4 ms·
I've also investigated bad science papers before, and blogged about it a few times. There's a lot out there. There's nothing special about getting a PhD that ma
by mike_hearn 9y ago
I've also investigated bad science papers before, and blogged about it a few times. There's a lot out there. There's nothing special about getting a PhD that makes you politically neutral or morally superior. The temptation to abuse maths to make your personal views seem "scientific" is irresistible for too many. My last public attempt:
https://blog.plan99.net/did-russian-bots-impact-brexit-ad66f08c014a https://blog.plan99.net/did-russian-bots-impact-brexit-ad66f...
I feel some sympathy with fellow Brit Nick Brown after reading this story about his work:
https://www.theguardian.com/science/2014/jan/19/mathematics-of-happiness-debunked-nick-brown https://www.theguardian.com/science/2014/jan/19/mathematics-...
(as an aside, totally unsurprised the guy works in computing, developing an intolerance for abuse of logic is a job hazard.)
But I don't know if this perspective is helpful to be honest. "Don't be a dick" is both quite an obvious principle and also rather useless when criticising people's work: some of the authors, and some of the people who want to believe the research is true, will inevitably view any criticism as being dickish. This is doubly true if you take the obvious next step and speculate as to whether the issues are mere oversights or (far more commonly, in my view) deliberate deceptions intended to further an agenda.
One problem is that genuine, honest mistakes tend to get picked up already by the existing peer review process. The really bad papers that get through are usually bad in support of some wider social mission, usually leading the public to some policy goal, and they don't get struck down because the peer reviewers share the same goals. Thus pointing out mistakes has no effect, because they already knew the science was bad. The only thing that can work is pointing out to the people who they're trying to influence that something has gone wrong.
There is a responsibility to be harsh in these cases. Failing to do so can simply let the issues fester and compound. From the Guardian article on Brown:
There were several psychologists, versed in non-linear dynamics, who smelt something fishy about the maths in the published paper. Stephen Guastello, from Marquette University, wrote a note of mild complaint to American Psychologist, which it chose not to publish because "there wasn't enough interest in the article". Guastello feels now that he should have been more forceful in his opinions. "In retrospect," he says, "I see how I could have been more clearly negative and less supportive of what looked like an article that could move the field forward if someone would follow up with some strong empirical work."
The story describes Brown's first debunking of a paper that used lots of clever looking math to reach obviously absurd conclusions about psychology. One author of the paper admitted that she had never understood the maths and the person who did create the maths (Losada) has refused to ever respond to the debunking in any way. All the reviewers were successfully intimidated by the maths and were unwilling to criticise it, allowing Losada to get away with it:
John Gottman, a leading authority in the psychology of successful relationships, wrote to Losada because he couldn't follow the equations. "I thought it was something I didn't know about, because he's a smart guy, Losada. He never answered my email," he says.
So the critics weren't critical enough and a paper based on pure mathematical bullshit racked up 350 citations. Vast amounts of time, money and effort was wasted. What does this say about the scientific process?
By the way, in case anyone thinks computing is immune to this sort of thing, it's not. Papers about "Russian twitter bots" are a current streak of bad science in computing, for obvious reasons: the authors want to undermine political trends they don't like. The last one I read claimed to have constructed a neural network that could detect Russian bots with 99%+ accuracy. Nowhere in the paper did it give any examples of what the network detected, nor what the ground truth they were comparing to was. When I emailed the authors to ask for the data so the study could be replicated, I was told the data was not available (despite being just a bunch of tweets i.e. public data).
It was literally an entire paper that looked scientific but boiled down to "we made a magic black box that without fail finds evidence of a vast conspiracy, but we can't show you. Trust us."
- AlexCoventry 9y agoI don't understand why this is being downvoted. It seems to be the most insightful, empirically grounded, high-effort comment in the thread.
- billforsternz 9y agoI noticed the same thing (and upvoted to push back). I am not 100% sure but I think there's a trend on HN whereby good comments are often being downvoted. I need to recheck the guidelines (on mobile now) but I thought downvote should be applied rarely. For example I don't downvote if I disagree, even when I am actively arguing in a thread. I reserve downvoting for trolling and other obviously bad faith comments.
- shaki-dora 9y agoWhen an article on how not be seen as a crank makes you want to post a long comment about your research into Russian bots, you should probably think again.
- ScottBurson 9y agoThis is exactly wrong. It's not having an opinion on a controversial topic that makes one a crank, it's whether that opinion is well supported and respectfully expressed. In fact, mike_hearn's first link is to a blog post where he debunks claims of evidence that Russian bots influenced the Brexit vote. (He doesn't offer evidence that that didn't happen; what kind of evidence could exist? But he does argue that the alleged evidence offered by others that there was such influence is unconvincing.)
- mike_hearn 9y agoIt's become a cliche, but that kind of low effort insult is exactly how people like Trump win. You can't argue your position so you resort to calling people who challenge you "cranks". The lazy assumptions that so often pervades these debates leads directly into over-enthusiastic trust of pseudo-science. I've been out all evening, and my post is now at +11. But if it was getting downvoted I am 99% certain it's because a lot of people really want to believe in the whole Russia/Trump/Twitter story and can't stand anything that challenges it, even incidentally. It meets their pre-existing biases and grants a feeling of moral and mental superiority: the other tribe might be brainwashed by something as trivial as a retweet, but there's no way I'd be like that. But that's the whole point of this blog post we're discussing: lots of people have very, very strong incentives to believe in pseudo-science dressed up as the real thing. That doesn't just include scientists themselves but also journalists, politicians and so on. The psychological paper that Nick Brown did a takedown of was cited by hundreds of papers and led to lucrative speaking careers. The underlying equations that made it seem like science vs mere opinion weren't connected to anything and boiled down to nonsense (a problem I've seen several times). Yet it took enormous efforts to do anything about it, and in the end the scientists who got caught are unrepentant: one of them didn't even bother to respond and the other agreed that the maths was nonsense but doesn't believe it really matters. The same thing has happened repeatedly in recent years. See the story of Amy Cuddy: her 'power pose' research was thoroughly debunked but she simply refused to accept it, despite her co-authors accepting that the research was wrong. She continues to milk the speaking circuit and has even been cited by the BBC as an example of an "inspiring woman", although her career is arguably built on deception. At this point, if you believe in Russian twitter bots influencing elections, why shouldn't you be the crank? Where is the evidence? All claims I've seen on this topic fall apart under examination. Even the premise is absurd - it relies on the belief that millions of people are so politically ungrounded their vote can be shifted merely by seeing a spammy retweet. That is a very dubious belief. But if you believe you have a reliable study that proves the existence of armies of bots capable of swinging elections, please do show it. I will enjoy examining it. However, it is not sufficient to merely point at high-status individuals and delegate to them, because - as this entire discussion shows - bad "science" is capable of spreading and even becoming a consensus far too fast.