>>for example, trials could only be offered as evidence for the efficacy of a drug if they had been registered in advance.

I had thought that that was the point of clinicaltrials.gov - forcing researchers to preregister studies (including methods and number of participants and, as I remember, either p-value or expected magnitude of result - but you can get one from the other two) in order to be accepted for publication.

Sounds like the FDA isn't using this?

That said, for anyone at sea with what that chart is illustrative of, check "How to Lie with Statistics", a book aimed at High Schoolers, so you can breeze through. I still remember the example used there correlating the salaries of Methodist ministers with the price of rum in Jamaica from the mid 60s when I read it.

Finally, re "Even if one vaccine is safe (itself a matter of debate) ". I'd agree that there have been insufficient studies on the number of childhood vaccines given at short intervals, but individual vaccines have been well studied - to the point that an IRB would be unlikely to agree to withholding one in a placebo group - but we have some natural experiments in progress (polio, mmr, measles) - none of which make me happy.

Other than that, a well written, well-reasoned article.