How to decide if my phenotype is skewed and which transformation (log, sqrt, etc.) to use?

Theory & background
1

  1. How do you decide if a variable is “skewed enough” to need a transformation (e.g. any rule of thumb for skewness values or visual criteria)?

  2. For left‑ vs right‑skewed data, which simple transformations are usually recommended (e.g. log, square‑root, reflection + log), and how do you choose between them?

  3. With typical GWAS sample sizes, is it acceptable to leave the phenotype untransformed if residuals look “okay”, or would you still recommend something like an inverse normal transformation?