I am analysing a dataset containing variables of different types: continuous, ordinal and categorical. To prioritise in which order to analyse the variables, I would like to evaluate the relatedness of all of these variables to one another.

In other words: is there a measure that takes two variables of any type and results in a number where a higher number indicates the variables are more related.


The analysis is a first step to modelling the data, so the relatedness should measure how well one variable would be as a predictor in regression or classification towards the other variable.

Work so far

I came across covariance analysis, which has different approaches depending on the types of the dependent and independent variable. I was not sure if the resulting $R^2$ can be used to order the variable pairs for my means. Secondly, Mutual Information seems to work for all variable types and could therefore be interesting, however I have not found an implementation of this method that would work for any variable type.

Additional information

We work in Python so measures implemented to work with Numpy are preferred.


Your Answer

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

Browse other questions tagged or ask your own question.