What I'd hope to see is randomized studies or studies that exploit some random variation for identification. The studies mentioned in the article all appear to

What I'd hope to see is randomized studies or studies that exploit some random variation for identification. The studies mentioned in the article all appear to rely on regressions controlling for observables. But of course there could be plenty of unobservable factors that are driving the result: the type of people that spend a lot of time in the sun are healthier for other reasons. I haven't read Weller's work though so maybe he has some RCTs too or maybe someone else does.