I'm new to data analysis and I've been messing it up by using Pearson's correlation coefficient to measure possible relationships between different variables in my data set, both dependent and dependent, discrete and continuous.

Today, I read that using Pearson's on categorical data is a no-no unless there are only two categories, and that those categories must be 0 or 1.

Then I read that using the correlation ratio $\eta$ is good for categorical variables with greater than 2 classes.

Then, I read that I have to consider whether the variable is nominal or ordinal before I choose how to measure relatedness.

What are the best practices for assigning a measure of relatedness to the following types of relationships?

  1. continuous-nominal
  2. continuous-ordinal
  3. nominal-nominal
  4. ordinal-ordinal
  5. nominal-ordinal

A little bit of theory is welcome as I'm not a stats guy and I'm always interested in diving below the surface. However, I'm mostly looking for a concise answer that allows me to move forward in my analysis.


I found a white-paper that outlined the answer to my question, with examples of specific use cases in the medical field. It was very easy to understand.

  1. continuous-nominal (Dichotomous) - Point Biserial
  2. continuous-ordinal - Kendall's $\tau_B$
  3. nominal-nominal (Both Dichotomous) - $\phi$ coefficient
  4. nominal-nominal (Non Dichotomous) - Goodman's $\lambda$
  5. ordinal-ordinal - Kendall's $\tau_B$
  6. nominal-ordinal - Rank Biserial

I hope this helps someone else.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.