Box-Cox or Yeo-Johnson transform as a preprocessing step to FAMD

[Dylan Mann-Krzisnik]

Hello FactoMineRers,

I am currently experimenting with FactoMineR's FAMD model for mixed categorical and quantitative data. I remember reading that FactoMineR commands will ensure a centred unit-variance normalization of the data.

I was wondering whether performing power transforms such as Box-Cox or Yeo-Johnson was also advised as a preprocessing step? I tend to use these transforms on quantitative variables when the data is non-Gaussian, but perhaps this is not advised for FAMD?

Also, if power transforms are advised, should they be performed prior or following data imputation?

Thank you,

Dylan