidemic https://www.afterbabel.com/p/phone-based-childhood-cause-epi...

kelseyfrog 3y ago
> Alcott and colleagues (2020) randomly assigned 2743 adults. I stopped there. We're talking specifically about teenage depression and the cited study is adults. If Haidt wants us to take results from one subgroup and apply it to another, he has to provide evidence. That's how this process works.
- neglesaks 3y ago
  You conveniently stopped reading at the first cite, before he provided more experimental evidence.
  - kranke155 3y ago
    Because he has no intention of being convinced by data that goes against his beliefs, which is an ego based anti scientific stance.
    - kelseyfrog 3y ago
      I'm much more interested in the rhetorical device, "I stopped reading at X and couldn't take anything after that seriously." It looks like I have some more learning to do before employing it successfully.
    - cap1434 3y ago
      Project much?
      - kranke155 3y ago
        how did I project? Where do you see me doing the same thing?
- hackerlight 3y ago
  This is weirdly uncharitable. Why would you stop there when there's 22 experimental studies[1] and the majority of the article left to read? [1] https://docs.google.com/document/d/1w-HOfseF2wF9YIpXwUUtP65-olnkPyWcgF5BiAtBEy0/edit https://docs.google.com/document/d/1w-HOfseF2wF9YIpXwUUtP65-...
- tomrod 3y ago
  > If Haidt wants us to take results from one subgroup and apply it to another, he has to provide evidence. That's how this process works. Perhaps considering where he does so, further in the article which you stopped reading.

nabla9 3y ago

That's not true. 1. Haid has evidence from randomized experiments with kids. 2. There are natural experiments caused by different regional adoption rates of high-speed internet. 3. When the effect size of harm correlates with the frequency of use in population, that's more than time series and spurious-correlations. Social Media and Mental Health: A Collaborative Review https://docs.google.com/document/d/1w-HOfseF2wF9YIpXwUUtP65-olnkPyWcgF5BiAtBEy0/edit https://docs.google.com/document/d/1w-HOfseF2wF9YIpXwUUtP65-...

kelseyfrog 3y ago
It's not possible to blind the experiment's test subjects. Again, all I'm asking for is the level of critique to be comparable to the strength of the claims. We are eager to critique lower stakes claims, but for some reason Haidt typically gets a pass here. All I'm asking for is consistency. I'm not sure how to put this, but effect size and causation are different things. It's a category mistake to compare.
- nabla9 3y ago
  > It's not possible to blind the experiment's test subjects. It's not necessary. Randomized experiment is good enough. So are natural experiments.
timmytokyo 3y ago
Mike Masnick has written a review of Haidt's book, and it's a nice rebuttal to each of these arguments. https://www.thedailybeast.com/the-coddling-of-the-american-parent https://www.thedailybeast.com/the-coddling-of-the-american-p... One of the points he makes -- and that I don't see much discussed when it comes to Haidt's work -- is that Haidt's prescriptions for a solution are dangerous. If Haidt is wrong, one of the most unsettling consequences of relying on his solutions is that we stop looking for the real causes of any degradation in childhood mental health. Even if he's right, his solutions will have all sorts of unintended consequences, including loss of privacy on the internet, lack of access by children to resources that are in fact helpful to their mental health, and violations of various legal and constitutional human rights frameworks.