3 ms·

I don't think this kind of fraud is even remotely close in scale to the cycling doping you mention. My primary reason for believing this is that as a scientist

I don't think this kind of fraud is even remotely close in scale to the cycling doping you mention. My primary reason for believing this is that as a scientist myself, I've never done it, and none of my coauthors has ever proposed to do something like that. I have seen other kinds of fraud (e.g. ghost authorship) but not result manipulation. But of course, If I were a fraudster myself I would also say that I've never done it :) So a more general argument: scientific "doping" is just not as straightforward as cycling doping. In cycling, it's clear that if you don't get caught (and have no scruples) you benefit from doping, it makes you stronger. In science, you first need an idea that looks original/promising, and then you do the experiments and see if the idea works or not. This kind of "doping" will only help if you (1) are good at generating ideas, but (2) these ideas often don't work. At least in my field, this is possible but far from the most common situation to be. And in other situations, there is not much benefit to faking results.

awesomeMilou 3y ago
> if you (1) are good at generating ideas I have been joking with my former boss (who is a professor) about how easy it is to come up with good ideas. We each have five to six One Billion Dollar ideas before breakfast, and another 1-2 while in the shower! /s Coming up with ideas is _ridicolously_ easy, executing on them is the hard part. If you're a seasoned researcher, it shouldn't be hard to guess at the "meta" of your field, you're probably aware of promising ideas discussed among your colleagues. What if you faked results, proving one of these ideas? And the bar for cheating is _suprisingly_ low. I'm not a researcher myself, but I've experienced how my work is torn apart in peer review. But if you're dealing with data, apparently no one makes sure to actually check it before approval.[0][1] [0] https://youtu.be/d2Tm3Yx4HWI?t=846 https://youtu.be/d2Tm3Yx4HWI?t=846 (analysis on the original data from the Francesca Gino scandal, which already looks suspicious without statistical analysis) [1] https://youtu.be/KsSuhP60qnI https://youtu.be/KsSuhP60qnI (faked data from the Hendrik Schoen scandal at Bell Labs, as long as you make outrageous claims no one can verify, you can continue to publish in top journals)
- Al-Khwarizmi 3y ago
  I suppose this is heavily discipline-specific. In my specific subfield of CS, the execution is often a Python program or bunch of scripts that can be (and often are) coded by an undergrad if given enough guidance on what exactly is wanted. And then of course the way of presenting it in the paper, choosing which data to show, etc. but cheating doesn't help much there. The hard part in my view is coming up with what Python program should be written (i.e. coming up with a relevant algorithm, experiment, evaluation methodology, etc.). I can see how in other fields, e.g. biomedical, execution could be more challenging and ideas more plentiful. Anyway, I don't think ideas (read: good ones, because obviously anyone can have nonsense ideas) can be a commodity in any field. It is quite widely accepted that, e.g., when applying for an ERC grant (the highest-regarded individual research grants in Europe), having a good idea is as important as the CV if not more. This wouldn't be the case if coming up with good ideas were so trivial.
  - awesomeMilou 3y ago
    > having a good idea is as important as the CV if not more I don't know about the acceptance criteria for ERC grants, but from the peer reviews I've read from one of my professors papers it feels like there's a bias towards accepting novelty of all else. My joke with ideas was meant more general, i.e. that you can have a few billion dollar ideas, but executing on them is much harder (due to lack of funds, time or skill). Additionally, depending on the demand of the field I can imagine it being easier to obtain funding for resarch. Imagine being in machine learning right now lol. I wasn't talking about grant acceptance though, but purely what kind of ideas can get you published and what data you have to fake in order to get there. In the case of Francesca Gino it doesn't sound like her fraudulent resarch was so costly that she would have to have applied for large grants in order to conduct it.
slingnow 3y ago
You make it sound so straightforward and simple. > and then you do the experiments and see if the idea works or not The problem is "if the idea works or not" can be very fuzzy. With a little p-hacking, anything can work! If nothing in your original data/experiment supports your hypothesis, you can just cherry pick some part of it that gives "significant" results.
noobermin 3y ago
>My primary reason for believing this is that as a scientist myself, I've never done it, and none of my coauthors has ever proposed to do something like that. Are you sure? Also a) how far along are you and b) how far from actual research work are you at this point?
- Al-Khwarizmi 3y ago
  I'm a full professor and PI of several projects. So sadly more management than actual hands-on research lately, but I've gone through all stages, of course. And yes, I have more than 100 papers and in their making I haven't fabricated data or seen anyone do so. If there is fake data in any of them, it would be because someone included it behind my back. But I doubt it's the case, as during the course of my career we tried various things that didn't work and we just moved on to other experiments. I've never had a student or collaborator even suggest "let's just edit the data and publish it". The most that we have done is to try to cast negative results in a positive light to make them publishable ("the results are promising...", "the results are mixed but show improvements for some dataset..."), but without touching the data or saying anything factually incorrect.

