Variables in the field of substance abuse tend to be highly positive skewed (e.g. number of drinks in a tipical session, total AUDIT Scores), when data comes from community samples. Since traditional correlation techniques (Pearson, Spearman) have limitations with this type of data, what could be a more adequate analysis technique for assessing the strenght of the association between variables?
Spearman's p or rank correlation only needs the data to be random and independent of others to be used, unlike Pearson correlation coefficient which insists on bivariate normal relationship and linear relation, in addition to 'randomness and independence of the data'
Essentially, Spearman is okay to use for highly skewed data.
You may want to look at copula models. Some of copula functions are extremely useful for investigating tail dependence, e.g. Clayton and Gumbel.
https://www.jstatsoft.org/article/view/v021i04/v21i04.pdf
Weidong
- Badges
- Science topic
- Similar topics
- Correction