* There are tons of randomized controlled trials of policy measures (malaria bednets, minimum income). Many measure long-run outcomes.
* Natural experiments can measure long-run effects. In economic history, sometimes that means centuries.
* Many other designs are plausibly causal. The right instrumental variable, or a regression discontinuity design. In some cases, even a simple diff-in-diff with panel data. This design, nope.
The treatment here is regular social media use and video game playing. This not something where you can randomize people to treatment conditions, like you can with malaria.
Likewise, tell me -- what would be a good instrument for estimating a causal effect of video game playing? What measure would plausibly affect intelligence only through video game playing? Where is there a natural experiment that allows for an RD design where young people on one side of the discontinuity play video games and the others do not?
We get it -- you've taken a causal inference or econometrics course and want people to know it.
No, my point is that bad research which conforms to people's priors should not be taken seriously.
There are many experiments on video game playing. Most of them are short run, obviously. But never underestimate researcher ingenuity. Here's a cute paper which uses as an instrument "did your roommate bring a video game to university"? Not beyond critique, but plausible: https://economics.uwo.ca/people/stinebrickner_docs/paper2.pd.... It's also relevant because the dependent variable is how much students study, and their resulting performance.
More to the point, if anyone seriously thinks video games will raise kids' IQ, and can persuade funders of it, they could simply give the treatment group an Xbox. That would be expensive - say $20K-ish - but much cheaper than the benefit of an extra IQ point or 2, over millions of children.
* There are tons of randomized controlled trials of policy measures (malaria bednets, minimum income). Many measure long-run outcomes.
* Natural experiments can measure long-run effects. In economic history, sometimes that means centuries.
* Many other designs are plausibly causal. The right instrumental variable, or a regression discontinuity design. In some cases, even a simple diff-in-diff with panel data. This design, nope.