awesomeMilou 3y ago

> if you (1) are good at generating ideas I have been joking with my former boss (who is a professor) about how easy it is to come up with good ideas. We each have five to six One Billion Dollar ideas before breakfast, and another 1-2 while in the shower! /s Coming up with ideas is _ridicolously_ easy, executing on them is the hard part. If you're a seasoned researcher, it shouldn't be hard to guess at the "meta" of your field, you're probably aware of promising ideas discussed among your colleagues. What if you faked results, proving one of these ideas? And the bar for cheating is _suprisingly_ low. I'm not a researcher myself, but I've experienced how my work is torn apart in peer review. But if you're dealing with data, apparently no one makes sure to actually check it before approval.[0][1] [0] https://youtu.be/d2Tm3Yx4HWI?t=846 https://youtu.be/d2Tm3Yx4HWI?t=846 (analysis on the original data from the Francesca Gino scandal, which already looks suspicious without statistical analysis) [1] https://youtu.be/KsSuhP60qnI https://youtu.be/KsSuhP60qnI (faked data from the Hendrik Schoen scandal at Bell Labs, as long as you make outrageous claims no one can verify, you can continue to publish in top journals)

Al-Khwarizmi 3y ago
I suppose this is heavily discipline-specific. In my specific subfield of CS, the execution is often a Python program or bunch of scripts that can be (and often are) coded by an undergrad if given enough guidance on what exactly is wanted. And then of course the way of presenting it in the paper, choosing which data to show, etc. but cheating doesn't help much there. The hard part in my view is coming up with what Python program should be written (i.e. coming up with a relevant algorithm, experiment, evaluation methodology, etc.). I can see how in other fields, e.g. biomedical, execution could be more challenging and ideas more plentiful. Anyway, I don't think ideas (read: good ones, because obviously anyone can have nonsense ideas) can be a commodity in any field. It is quite widely accepted that, e.g., when applying for an ERC grant (the highest-regarded individual research grants in Europe), having a good idea is as important as the CV if not more. This wouldn't be the case if coming up with good ideas were so trivial.
- awesomeMilou 3y ago
  > having a good idea is as important as the CV if not more I don't know about the acceptance criteria for ERC grants, but from the peer reviews I've read from one of my professors papers it feels like there's a bias towards accepting novelty of all else. My joke with ideas was meant more general, i.e. that you can have a few billion dollar ideas, but executing on them is much harder (due to lack of funds, time or skill). Additionally, depending on the demand of the field I can imagine it being easier to obtain funding for resarch. Imagine being in machine learning right now lol. I wasn't talking about grant acceptance though, but purely what kind of ideas can get you published and what data you have to fake in order to get there. In the case of Francesca Gino it doesn't sound like her fraudulent resarch was so costly that she would have to have applied for large grants in order to conduct it.

slingnow 3y ago

You make it sound so straightforward and simple. > and then you do the experiments and see if the idea works or not The problem is "if the idea works or not" can be very fuzzy. With a little p-hacking, anything can work! If nothing in your original data/experiment supports your hypothesis, you can just cherry pick some part of it that gives "significant" results.

noobermin 3y ago

>My primary reason for believing this is that as a scientist myself, I've never done it, and none of my coauthors has ever proposed to do something like that. Are you sure? Also a) how far along are you and b) how far from actual research work are you at this point?

Al-Khwarizmi 3y ago
I'm a full professor and PI of several projects. So sadly more management than actual hands-on research lately, but I've gone through all stages, of course. And yes, I have more than 100 papers and in their making I haven't fabricated data or seen anyone do so. If there is fake data in any of them, it would be because someone included it behind my back. But I doubt it's the case, as during the course of my career we tried various things that didn't work and we just moved on to other experiments. I've never had a student or collaborator even suggest "let's just edit the data and publish it". The most that we have done is to try to cast negative results in a positive light to make them publishable ("the results are promising...", "the results are mixed but show improvements for some dataset..."), but without touching the data or saying anything factually incorrect.