Log scale/transformation for sensory testing

Mechanical perception has been shown to follow the Weber-Fechner law, which states that the perceived intensity of a mechanical stimulus is proportional to the logarithm of the stimulus strength.

According to Mills et al (2012), it follows that when we test mechanical thresholds, we should be representing the withdrawal thresholds logarithmically, rather than linearly as most researchers do. Definitely check out this paper. It’s well worth a read.

I now represent my PWTs in log(g) (with linear coordinates), rather than just as grams. I find that it makes my data tighter, and more importantly, as Mills argues, it makes the data more consistent with a normal distribution, which is a key underlying assumption of parametric stats (t-tests, ANOVAs, etc.). Truly, if we use PWTs in grams, we should be using non-parametric tests, although many pain papers don’t do this.

So thinking logarithmically about PWTs has multiple benefits. What do you think? What do you do?

@jmogil, @fmoehring @ram_kandasamy

1 Like

Hey Alex,

I definitely agree, our lab uses log (g) for mechanical stimuli. We are also a behavioral pharmacology lab so the X axis can often be dose which is also represented on the log scale, but I also think it is a better representation of the data. Initially, I was using a linear scale but I have since switched over for the reasons you and the paper have mentioned.

However, linear grams is still probably the most common in the literature.


1 Like

I agree with the LegakisL and achamess. We are also using log scale in our lab to show PWT (von Frey filaments) and LegakisL pointed out the most important reasons: normal distribution that allowed the use of parametric tests. Unfortunately, there are some referees that do not understand and in some revisions they ask to present the absolute
values. We just published a paper (Sci Rep. 2016 May 27;6:26955. doi: 10.1038/srep26955) in which we presented both scales to attend the referee.

1 Like

@thicunha Thanks so much for sharing your experience! And excellent paper. I was excited when it came out since the MPL model looks simple and robust and reproducible. I hope to try it out soon.