I've been working through some regressions, and I was told by a credible source to not worry about transforming the predictor(s) but be sure that the univariate response values are distributed close to Normal and of course check normality of residuals.
I'm perplexed because I've been following the examples of the book "An R Companion to Applied Regression" (John Fox, 2018) who has emphasized transforming both the response and predictor(s). I've been using the Box-Cox transformation method for multivariate normality on my predictors which seemed to help but now I'm not sure if it's the appropriate thing to do to produces relevant results.
There's a similar question here (10yo) that doesn't really have an answer so I'm giving it another shake here.
TLDR: multiple (predictor) linear regression -- Transform univariate response and predictors or just the response and maybe scale/center the predictors?