To be honest I'm surprised it's not higher. Relevant discussion: <a href="https://news.ycombinator.com/item?id=17789308" rel="nofollow">https://news.ycombinator.com/item?id=17789308</a><p>The bottom line is that there's not much incentive for doing the boring statistical validations (I can tell you that <i>no one</i> likes doing statistics apart from statisticians, and not even all of them) and verifying that everything is reproductible, and a <i>huge</i> incentive for, let's say, 'arrange' a thing or two so that the paper looks better. So many people just kinda do it, and it slips through the cracks because:<p>-In many fields, the reviewers are not statisticians themselves<p>-No one really bothers to download the data, setup the environment and libraries, and reproduce the exact steps taken to obtain the very same figures in a paper. Which is understandable given how doing all of that can be such a chore. Anyway, most of the time the exact steps aren't even described. No, jupyter notebooks aren't a solution either (it would take too long to explain why and the post is already long).<p>-In many cases, the results turn out to be true anyway so people don't notice they were initially put forward with fraudulent validations<p>-When they turn out to be false, people just shrug and move on with what's actually true. Bogus results often fail to stand the test of time and get forgotten quickly despite initial hype. No one bothers to say: "Hey, that paper from 8 years ago is bullshit and their authors are hacks!" because nobody cares.<p>-There's a huge psychological barrier to actually call out one of your peers and affirm that they're an impostor and their work is bogus. Especially when said impostor happens to be a big name in the field, that part of you still doesn't believe they would commit fraud, not to mention the social repercussions and backlash of doing such an accusation. We scientists aren't a very confrontational bunch.<p>So most of the time it kinda works and we're all bumbling along hoping to find some modicum of truth at the end of the road with minimal harm done. But of course sometimes you get these guys who kick off their entire career on a high profile, much hyped fabricated result (Woo-Suk) or even a <i>series</i> of bogus papers (Sato), and <i>that</i> may lead to long-term harm. The good news is that hyped papers or very prominent figures quickly attract scrutiny from their peers, and sooner or later reality catches up with them as labs around the world fail to replicate their 'breakthroughs'.<p>All in all, I'd say we're doing fine. We're just not the ethereal source of truth that some people hold us to be, the very same people who, after claming that 'God is dead', are very quick to replace Him with His sillicon-based equivalent around which we would act like priests, except with lab coats in lieu of clerical garments.