I also liked the emphasis on non-classical measurement error. The study described in the article about cancer patients who "remember" eating diets higher in fat than they themselves reported prior to the appearance of the cancer shows the power of lay theory at work. The women had adopted a popular theory regarding the cause of cancer and then (unconsciously) adjusted their reports of past eating behavior to match the theory.
I would have liked to learn more about what variables were conditioned on in the "case-control" studies. The natural reaction of economists to the odd findings from these studies would have been to go hunting for unobserved confounders.
Finally, I was a bit puzzled by this bit:
What is clearly needed, experts agree, is a single, definitive, unbiased study — “one trial,” to borrow Paget’s terminology.
Having one (supposedly) definitive study is great for journalists as it reduces their reading load and simplifies their story lines. And in contexts where larger randomized trials on humans with reasonable external validity are possible, perhaps the one grand study really is the goal. But what is described in the article is an animal study so there will always be external validity concerns. In the world of economics, journalistic focus on the "one easily explained, well-published study by famous people at a leading university" sometimes leads to the provision of a very misleading picture of the overall body of evidence and the overall level of agreement in the field. Better to do the work of actually reading, and reporting, the literature.