I am going to log transform a scale due to extreme skewness; I am using RStudio for my analyses (linear regression). Some of the values computed are "-Inf" or infinite values. To address this, I was thinking of adding 1 to my log10 code: y =log10(x + 1). Previous posts indicate that this would address my problem. Are there specific guidelines for reporting my results? Do I have to "back transform" anything during interpretation?
Asked
Active
Viewed 46 times
0
-
Before you do this, you might want to study your options and their potential effects on the results. See https://stats.stackexchange.com/questions/30728 *inter alia.* There's a huge difference between transforming the explanatory and response variables, too: the answer to your question depends on which variable you transform. – whuber Apr 20 '18 at 14:37
-
Also note that back-transformation will be impossible, which greatly reduces the interpretability of the result. – Knarpie Apr 20 '18 at 15:16
-
@Knarpie Could you elaborate? Since the function $x\to \log_{10}(x+1)=y$ is explicitly invertible and surjective, you must have in mind something different than just finding $x$ in terms of $y.$ – whuber Apr 20 '18 at 15:42
-
Yes sorry, I am referring to the interpretability of the regression coefficients in function of the original variable – Knarpie Apr 22 '18 at 19:22