To reliably identify causes, as opposed to correlations, a higher standard of evidence is required: the controlled trial. In its simplest form: recruit a group of subjects, and assign half of them a diet for, say, 15 years. At the end of the trial, assess the health of those in the intervention group, versus the control group. This method is also problematic: it is virtually impossible to closely supervise the diets of large groups of people. But a properly conducted trial is the only way to conclude with any confidence that X is responsible for Y.

...scientific inquiry prone to the eternal rules of human social life: deference to the charismatic, herding towards majority opinion, punishment for deviance, and intense discomfort with admitting to error. Of course, such tendencies are precisely what the scientific method was invented to correct for, and over the long run, it does a good job of it. In the long run, however, we’re all dead, quite possibly sooner than we would be if we hadn’t been following a diet based on poor advice.