5 ms·

I'm not sure how fake data is substantially different from any other part of a fake paper. Just like how randomly-generated sentences might sound reasonable, fa

I'm not sure how fake data is substantially different from any other part of a fake paper. Just like how randomly-generated sentences might sound reasonable, faked data might look like real data. In both cases, we expect peer review to engage and to consider how data was sourced. This happened in the "Who's Afraid of Peer Review?" incident [0], where the offending papers were intentionally designed to have many red flags that would have helped reviewers understand that the data was bogus. More worryingly, data cannot model itself, so the quality of data should be largely irrelevant to any models which purport to explain it; as a result, a paper's contribution to science needs to be structured to work with other results, and cannot simply proclaim that its modelling is correct because some given equation fits the observed data. To pick a situation where your expertise shines, imagine that somebody submits a paper about a 3-SAT solver, and they include data that shows their solver doing not just extremely well on standard problem sets, but asymptotically scaling better than exponentially. The reason that you might doubt the honest presentation of the data is because you know, from having studied the field, that such behavior is unlikely. [0] https://en.wikipedia.org/wiki/Who%27s_Afraid_of_Peer_Review%3F https://en.wikipedia.org/wiki/Who%27s_Afraid_of_Peer_Review%...

tptacek 6y ago
No, we don't expect peer reviewers to investigate the sourcing of data. A journal reviewer might be assigned a workload of dozens of papers to be completed, in snippets of spare time, over the course of a month or two. Peer reviewers aren't equipped to do that kind of vetting; that's what replication is for, and replications are their own research projects. It's really surprising what kinds of expectations people seem to have about peer review and what it can reasonably accomplish. "Who's Afraid Of Peer Review" targeted fee-charging open-access journals, which are financially incentivized to accept random papers. Journal reviewers are generally unpaid postgraduate academics.
- jxramos 6y ago
  Spot on about unreasonable expectations; omne ignotum pro magnifico est always pokes its head when guessing on unfamiliar processes and procedures and what not. So funny how human nature when it assumes, assumes big grand intensive things! I'm frequently guilty of it myself. This is news to me about the reasonable costs absorbed from peer review, makes so much sense the level of vetting that's employed.
- Kednicma 6y ago
  I don't want to quibble with your experiences of peer review; your experiences are valid. But what does a journal provide, then, exactly? Are they relics of the era before cheap self-publication?
  - tptacek 6y ago
    "A panel of experts from this field read this paper and felt it likely to make significant contributions". For high-impact journals, you can add "so much so that it was selected to be among the 25% of submissions this cycle to be published". What peer review does not say is "a panel of experts carefully vetted this paper to ensure that its conclusions are accurate". In fact, part of the premise of replication projects is that peer review doesn't say that. I've only been a reviewer a couple times, and only in computer science. Other people on HN have experience reviewing for other hard science venues. Maybe their experiences are different. But I think it's notable that you don't hear that in comments about what this hoax exposed; has there been any HN comment from a reviewer saying that they were expected to rigorously vet submissions, the way a PhD board does with a thesis? Daniel Engber makes many of these points better than I do, here: https://slate.com/technology/2018/10/grievance-studies-hoax-not-academic-scandal.html https://slate.com/technology/2018/10/grievance-studies-hoax-...
    - thu2111 6y ago
      The problem is not so much the system as how it's presented: the media presents "peer-reviewed study" as a gold standard, practically synonymous with correct. And unfortunately academia absolutely encourages this practice. The message that peer review is largely meaningless hasn't really been made by academia, for obvious reasons, and thus stupid papers getting accepted by journals will continue to be interesting and news-worthy until people get the message that peer review doesn't mean much. The world will then move on to "replicated" as a gold standard. This will be better but not by much. Just a few weeks ago Imperial College London published a press release claiming their Report 9 results from their COVID-19 simulator program had been replicated. Worth a press release because after it was open sourced it was discovered to be filled with non-deterministic behaviour, even with fixed RNG seeds. Unfortunately the press release was fraudulent. The report they cited as evidence of replication was by a friendly academic. He said he was able to replicate the results, then admitted every number he got out was different, some by 10%-25% different. This is the output of a computer simulation so the allowable difference is 0%. Despite this, Nature and other outlets proceeded to report that ICL's COVID model was "replicated". In the end, academia is strongly incentivised to appear credible, not actually be so. Being credible requires you can reliably turn your findings into something useful, as when in the realm of pure theory you can never be sure your findings actually hold up in reality. Corporate research has this attribute (eventually), academic research doesn't, as academics are rewarded for writing clever sounding papers not being correct.
- free_rms 6y ago
  But did the authors of "who's afraid of peer review" face censure? The only difference is who's ox got gored.
  - tptacek 6y ago
    No, that's not the only difference. In fact, the two projects were more dissimilar than similar. Notably, the point of "Who's Afraid Of Peer Review" is that no reasonable reviewer could accept the paper for any reason; the papers (they were all practically identical) contained self-contradicting data; in fact, the plots in the paper were contradictory, so you couldn't even just skim the abstract and the graphics and then reasonably accept the paper. And, of course, the WAOPR papers were claiming to have cured cancer. The WAOPR papers were designed to be trivial to spot. That's not the case for the S2 papers; the "dog park" paper, for instance, intricately describes 1000 hours of field work. And, of course, there's the distinction that WAOPR papers targeted commercial services asking for money to publish papers, not volunteer peer review time.
    - free_rms 6y ago
      You're saying it's reasonable to publish "main kampf rewritten using woke language", and "a study on how men should regularly use dildos on themselves to improve their social attitutes"? If it was just the dog park paper, I could see your point, but come on, those papers are completely farcical.

