Style manuals sound like they ought to be boring things, full of arcane points about commas and whatnot. But Wikipedia’s style manual has an interesting admonition: Be bold. The idea is that if you see something that could be improved, you should dive in and start making it better. Don’t wait until you are ready to be comprehensive, don’t fret about getting every detail perfect. That’s the path to paralysis. Wikipedia is an ongoing work in progress, your changes won’t be the last word but you can make things better.
In a new editorial at Psychological Science, interim editor Stephen Lindsay is clearly following the be bold philosophy. He lays out a clear and progressive set of principles for evaluating research. Beware the “troubling trio” of low power, surprising results, and just-barely-significant results. Look for signs of p-hacking. Care about power and precision. Don’t confuse nonsignificant for null.
To people who have been paying attention to the science reform discussion of the last few years (and its longstanding precursors), none of this is new. What is new is that an editor of a prominent journal has clearly been reading and absorbing the last few years’ wave of careful and thoughtful scholarship on research methods and meta-science. And he is boldly acting on it.
I mean, yes, there are some things I am not 100% in love with in that editorial. Personally, I’d like to see more value placed on good exploratory research.* I’d like to see him discuss whether Psychological Science will be less results-oriented, since that is a major contributor to publication bias.** And I’m sure other people have their objections too.***
But… Improving science will forever be a work in progress. Lindsay has laid out a set of principles. In the short term, they will be interpreted and implemented by humans with intelligence and judgment. In the longer term, someone will eventually look at what is and is not working and will make more changes.
Are Lindsay’s changes as good as they could possibly be? The answers are (1) “duh” because obviously no and (2) “duh” because it’s the wrong question. Instead let’s ask, are these changes better than things have been? I’m not going to give that one a “duh,” but I’ll stand behind a considered “yes.”
* Part of this is because in psychology we don’t have nearly as good a foundation of implicit knowledge and accumulated wisdom for differentiating good from bad exploratory research as we do for hypothesis-testing. So exploratory research gets a bad name because somebody hacks around in a tiny dataset and calls it “exploratory research,” and nobody has the language or concepts to say why they’re doing it wrong. I hope we can fix that. For starters, we could start stealing more ideas from the machine learning and genomics people, though we will need to adapt them for the particular features of our scientific problems. But that’s a blog post for another day.
** There are some nice comments about this already on the ISCON facebook page. Dan Simons brought up the exploratory issue; Victoria Savalei the issue about results-focus. My reactions on these issues are in part bouncing off of theirs.
*** When I got to the part about using confidence intervals to support the null, I immediately had a vision of steam coming out of some of the Twitter Bayesians’ ears.