Meg recently wrote a post acknowledging that crying in science was pretty common place. It really touched a nerve and went viral. Meg’s opening syllogism was masterful: humans cry, scientists are human, therefore scientists will cry.
I want to touch on an even more sensitive syllogism: humans make mistakes, scientists are human, therefore scientists will make mistakes. And a corollary – some mistakes will make it into print.
People obsessed with preserving a united front against science deniers might try to pretend this isn’t true. But it is true. This rarely acknowledged truth about scientists is fresh in everybody’s minds because of a recent retraction of an ecology paper (due to an honest mistake). I’m not even going to link to it since it is a distraction from my main point to single out one group of individuals when I’m talking about collective responsibility (but if its too distracting not to know Jeremy linked to it on Friday).
What I am finding revealing is not that a retraction occurred but other people’s reactions to the fact that a retraction occurred. There seems to be a lot of distancing and blaming. The first commentor on retraction watch even went one step further and very sloppily and inaccurately started throwing around the phrase “fraud scandal” (really? the topic of mistake is so taboo we can’t differentiate the profound difference between mistake and fraud?)
My reactions were rather different, going in order of occurrence, and probably progressively more profoundly were:
- Ouch – I feel bad for the authors
- I’m impressed with the way the authors handled this – it took a lot of courage
- That’s science working the way it is supposed to
- It could have been me
There’s no need to expand on the first one (except its worth noting I don’t know any of the author’s personally so this was more of a 1 degree removed member of my community form of empathy).
But I think it is worth dwelling on the second one for a moment. It must have been very tempting to bluster and deny that there were substantive enough mistakes to require a retraction and hoped this faded away. We all know this strategy has a decent shot at working. In an infamous case in evolution (UPDATE the link in Jeremy’s post is broken – follow this link), it worked for years until a co-author took it upon himself to self-publish and blow the whistle (nobody talks about this but the journals have an obvious interest in not highlighting a mistake). But these author’s didn’t weasel in any fashion. And they thought about the good of science before the good of their careers. Good for them!
As for the 3rd reaction – this is not a failure of science. It is a success of science! It is science working as it is supposed to. And it is exactly why science has a claim to a degree of rigor that other modes of thought don’t have. The reason my syllogism doesn’t eliminate science as a paragon of correctness is that – contrary to the popular view about lone geniuses – science is not about individuals or single papers. It is about the community and the total body of evidence. One individual can be right, wrong, a crack-pot, a genius, mistaken, right for the wrong reasons, and etc. But the community as a whole (given time) checks each other and identifies wrong ideas and mistakes. The hive mind will get the important things right with some time. If you read the details, this is exactly what happened. Good for science!
The last reaction is the touchiest of all (it could have been me*). Of course I do not knowingly have any mistakes in print. But I could have a mistake out there I don’t know about. And I’ve caught some that came close. And I could make one in the future. Should I be thinking that? Should I be admitting that in a public blog? I sure hope your answer to both of these questions is yes. If I’m not asking the first quesiton (and admitting the possibility) how can I be putting my best effort into avoiding mistakes. The same for the community context. And I’m pretty sure any other honest scientist cannot say they are 100% sure they never had made a mistake and never will make a mistake. 95% sure – I hope so. Maybe even 99% sure. But 100% sure? I don’t trust you if that is what you claim. Every lab I’ve ever worked in or been close to (meaning dozens) have challenges and errors with data and coding and replicability of analysis. Most of them are discovered and fixed (or sadly prevent publication). But has anybody here ever run an analysis, gotten a particular t-statistic/p-value and written it up, and then run the analysis later and gotten a slightly different number and never been able to recreate the original? Anybody have one or two sample IDs that got lost in the shuffle and you don’t know what they are? These are admittedly small mistakes that probably didn’t change the outcome. But it is only a difference of degree. And I bet most of you know of bigger mistakes that almost got out the door.
I want to speak for a minute more specifically about coding. In this day and age nearly every paper has some coding behind it. It might just be an R script to run the analyses (and probably dropping some rows with incomplete data etc along the way). But it might be like the stuff that goes on in my lab including 1000+ line computer simulations and 1000+ line big data analysis. Software engineers have done a lot of formal analysis of coding errors. And to summarize a lot of literature, they are numerous and the best we can do is move asymptotically towards eliminating them. Getting rid of even 90-95% of the errors takes a lot of work..Even in highly structured anti-error environments like NASA or the medical field mistakes slip through (like the mis-transcribed formula that caused a rocket to crash). And science is anything but a highly-structured anti-error environment (and we shouldn’t be – our orientation is on innovation). In a future post, I will go through some of the tricks I use to validate and have faith in my code.. But that would be a distraction here (so you might want to save your comments on how you do it for that post too). The bottom line though is I know enough software engineering not to fool myself. I know there are errors in my code. I’ve caught a couple of one line mistakes that totally changed the results while I was in the middle of writing up my first draft. I think and hope that the remaining errors are small. But I could be wrong. And if I am wrong and made a whopping mistake, I hope you find my mistake!
The software industry’s effort at studying errors was just mentioned. But the medical and airline industries have recently devoted a lot of attention to the topic of mistakes as well (their mistakes are often fatal).The Institute of Medicine released a report entitled “To Err is Hman” with this telling quote:
“.. the majority of medical errors do not result from individual recklessness or the actions of a particular group–this is not a “bad apple” problem. More commonly, errors are caused by faulty systems, processes, and conditions that lead people to make mistakes or fail to prevent them.”
Broad brushing the details, both medicine and the airlines have come to the conclusion that the best way to avoid mistakes are to 1) destroy the myth of infallibility, 2) eliminate the notion that raising the possibility of a mistake is offensive, 3) introduce a culture of regularly talking about the possibility of mistakes and analyzing mistakes made for lessons learned, and 4) make avoiding mistakes a collective group responsibility.
I think arguably science figured this all out a couple of hundred years ago. But it is worth making explicit again. And per #3 it is worth continuously re-evaluating how we’re doing. In particular we do #4 extremely well. We have peer review, post-publication review (which is stronger for prominent and surprising results), attempts at replication etc. We’re professional skeptics. We also do pretty well at #2; you expect and accept your work being criticized and picked apart (even if nobody enjoys it!). #1 is more of a mixed bag. I’ve heard a lot of “it could never happen in my lab” comments recently, which is exactly the myth of infallibility. And the same for #3 – I haven’t yet heard anybody say “I’m going to change X in my lab” in response to the recent incident. And more generally across #1-#4, I would suggest that coding is novel enough in ecology that we have not yet fully developed a robust set of community practices around preventing coding errors.
In conclusion, I am sure somebody is going to say I am glorifying mistakes in science. I’m not. Mistakes* are unfortunate and we all need to (and I think all do) put a lot of effort into avoiding them. But I sincerely believe there is no way to guarantee individual scientists do not make mistakes. At the same time, I also sincerely believe that a well constructed scientific community is robust enough to find and correct all important mistakes over time. Which means it really matters whether we respond to mistakes by finger pointing or examining our common culture and how to improve it. The later is the conversation I want to have.
*Probably important to reiterate here that I’m talking about mistakes, not fraud. Whole different kettle of fish. I presume most people can see that, which is why I am not belaboring it.