1

I have a very basic question, but I am having difficulties finding an answer.

What would be a simple measure of dependency between a categorical and a numerical variable?

For two numericals it could be a linear correlation coefficient, for two categoricals - mutual information. But I don't want to discretize the numerical variable, or convert the categorical variable into numerical.

inzl
  • 1,183
  • 8
  • 18
  • 1
    Is your question different from a number of posted already, such as [this](http://stats.stackexchange.com/q/20022/3277) or [this](http://stats.stackexchange.com/q/73065/3277). By the way, which of the two variables - nominal or interveal - is the dependent one? – ttnphns Dec 15 '14 at 09:16
  • Thanks for pointing out these answers. If I understand correctly, computing "eta" coefficient requires discretization of the numerical variable, that is, converting into intervals. I am interested, if there is a simple measure that would work with raw numerical variable. – inzl Dec 15 '14 at 09:40
  • The numerical variable is the dependent one. – inzl Dec 15 '14 at 09:41
  • Eta is for numeric as the dependent, and requires no discretization of it. – ttnphns Dec 15 '14 at 10:29

0 Answers0