Data

Hey fellow data nerds, thief is trying to log-transform a set of right-skewed target variables for linear regression, suddenly had this question about what is the log's base. Gemini says it's a natural log (base e)?Does it matter?

Defi. Ethereum. Physics.

Data at @quotient.

Nah, they’re all a constant away from each other so it doesn’t matter. Natural is fine. @monteluna is the guru and can confirm

OK there's *some* situations where it does matter in extreme skew cases. The issue is a base 10 log transform isn't powerful enough, so its only good in cases where your data is spread over 3 or more orders of magnitude.

I just looked up this article that shows a good example and might help.

Aviator, diver, Etherean, nexialist, PhD in aviation and climate science, host of /diving /ethfinance /forteantimes /geopolitics /science /singapore /thomas

my takeaway from that article is based on "how skewed" that your raw data is, the appropriate transformation function varies:

Yeah, also I don't know why Box-Cox isn't listed here.

https://www.statisticshowto.com/probability-and-statistics/normal-distributions/box-cox-transformation/

OK there's *some* situations where it does matter in extreme skew cases. The issue is a base 10 log transform isn't powerful enough, so its only good in cases where your data is spread over 3 or more orders of magnitude.

I just looked up this article that shows a good example and might help.

https://anatomisebiostats.com/biostatistics-blog/transforming-skewed-data/