Emil O W Kirkegaard

It's not clear how many are poorly described and how many are just wrong. Supposedly hedge funds that invest in medicine assume that 50% of reported results are wrong.

In nutrition/food science it's even worse. It's acceptable to survey people about a few dozen foods they eat and medical conditions they might have, then, Brian Wansink style, publish any association where p=.049.

I can't comment on nutrition science, which seems quite sloppy. However, in molecular biology/cancer research, at the cutting edge of knowledge, the degree of technical complication and expertise is huge, a certain failure rate is expected, it can be genuinely 'hard' to reproduce

Analogy, if you took a group of qualified engineers, and gave them the requisite resources, + written instructions, and asked them to reproduce a functioning space shuttle, they could probably do it, but there would be a failure rate

Yet most people accept that the space shuttle worked (most of the time)

Maybe. But the claim you're making is controversial. Cancer researchers were not entirely confident in these results—that's why they tried to replicate them. After they failed, some people offered post hoc explanations: too complex, original methods not described in detail, etc.

OK, if you pick out studies that your experience tells you are less likely to reproduce, then you could get a high fail-to-replicate rate. Amgen has never revealed their studies, so we don't know how cherry picked their sample was, or whether their replication attempts sucked.

I'm a molecular biologist who knows virology extremely well and cancer research somewhat. The claim that that "47/53 studies are incorrect" is simply not a fair reflection of the overall field.