Welcome to the Society for the Suppression of the Correlation Coefficient
by Thom
Most of what is known about the history Society for the Suppression of the Correlation Coefficient comes from a few remarks by John Wilder Tukey (Tukey, 1954) who wrote that he was “an early member of [Charles P. Winsor’s] informal society for the suppression of correlation coefficients – whose guiding principle is most correlation coefficients should never be calculated.” Charles P. Winsor was an important influence on Tukey and best known in statistics for the form of sample of trimming known as winsorising.
The aim of this blog is to further the aims of the society and educate about the dangers of unfettered application of the correlation coefficient. I like to think that this follows the spirit of the original informal society by spreading a serious message in a gently humourous way.
References
Tukey, J. W. (1954). Causation, regression and path analysis. In O. Kempthorne, T. A. Bancroft, J. W. Gowen & J. L. Lush (Eds.), Statistics and mathematics in biology (pp. 35-66). Ames, IA: Iowa State College Press.
So, what is the replacement you propose ?
Good question. In general you want to use an unstandardized coefficient e.g., a regression coefficient in the original units. This is obvious in some situations – but the question becomes difficult when you want to compare effects using completely different measures on different scales. Sometimes there may be insufficient information to make a sensible comparison, other times you can convert to a common scale (depending on what you want to do), and yet other you can make reasonable assumptions to make the comparison. This could involve a lot of work or require you to collect more data – but the problem isn’t solved (just hidden) by using standardized coefficients.