4 ms·

I'm not convinced how any of this shows causation. Probably the strongest point in this substack pointing to causation is the use of longitudinal studies. Haidt

I'm not convinced how any of this shows causation. Probably the strongest point in this substack pointing to causation is the use of longitudinal studies. Haidt is using a meta-analysis of long-term specific (as opposed to short term studies which display no effect, there's merit here as a longer longitudinal study is probably better) longitudinal studies to make this point, so essentially saying that a meta-analysis of long-term longitudinal studies is required to show that causation can be established; this is all indirect. I still think that to prove this point we need to make an experiment that actually attempts to show causation. I'm also not sure if it's worth discussing this on HN because this topic tends to attract commenters of one opinion, but I also think it's worth making opposition known. I fully expect this opinion to be unpopular. EDIT: Indeed, doing a shallow dive into the referenced longitudinal study links shows that none of them establish any form of control. One of the studies linked's abstract (I don't have access to the full paper) has an intervention model but doesn't split the initial cohort in order to establish the causal effect. N.B. I am aware that in social science it's hard to find large enough cohort sizes to establish effects and that splitting a cohort to establish causality could destroy the statistical power of the study altogether.

cubefox 4y ago
Do you have an alternate hypothesis which could explain the results of all the studies the cites?
- Karrot_Kream 4y ago
  Of course, I have a few. The case for correlation is clear, and numerous studies of varying statistical power have proved it. The question is causation. So if social media usage isn't causal, then what is? I think most people who are skeptical of causation would point to a hidden latent variable that is actually causal. One way this could take place is some feature, say "susceptibility to fomo", that's actually causal. Folks who have a higher susceptibility might see frequent, sharp decreases in mental health outcomes from social media usage. Folks with low susceptibility might be relatively unaffected from social media usage. It could be interesting to see what the prevalence of this susceptibility is in the general population, whether there are certain cohorts that have high prevalence, and whether there are interventions that can be performed to decrease this prevalence. A similar analogous cause is the presence of schizophrenia in an analysis of psychedelic use. Most folks are fine using psychedelics infrequently. However, folks with a predilection for schizophrenia frequently have bad outcomes from psychedelic use and may even accelerate the onset of latent schizophrenia. The above "fomo susceptibility" might even be too simplistic a model. There might be a cluster of features, combinations of past experiences and perhaps biological traits, that actually act as "fomo susceptibility." I loosely think the real model works something like this. Age may be a controlling factor as well. Regardless, I'm not convinced there's causation here. Social media usage research is like a lot of other poorly understood psychological and mental health phenomenon where studies are underpowered, causality is hard to establish, and effect sizes are small which all lead to conflicting conclusions.

cubefox 4y ago

Do you have an alternate hypothesis which could explain the results of all the studies the cites?

Karrot_Kream 4y ago
Of course, I have a few. The case for correlation is clear, and numerous studies of varying statistical power have proved it. The question is causation. So if social media usage isn't causal, then what is? I think most people who are skeptical of causation would point to a hidden latent variable that is actually causal. One way this could take place is some feature, say "susceptibility to fomo", that's actually causal. Folks who have a higher susceptibility might see frequent, sharp decreases in mental health outcomes from social media usage. Folks with low susceptibility might be relatively unaffected from social media usage. It could be interesting to see what the prevalence of this susceptibility is in the general population, whether there are certain cohorts that have high prevalence, and whether there are interventions that can be performed to decrease this prevalence. A similar analogous cause is the presence of schizophrenia in an analysis of psychedelic use. Most folks are fine using psychedelics infrequently. However, folks with a predilection for schizophrenia frequently have bad outcomes from psychedelic use and may even accelerate the onset of latent schizophrenia. The above "fomo susceptibility" might even be too simplistic a model. There might be a cluster of features, combinations of past experiences and perhaps biological traits, that actually act as "fomo susceptibility." I loosely think the real model works something like this. Age may be a controlling factor as well. Regardless, I'm not convinced there's causation here. Social media usage research is like a lot of other poorly understood psychological and mental health phenomenon where studies are underpowered, causality is hard to establish, and effect sizes are small which all lead to conflicting conclusions.