Spurious correlations--causes and cures.

Abstract

Although the Pearson Product--Moment Correlation Coefficient is one of the most widely used statistics in the health and behavioral sciences, it is not always appreciated that the critical assumption of bivariate normality underlies its interpretation. When variables have marginal distributions which are skewed or have heavy tails which produce outliers, correlations may be either spuriously large or small. Having diagnosed problems through exploratory data analysis, one must take the appropriate corrective action, such as re-expressing (transforming) variables or selectively discarding discordant observations.