Yes, many of us have also taken a college statistical course and understand correlation != causation. Yes, pop-sci articles will more heavily infer there is a causation, and this is bad for public trust in science. Also, dismissing all correlation studies is bad for public trust in science. There's several very good reasons why we do correlation studies, and they actual return interesting data.
I'd like to point out that blood alcohol levels are not 1 to 1 connected to level of impairment, but still serve as a useful indicator for ability to drive. Those with high tolerances behave differently than those with lower tolerances. The current Cannabis test is far from perfect, but seems to be the best proxy we have available for empirical evidence of level of impairment.
There's no methods of field sobriety testing that are actually reliable, what it's best at doing is allowing police to get probable cause even if a sober driver just happens to not be good at something they've not done before and are being asked to do in the dark and cold outside their car in a high stress moment; or, worse, if the police officer judging them is just biased and wanting to subjectively decide they failed.
I'm pretty sure lawyers' advice is generally to say no when asked to take a field sobriety test, as you're basically only asked to do it if the police already think you're going to fail and therefore will be at minimum subconsciously biases towards expecting that. Much better to only let them do any breath/blood tests they can legally insist on. (At least, if you are indeed sober. I don't know what the best advice is if you're going to fail those tests, maybe in that case a tiny chance of being convincing with a field sobriety test is worth the chance?)
I'd like to point out that blood alcohol levels are not 1 to 1 connected to level of impairment, but still serve as a useful indicator for ability to drive. Those with high tolerances behave differently than those with lower tolerances. The current Cannabis test is far from perfect, but seems to be the best proxy we have available for empirical evidence of level of impairment.