I have been working on some data in collaboration with a statistician. As the data was not normally distributed a log-transformation was performed. However, it was explained to me that we cannot report effect sizes of our findings in the original units because of this transformation. We therefore resorted to using percentages.
I need to explain the reason for this in simple terms to collaborators (who are in the clinical field, not statisticians). Does anyone have suggestions of how to do this?
The communication problem here seems two-fold.
You seem to be implying that your statistician colleague advised percents, but on the face of it that makes no sense here; if you applied a logarithmic transformation, then an exponential transformation would reverse that, but secondary corrections might also be in order. Presuming your statistician is competent, I think we need much more detail on what was done to make sense of it.
Conversely, what is your problem with your collaborators? That they don't understand percents? That they would not understand exponentials?
It seems that you are missing the scope here for generalized linear models with logarithmic link which would automatically yield predictions on the scale of the response and obviate any need even for back-transformation, let alone for secondary corrections.
