When is it appropriate to use a Kendall over a Spearman correlation?
I have categorical variables and non-parametric continuous variables. I am running a correlation matrix on the variables. I am leaning toward opting for a Kendall correlation. I have extensively read many forums and articles about which to use. It seems the Kendall is typically used for smaller datasets but gives a "more conservative" value and confidence interval. Is there a reason this would not make sense to be used?
Hello Afton,
If all of the categorical variables have two levels, I'd use the Spearman.
If one or more of the categorical variables has three or more levels, then you'd have to compute many correlations to capture all pairs using a method like Spearman (or Kendall's tau). In this instance, Kendall's coefficient of concordance is usually easier (and requires only one computation per variable pair).
It is important to note that the nature/distribution of the Spearman rho is qualitatively different from that of either of the Kendall indices. So, don't compare values across methods.
Good luck with your work.
In the normal case, the Kendall correlation is preferred than the Spearman correlation because of a smaller gross error sensitivity (GES) (more robust) and a smaller asymptotic variance (AV) (more efficient). If you are interested in other cases, you may compute their GES and AV by yourself.
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers