What should be the variable type for: ratio variables and ordinal variables?

Curious

1) e.g. percentage of population
2) customer rating (doesn't setting the type to polynominal lead to loss of information of hierarchy?)

SGolbert

Hi,

I'm happy to see you posting again, but nonetheless I would like to ask you to specify a bit more on your posts. We need context and whenever possible sample data. And if you have a sample process, that would be a great way to get the help started!

Regards,

Sebastian

Telcontar120

Without sample data it is hard to know for sure, but a priori it seems that a percentage or ratio should be attribute data type numeric/real and a rating should probably be converted to a numeric/integer (this may involve transformation on your part first if you have it as labeled text in the raw form).

kypexin

Hi @Curious

customer rating (doesn't setting the type to polynominal lead to loss of information of hierarchy?)

I would suggest that you follow a simple rule, if you can directly compare or order different values on 'bigger-smaller' scale, than it should be of a numerical type. Which is most likely the case with a rating, as one can be higher than the other. Also keep in mind that variable type matters in case the variable is a target (label), which you are trying lo predict. In case of numerical label you face regression problem, while with polynomial label it's classification problem, which means different models should be used.

Curious

For instance, if we have variables: 

% of lower status of the population (with values e.g. 12.43) 
proportion of non-retail business acres per town (with values e.g. 7.87)
average number of rooms per house (with values e.g. 6.575)
index of accessibility to highways (with values e.g. 1,2, 3, 4, 5)

Thank you!

kypexin

Hi @Curious

Here all variables are logically numerical, first three being real and last one integer.