Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Is there a general principle on whether one should compute pearson correlation for two random variables X and Y before taking their log transform or after? Is there a procedure to test which is more appropriate? They yield similar but different values, since log transform is non-linear. Does it depend on whether X or Y are closer to normality after log? If so, why does that matter? And does that mean that one should do a normality test on X and Y versus log(X) and log(Y) and based on that decide whether pearson(x,y) is more appropriate than pearson(log(x),log(y))?

share|improve this question
@vinux has a nice answer, and provides an informative link for understanding the role of normality in correlation. I just wanted to point to this question: which is very good for understanding what logs do in regression. – gung Feb 13 '12 at 17:46

Correlation(pearson) measures a linear relationship between two continuous variables. There is no such choice for (X,Y) or (log X, log Y). Scatter plot of the variables can be used for understanding of the relationship.

The following link may answer regarding normality issue. link

share|improve this answer

Because $\log(X)$ and $\log(Y)$ are monotonic transformations of the data $X$ and $Y$, you might also choose to use Spearman's rank correlation ($\rho_S$) and not worry about transforming your data, as you would get $\rho_S(X,Y) = \rho_S(\log(X),\log(Y))$

share|improve this answer

Pearson correlation is for parametric testing and is more powerful than non-parametirc test. Thus, we opt to use transformation before any non-parametric procedures. Transform your data and get pearsons correlation. That's it.

share|improve this answer
@ abi: Depending on sample size, Spearman's and Kendall's coefficients are relatively similar in terms of power and MSE to Pearson's with normally distributed data, and are far superior with even slight data contamination. – Patrick Aug 24 '13 at 3:21

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.