I have four questions representing one variable called work family conflict. They are all ordinal. And I added all the items and made it one variable.

After adding them up, I got code ranging from 1 to 16. Now 16 means high conflict and 1 means low. I have to to run the regression, how can I do it. Work family conflict is independent variable and my dependent variable is self reported health which is a likert scale.

  • Should I recode and categorized it to high, medium and low and run the ordinal regression?
  • Or can I run the ordinal regression without categorizing?
  • Is it right to run the ordinal regression when dependent variable has likert scale?
  • $\begingroup$ You can do it both way. Your dependent variable is ordinal and you have a categorical independent variable. $\endgroup$
    – SmallChess
    Feb 2, 2017 at 8:47
  • $\begingroup$ If you have enough cases then leave it as 0 to 16 (you did mean that, not 1 to 16?). Why waste data? $\endgroup$
    – mdewey
    Feb 2, 2017 at 14:12
  • $\begingroup$ I am not confused about whether it should be 0 or 1, I am confused as to how I can run the regression with. If I run the regression I will get value for each from 1 to 16 but I don't want that. I want one value that is for the conflict. $\endgroup$ Feb 2, 2017 at 14:21

1 Answer 1


Summing or averaging ordinal variables is debateable, but proceeding anyway, ordinal regression (PLUM) gives you a threshold coefficient for each level, but you still get only one coefficient for your independent covariate.


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.