We regularly get questions about log transforming dependent variables, which in some quarters appears to be seen as a magic silver bullet that solves all problems, because apparently, "everything is normally distributed after log-transforming" (which, to be explicit, is wrong, and not even useful). Here are a few useful prior threads in this vein (contributed by User1865345 and mkt):
- When (and why) should you take the log of a distribution (of numbers)?
- In linear regression, when is it appropriate to use the log of an independent variable instead of the actual values?
- Interpretation of log transformed predictor and/or response
- Interpreting log-log regression with log(1+x) as independent variable
As in this analogous question on imbalanced data, it probably does not make sense to attempt to merge them into the One Question To Rule Them All As A Canonical Duplicate. However, I would propose that we collect a couple of useful threads on this topic here (threads tagged both "data-transformation" and "logarithm" might be a good start) to use as duplicate targets, depending on the nuance of the newly posted question, and ask the moderators to slap the faq tag on them.