tptacek 6y ago

No, we don't expect peer reviewers to investigate the sourcing of data. A journal reviewer might be assigned a workload of dozens of papers to be completed, in snippets of spare time, over the course of a month or two. Peer reviewers aren't equipped to do that kind of vetting; that's what replication is for, and replications are their own research projects. It's really surprising what kinds of expectations people seem to have about peer review and what it can reasonably accomplish. "Who's Afraid Of Peer Review" targeted fee-charging open-access journals, which are financially incentivized to accept random papers. Journal reviewers are generally unpaid postgraduate academics.

jxramos 6y ago
Spot on about unreasonable expectations; omne ignotum pro magnifico est always pokes its head when guessing on unfamiliar processes and procedures and what not. So funny how human nature when it assumes, assumes big grand intensive things! I'm frequently guilty of it myself. This is news to me about the reasonable costs absorbed from peer review, makes so much sense the level of vetting that's employed.
Kednicma 6y ago
I don't want to quibble with your experiences of peer review; your experiences are valid. But what does a journal provide, then, exactly? Are they relics of the era before cheap self-publication?
- tptacek 6y ago
  "A panel of experts from this field read this paper and felt it likely to make significant contributions". For high-impact journals, you can add "so much so that it was selected to be among the 25% of submissions this cycle to be published". What peer review does not say is "a panel of experts carefully vetted this paper to ensure that its conclusions are accurate". In fact, part of the premise of replication projects is that peer review doesn't say that. I've only been a reviewer a couple times, and only in computer science. Other people on HN have experience reviewing for other hard science venues. Maybe their experiences are different. But I think it's notable that you don't hear that in comments about what this hoax exposed; has there been any HN comment from a reviewer saying that they were expected to rigorously vet submissions, the way a PhD board does with a thesis? Daniel Engber makes many of these points better than I do, here: https://slate.com/technology/2018/10/grievance-studies-hoax-not-academic-scandal.html https://slate.com/technology/2018/10/grievance-studies-hoax-...
  - thu2111 6y ago
    The problem is not so much the system as how it's presented: the media presents "peer-reviewed study" as a gold standard, practically synonymous with correct. And unfortunately academia absolutely encourages this practice. The message that peer review is largely meaningless hasn't really been made by academia, for obvious reasons, and thus stupid papers getting accepted by journals will continue to be interesting and news-worthy until people get the message that peer review doesn't mean much. The world will then move on to "replicated" as a gold standard. This will be better but not by much. Just a few weeks ago Imperial College London published a press release claiming their Report 9 results from their COVID-19 simulator program had been replicated. Worth a press release because after it was open sourced it was discovered to be filled with non-deterministic behaviour, even with fixed RNG seeds. Unfortunately the press release was fraudulent. The report they cited as evidence of replication was by a friendly academic. He said he was able to replicate the results, then admitted every number he got out was different, some by 10%-25% different. This is the output of a computer simulation so the allowable difference is 0%. Despite this, Nature and other outlets proceeded to report that ICL's COVID model was "replicated". In the end, academia is strongly incentivised to appear credible, not actually be so. Being credible requires you can reliably turn your findings into something useful, as when in the realm of pure theory you can never be sure your findings actually hold up in reality. Corporate research has this attribute (eventually), academic research doesn't, as academics are rewarded for writing clever sounding papers not being correct.
free_rms 6y ago
But did the authors of "who's afraid of peer review" face censure? The only difference is who's ox got gored.
- tptacek 6y ago
  No, that's not the only difference. In fact, the two projects were more dissimilar than similar. Notably, the point of "Who's Afraid Of Peer Review" is that no reasonable reviewer could accept the paper for any reason; the papers (they were all practically identical) contained self-contradicting data; in fact, the plots in the paper were contradictory, so you couldn't even just skim the abstract and the graphics and then reasonably accept the paper. And, of course, the WAOPR papers were claiming to have cured cancer. The WAOPR papers were designed to be trivial to spot. That's not the case for the S2 papers; the "dog park" paper, for instance, intricately describes 1000 hours of field work. And, of course, there's the distinction that WAOPR papers targeted commercial services asking for money to publish papers, not volunteer peer review time.
  - free_rms 6y ago
    You're saying it's reasonable to publish "main kampf rewritten using woke language", and "a study on how men should regularly use dildos on themselves to improve their social attitutes"? If it was just the dog park paper, I could see your point, but come on, those papers are completely farcical.