0

There is a set of numerical and categorical independent variables and one dependent variable that is categorical. Some of the categories are far to be related with the rest, but most of them are very related between them, and the difference between one category and another is based on small nuances.

Then, there is some way to measure how related are the categories, among them, in the dependent variable??

juanmah
  • 101
  • 3
  • What kind of correlation? Pearson / Spearman are defined only for continuous variables. – user2974951 Aug 30 '19 at 11:31
  • That's my question – juanmah Aug 30 '19 at 11:35
  • There is no real correlation measure between nominal values. – user2974951 Sep 03 '19 at 05:57
  • I disagree, think that the categorical dependent variable is a color. It could be Red, Green, Brown, Yellow and Aquamarine, Turquoise Blue, Sky Blue, Pacific Blue, Cerulean, Navy Blue, Denim, Indigo... The blue nuances are more related between them than the other colors (this is real). And if a yellowish color is predicted as Cerulean is worst than a Turquoisish color is predicted as a Aquamarine (my eyes can not notice the difference). There is a way to 'measure' the wrongness of the prediction error? – juanmah Sep 03 '19 at 06:27
  • You can say that certain colors are more likely to appear together than others, but this is not correlation. You could think of this in terms of frequency, whether the frequencies between colors are different, in which case you could use something like a Chi-square test. – user2974951 Sep 03 '19 at 06:30
  • Correlation: a mutual relationship or connection between two or more things. It is. Don't get confused with an specific type of correlation coeficient, that has it's own assumptions and boundaries. This specific correlation coefficient could no be applied to this case. I'm asking for the existence of some type of, let's say coefficent (or not), adequate to the case I'm presenting. – juanmah Sep 03 '19 at 06:39
  • 1
    Have a look at https://stats.stackexchange.com/questions/108007/correlations-with-unordered-categorical-variables?noredirect=1&lq=1 – user2974951 Sep 03 '19 at 06:48

0 Answers0