Comment by D-Machine | Hacker News Reader

Comment by D-Machine | Hacker News Reader

Parent Full thread

You're gish-galloping a bunch of random studies, most of which appear to be cross-sectional. I've linked a broad review.

Even the longitudinal studies are poor here. See, for example, as this Nature article notes:

    "The study has multiple limitations that need to be considered. First, to interpret the parameters from our analyses as estimates of causal effects one would need to adopt the following assumptions: (a) there are no time-varying unobserved confounders that impact the relation between social media use and life satisfaction; (b) the model adequately accounts for unobserved time-invariant confounding through the inclusion of a random intercept; (c) there is no measurement error in the variables; (d) the time interval between studies (one year) is the right length to capture the effects of interest; and (e) the bidirectional links estimated by our longitudinal model are linear in nature. Only if these assumptions are met can this observational study be said to capture the causal effects between social media and life satisfaction. Second, the data are self-report and therefore only allow inferences about the impact of self-estimated time on social media, rather than objectively measured social media use." https://www.nature.com/articles/s41467-022-29296-3#Sec2

I'd also suggest looking at the coefficients (effect sizes) in the above (standardized regression coefficients barely approaching 0.2 - and this is one of the stronger findings), and other articles. The effects here, even if we were to pretend they were clearly established, are incredibly tiny. Examples:

- social media explaining only 0.4% of variance (https://pubmed.ncbi.nlm.nih.gov/30944443/)

- social-media/mental-health effect around β = .061 (https://christopherjferguson.com/Social%20Media%20Meta.pdf)

These are basically nothing, and yet you have such absolute confidence from people that social media is this big harmful thing. The evidence just isn't there.

~Qualifying all of this with the caveat that I'm not a social scientist.~

I really appreciate your response, which made me realize this is more nuanced than I thought at first blush. (I wrote that reply before you linked the National Academies review, which was quite helpful.)

Maybe I am truly morally panicked, but I'm really hesitant to brush aside the evidence as "basically nothing." Small effects are worth paying attention to (https://psycnet.apa.org/record/2022-26026-014). And quantifying the harms of social media use is less like quantifying the harms of, say, smoking cigarettes, because social media has (indeed, is built on) network effects. You could plop a kid from the 50's into a modern American neighborhood and their mental health might decline even if they don't use social media because of the way it has changed childhood. When the social life of an entire generation is transformed by technology, it stops making sense to ask "How much does one hour of Instagram hurt mental health?" Yet that is essentially the question all the studies are designed to answer, and even still, we see an effect.

Anyway, thanks for taking to time to explain your perspective.

Yes, sorry about the edits.

IMO the correct framing most supported by the evidence is that a few vulnerable populations (most likely within a certain age bracket, and already dealing with other significant life issues) are definitely at risk of nontrivial harm from social media. Otherwise, for the majority, it is likely a wash. And since the bias in studies is to look for negatives (much like drug studies don't properly measure negative side effects until much later), we also can't say if there aren't subgroups benefiting hugely as well (e.g. autistic people that can connect online, or other people with unusual interests that before would have remained isolated and disconnected, feel crazy, etc).

So, yes, small effects can be important. The observed effects for social media are quite small and consistent with being driven solely by a small, vulnerable population, however, and this make broad social injunctions and moral panic less defensible.

Also, these are effects on self-report or other psychological instruments, where in general you need to find something like a minimal clinically important difference before you can determine if you care practically about these things. E.g. if a standardized regression coefficient of 0.1 translates to a 1-point shift on a 20-point life-satisfaction scale, is this even something I (or anyone) can even notice? Does it even rise above measurement error?

Since no one is doing that kind of research here, we just get these (relatively meaningless) standardized effect sizes, and it is really basically impossible to know if we should care at all about the observed effects. This is mainly what I mean by the evidence being flimsy / weak, it is really too inconclusive to drive decisions at this point.

Reply on news.ycombinator.com