1
$\begingroup$

A product is rated 1 for poor and 10 for excellent. Does it strictly fall into either category - quantitative or categorical?

(Although it seems like it is categorical, I was wondering if treating it as quantitative might be helpful in terms of not having an excessive number of columns and perhaps increasing accuracy)

$\endgroup$
1
  • 1
    $\begingroup$ It is ordinal. In that, it differs from categorial variables and from any for which you could say the difference between 5 and 6 is just as big as between 4 and 5, etc. $\endgroup$ – Michael Hardy Dec 23 '18 at 18:08
1
$\begingroup$

It's neither nominal nor numerical. It's ordinal. An example for a nominal variable is sex. It can be male or female and there is no connection between them. You want to encode nominal variables as one hot. That way you specify that there is no dependence between the different values. A typical example for an ordinal variable is a size of a t-shirt. It can be either S, M or L. But there is a dependence between them in a way that M is bigger than S and L is bigger than M. So you want to encode it as 0, 1, 2 for S, M and L correspondingly. The good thing about your case is that it's already done for you. So the only thing you want now is to normalize the variable to scale the values to be between 0 and 1 and you are good to go.

$\endgroup$
3
  • $\begingroup$ It doesn't make sense to normalize the values like that, because, as you say, it is ordinal and not numeric. M is not halfway between S and L. $\endgroup$ – The Laconic Dec 24 '18 at 3:10
  • $\begingroup$ It does make sense. What if you have N numeric variables with range from 0 to 1 and 1 ordinal variable that is not normalized. Then you encounter the same issue as having non normalized numerical variables. $\endgroup$ – gorjan Dec 24 '18 at 17:01
  • $\begingroup$ I think my point wasn't clear. I'm not saying it doesn't make sense to normalize a variable in general (although there is not an "issue" with having non-normalized numerical variables in general either). I'm saying it doesn't make sense to treat an ordinal variable as if it is an integer. It's ordinal, not numeric. It needs to be turned into some numeric variable or variables, but not by simply mapping the values to 0, 1, 2, etc. Whether you normalize it or not is beside the point, and might not be necessary in any case. $\endgroup$ – The Laconic Dec 24 '18 at 17:10

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.