Last year I published a paper with my friend Johan on the evils of log transformations in the analysis of count data. We went for the subtle title “Do not log-transform count data“. Now, I was aware that this wasn’t the only statistical sin perpetrated by ecologists (and other scientists), and there’s quite a seam to be mined if only you know a bit more than non-specialists.
Evidently, Dave Warton has decided to mine the same seam. And how can you resist this title for his new paper?
The arcsine is asinine: the analysis of proportions in ecology
If I had it on, I’d take my hat off to salute him.
I have mixed feelings about these sorts of papers, Bob. One the one hand, I really like them, in a confimation bias sort of way. Let’s approach data analysis properly if we want to understand the natural world as best as we can. I’m a big fan of logistic regression as well. It’s my ‘every problem is actually a binary outcome in disguise’ hammer.
On the other hand (possibly the one holding your hat), I get a bit annoyed that these sort of papers need to be published. We’re supposed to be trained scientists, for gawd’s sake. Why don’t we think about these things before going for a black-box statistical approach? And why do so many papers using these
fawltyfaulty methods slip through the review process and still get published, requiring us to read papers that remind us what we’re doing is wrong? Sheeesh. I’m away to do a t-test on some ordinally ranked data.
Damn you Bob, I’d just seen the asine/asinine paper and was mustering the energy to blog about it, and you’ve beaten me to it! Why can’t you just stick to the county cricket blog?!
Mike – I kind of agree, but it is bewildering I guess if you’re not really interested in the stats, just performing the analysis as a necessary evil to get your work published – you’ll look at what everyone else does, or what’s suggested in your (ageing) copy of Zar or whatever, and do that. I kind of suspect that, if the specific transformation you use massively changes your conclusions, you probably shouldn’t read too much into them anyway. But yes, we should be doing things correctly, and it’s useful as an editor / reviewer to be able to point authors to this kind of paper, as well as others pointing out the folly of, for e.g., using residuals as data, or using stepwise variable selection methods…
Mike – I feel you pain. Or would do if it wasn’t an easy way of getting papers published. But it is difficult to keep up with new techniques. As Tom writes, everyone uses their old copy of Zar or Ranta, without being aware that these books were written based on what their writers learned as students.
I wonder how we can break this cycle – would another textbook really help? Especially as it would mean presenting new ideas that teachers wouldn’t want to use because they think their way is the right way.
What do you think I was doing whilst waiting for the start of play?
That’s an excellent point, Tom. I’m now going to go and check I haven’t used any residuals as data in any of my papers.
Otherwise, I’ve recently pointed out elsewhere that we should strive to maintain scientific integrity, which means keeping up to date with statistical approaches, even (especially) if this means your data doesn’t give you than answer you’d like to get from it. I’m afraid I can’t sympathise with a scientist who is unwilling to understand why they are analysing their data the way they are. Keeping up with statistical techniques is as important as keeping up with other methodological techniques. If nothing else, they can collaborate with a statistician in the design phase of a project, which has the associated benefit of keeping people like Bob off the streets at night. And during the day.
I do agree with you Mike. A particular bugbear of mine is authors who basically do the analysis right, then present it conducted in the wrong way too, ‘because that’s the method people usually use’ – although I suspect that such things often arise because of the requests of referees… So while I can understand why people may use wrong / out of date methods, I completely agree that they shouldn’t, and I think that part of the job of a well-functioning peer-review process is at some stage to point out the error of their ways. And ideally to suggest that they ask Bob how to do it properly…
Do you ever get out, Bob?
Bob, the DOI link to your paper is broken. Is it possible to fix it? I would like to read your paper. Of course, I could look you up on PubMed (I’d do that if the link can’t be fixed), but this way is easier… 🙂
Ah, thanks. It should work now.