The problem being, RCTs are tricky to set up for social variables to have enough statistical power, while avoiding the "test bias" which is known to change behavior. (Essentially a kind of placebo effect.) It has to be huge.

And how do you produce a social placebo to detect such effects by comparison? No treatment is not the same as known fake treatment.

In fact, even the observational study here is quite underpowered and not multicultural. It is to be taken with a few bags of salt.