I have a dataset mostly consisting of single items and I have been analyzing them with ordinal regression and other ordinal analyses. However, I also have a likert-type scale and in theory I would like to include it as an independent variable in analyses with ordinal independent and dependent variables. Another researcher told me that it is possible to essentially categorize the responses to likert-type scales in order to treat is an ordinal variable, for example like this:

  • Scale average 1-1.99 becomes 1
  • Scale average 2-2.99 becomes 2

And so on. In this way, responses to the scale become a single ordinal variable.

My hunch is that this is not appropriate and that I should find another way. Or does anyone know if this approach is actually valid?


1 Answer 1


Why do you want to make it ordinal? Ordinal independent variables are hard to deal with. (Ordinal logistic regression can be a good option when the dependent variable is ordinal).

You apparently have an average (or other combination) of some Likert items. Since you were willing to take their average, you are treating them as continuous. You can then treat the average as continuous. And that's what I would do.

  • 4
    $\begingroup$ Ordinal independent variables are easy to deal with in Bayesian models using the R brms package. But whatever you do don’t lose any of the information in the original variables by binning them. It’s better to use quadratic models in the predictors, even with not quite interval-scaled predictors, than losing that information. $\endgroup$ Feb 19 at 13:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.