log-transformation of values < 1.0 to meet normal distribution for a mixed ANOVA

NilsPixa

Hi all,

I want to conduct a mixed ANOVA, but my data does not meet the normal distribution assumption at every time point included in the ANOVA, so I decided to perform a log-transformation to reach the normal distribution. However, I realized that I have variables in my dataset containing values smaller than 1.0 (e.g., 0.3, 0.7, etc.), which might be a problem for log-transformation. Does anybody has any advice or can help with how to deal with log-transformation of values < 1.0?

Many thanks

Nils

EJ

Hi Nils,

You can first add a 1, and then do the transform. Or you could multiply all values x by a constant c, so that the smallest x gives cx>1; you then have log(cx), which is log(c) + log(x) and the constant should not affect your inference.

Also, there are of course other transformations you could use, see this paper:

Ratcliff, R. (1993). Methods for dealing with reaction time outliers. Psychological Bulletin, 114(3), 510–532. https://doi.org/10.1037/0033-2909.114.3.510

Cheers,

E.J.

NilsPixa

Dear E.J.,

Thank you very much for your answer and the reference to Ratcliff. I will go for it!

Best

Nils