Correlation between dichotomous and continuous variables
I'm using the operator "Correlation Matrix" on rapidminer, which I believe uses Pearson Correlation, and the operator is able to calculate correlations for every variable type, including binominals (dichotomous) and polinominals.
I would like to know: how exactly is the operator calculating the correlation for example between a binominal and a numerical attribute? Wouldn't a pearson correlation only allow numerical variables? Does it simply convert binominals to 0 and 1, or is it doing something else?
Thanks in advance,
Filipe G.B.
Tagged:
0
Best Answer

Telcontar120 Moderator, RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,635 UnicornI believe it is doing sequential integer coding for any nominal attributes. This is of course highly questionable for polynominal data in terms of correlation interpretability, but for binominal data it does make sense.1
Answers
Hey,
RapidMiner uses internally a mapping to integers for all nominal types. This mapping is used for the corellation. It's somewhat statistical not too good. That's why we throw a problem if you do it.
~Martin
Dortmund, Germany
Thank you for you answer
Hi, I am a new user in RapidMiner
Actually I have 31 attributes with 10K of instances.. I want to make correlation matrix in order to make a relationship between the attribute. The problem is I have many types of data which is nominal, polynominal and numerical data..May I know what are the process of correlation matrix for many types of data?
As explained previously in this thread, typically "correlation analysis" only applies to numerical variables. What would you expect the correlation coefficient for nominal data to tell you?
If you want to use nominal data with correlation, you are better off recoding it as a series of binominal/dummy variables first.
Lindon Ventures
Data Science Consulting from Certified RapidMiner Experts