[jyxent]

You misunderstand. The issue is that many more papers should be retracted. A paper this major that couldn't be replicated still took another decade to be retracted.

There are many more papers that are less cited that will never get looked at in enough detail to discover fraud.

[arp242]

> The issue is that many more papers should be retracted.

The reality is we don't know. Should it be more than 8 out of 10,000? Almost certainly, since quite a lot of papers barely get any attention after publication.

But should the number be 10 out of 10,000? 20 out of 10,000? 100 out of 10,000? We don't know.

And boring papers without any interesting conclusions published just for CV reasons also don't really matter, and the 10,000 figure is also wrong, and we also don't really know what the correct value for that figure is.

[mike_hearn]

It's going to be really field dependent. Computer science has fewer problems than social psychology for example.

To put a figure on it, consider that about 0.04% of papers are retracted but about 2.5% of scientists admit in polls to having engaged in fraud. Obviously though most fraudsters won't admit to it (they often convince themselves it's not fraud via mental gymnastics), so another way to approach it is to ask them to estimate what proportion of other scientists commit fraud. Then they answer 10%.

It's a bit hard to know how many papers are published per year, I've seen estimates of anywhere from 2 to 5 million. If 10% of scientists are committing fraud and they were all detected, that would yield on the order of maybe half a million to a million retractions per year, as we'd expect fraudulent scientists to be much more productive than the ones that play fair.

But fraud isn't the only reason that papers should be retracted. You'd expect serious non-fraud problems to be a reason too. If papers were retracted for all the reasons non-scientists imagine papers are retracted, e.g. being foundationally based on another paper/dataset that's been retracted, using invalid methodologies, internally inconsistent statistics, serious software bugs etc, you could probably get up to 50% or more in some fields.

This sounds absurdly high but is the number estimated by Richard Horton, editor of The Lancet. And according to Marc Andreessen, when he asked if it could be true that half of all biotech research was fake the head of a major US funding agency actually laughed at this figure and said no, in biotech it's more like maybe 90% is fake. So the problem does seem to be huge.

[wolverine876]

You're just making up results, sans data. Your paper won't even pass peer review - on HN.

[jyxent]

And you downplay evidence of fraud. You mentioned studies where 1/3 to 1/2 of reviewed papers are found to not be replicable and say they get similar but weaker results. This is evidence of p-hacking or data manipulation. You should expect far less papers to not be replicable given proper statistics.