>the regression is going to end up with non-zero coefficients on it
This is just not true. Plenty of regressions have coefficients that arent statistically distinct from zero.
>But this approach cannot tells us if those inputs are drivers, much less the main ones.
Again, untrue. There are plenty of statistically appropriate ways to estimate causality. You might consider looking into the latter-day work of Judea Pearl, a well known computer scientist. "The Book of Why" seems like a decent place for you to start, because ut is for the layperson, and you have a ton of fundamental errors in your statements of "fact."
>unfortunately with media, politicians, and the general public so science illiterate that they're unable to engage with this research on its own terms
You should also add "the confidently incorrect" to your list!
> This is just not true. Plenty of regressions have coefficients that arent statistically distinct from zero.
Sure. But you are responding to something I didn't claim, or at least didn't intend to claim. If you throw in a spurious piece of data that happens to exhibit the same trend, it's going to end up a non-zero coefficient. But that doesn't mean there is a causal relationship.
> There are plenty of statistically appropriate ways to estimate causality.
I didn't claim otherwise, but that doesn't help for an extremely underpowered analysis which not only didn't even consider causality but didn't consider alternative hypothesis. (In particular, it didn't even consider the null hypothesis except perhaps in some p-hack sense that they may not have published at all if the none of their coefficients had significance according to R's GLM or whatever package they used). That wasn't it's goal, I'm not even accusing the authors of bad science (at a minimum it passed the bar to get published)-- but the conclusions the media were drawing from it couldn't be supported by the work. It's an easy error to make because there is a gap between what we want to know and what we have the data to tell us.
This is just not true. Plenty of regressions have coefficients that arent statistically distinct from zero.
>But this approach cannot tells us if those inputs are drivers, much less the main ones.
Again, untrue. There are plenty of statistically appropriate ways to estimate causality. You might consider looking into the latter-day work of Judea Pearl, a well known computer scientist. "The Book of Why" seems like a decent place for you to start, because ut is for the layperson, and you have a ton of fundamental errors in your statements of "fact."
>unfortunately with media, politicians, and the general public so science illiterate that they're unable to engage with this research on its own terms
You should also add "the confidently incorrect" to your list!