[Bruce Weaver]

It's strange that you can't view the article--I'm seeing it as freely available even when looking at it from home (outside the university domain). In any case, the authors have also uploaded it to ResearchGate:

• https://www.researchgate.net/publica...ons_in_ecology

[Joseph Coveney]

the authors have also uploaded it to ResearchGate

Thanks, Bruce. I guess that it was the lock-down browser settings; I could download it from there eventually.

Anyway, to me a transformation is a transformation, and I'd grant that the logit transformation might have advantages in interpretability some contexts, but I have my doubts about such an advantage in the OP's. In planning for the experiment, I would have probably looked into inverse-normal transformation over logit (and would have probably checked aligned ranks instead of straight Friedman's) for test size verification and power analysis. But most likely, I would have remembered the arcsine transformation from my days with Winer's second edition, and have favored that. The method's old enough that there are recommendations of what to do with zeroes and ones that you can point to as authoritative in defending what you have done in a particular protocol.

In my experience with health-related quality-of-life visual analog scales on paper, there are typically a bunch of marks at around 2 or 3 cm, a big bunch at around 5 cm, another bunch at around 7 or 8 cm, and a handful at—or within a millimeter or so—of the zero- and 10-cm ends (in addition to the duplicate marks and the circles, Xs, slashes and smudges that aren't on the line at all). So, to use a logit transformation as in that article, you'd need to decide what to do with the 0% and 100% responses, and from what I saw when scanning the article, the logit transformation is a little sensitive to what you choose to do.