My LinkedIn home page has been flooded with posts from a big machine learning group I joined on a whim yesterday. And dear lord, I don't know whether to weep for the state of the world or be more optimistic about my job prospects.
The latest post involves wild excitement over a 'cheatsheet' about how log transformation is a 'hack' that fixes general model problems.
Say I have some historical data e.g., past stock prices, airline ticket price fluctuations, past financial data of the company...
Now someone (or some formula) comes along and says "let's take/use the log of the distribution" and here's where I go WHY?
Questions:
WHY should one take the log o...
I'm wondering if it makes a difference in interpretation whether only the dependent, both the dependent and independent, or only the independent variables are log transformed.
Consider the case of
log(DV) = Intercept + B1*IV + Error
I can interpret the IV as the percent increase but how does...
@mkt: do we need something similar to our meta thread on "unbalanced data" for the log transform? stats.meta.stackexchange.com/q/6349/1352 Perhaps we could collect a couple of good questions carrying both the "data-transformation" and the "logarithm" tags and ask the moderators to slap a "faq" tag on them - that would make finding them easier.
@StephanKolassa Yes, I think that would be worthwhile! There's quite a few of them. I VTC'd a couple as dupes of stats.stackexchange.com/questions/576504/… but it isn't comprehensive. And there is good info across some other threads.
I'm wondering if it makes a difference in interpretation whether only the dependent, both the dependent and independent, or only the independent variables are log transformed.
Consider the case of
log(DV) = Intercept + B1*IV + Error
I can interpret the IV as the percent increase but how does...
