Data Falsificada (Part 1): “Clusterfake” – Data Colada | Hacker News Reader

Data Falsificada (Part 1): “Clusterfake” – Data Colada

datacolada.org

43 pointsmalshe7 comments

I hope they use these techniques to develop an ML based methodology for detecting academic fraud. It seems like small potatoes, but the knock on effects are huge. The most obvious one is that academic studies are used as pretexts for public policy, and notably some very controversial ones where anyone who challenges the "science" will be the subject of a political opposition research campagin.

The other issue is that if you are among the cohort of people who have been declined a job, promotion, or cannot immigrate over degree requirements, academic fraud is utterly disgusting. Good people lose livelihoods over tiny missteps, and terrible people appear to make livelihoods on giant frauds. The people who were too stupid to cheat will find these frauds very difficult to forgive.

> The people who were too stupid to cheat will find these frauds very difficult to forgive.

Did you mean the people who were too “honest” instead of “stupid”? Otherwise you comment makes sense.

The academic cheaters you read about in these articles are always obvious cheats! I believe those to be in the minority, but who knows.

In OP's example, they copied the exact same lines to get the effect. With some more work they could've hidden their tracks far better.

Now if you build ML that detected good cheats you have a huge mess at your hand: the false-positives won't be zero! Are you OK to raise potentially career-destroying accusations based on a black box model? I am not.

(I could raise my usual point and say 'academia runs after pointless measurements, numbers of papers produced etc., and as long as you have those pointless goals you'll have cheats, so the system itself should be changed', but the older I get the less that change seems possible, wanted, or even considered.)

Confusing title, but quick read and worth it. Data collector only had to swap 3 data points with 3 others to get the fake effect in their study, and the other authors seemingly had no idea.

Indeed the title is confusing. I wanted to editorialize it to reflect the key point more accurately but I think HN usually doesn’t like it.

> That’s right: Two different people independently faked data for two different studies in a paper about dishonesty.

Maybe these research fraud sleuths' very limited capacity to investigate was drawn disproportionately to papers about dishonesty (because that is their interest)?

So the prevalence of fraud could be similar to that in many other fields?

Reply on news.ycombinator.com