My outcome variable includes six categories but I'm not sure whether they count as "fully ordered". The categories measure the extent of added value of a certain intervention: 1 (major added value), 2 (considerable added value), 3 (minor added value), 4 (non-quantifiable added value), 5 (no added value), 6 (less value).

My concern is, however, that category 4 (non-quantifiable) is defined as showing that there is added value but that it cannot be quantified because of a lack of data, i.e. the benefit could be either major, considerable or minor. It doesn't mean that category 4 is smaller than category 3 for example.

Would it still be possible to use an ordered logit model given that one of the six categories doesn't really fit in the order? Or can the ordered logit model be discarded even without having tested for proportional odds?

  • $\begingroup$ I would suggest a two-part model. One that models the distinction between quantifiable and non-quantifiable benefit, and conditional on being quantifiable, you can model the degree of impact. You can find examples of such models in Chapter 11 here. You might well find that quantifiable or non-quantifiable model is non-informative, in which case you can simply drop category 4 and model the second part only. $\endgroup$ – tchakravarty Aug 12 '16 at 8:31
  • $\begingroup$ Great thank you for your reply and the link to the book. I will look into that! $\endgroup$ – Statsquestions1 Aug 12 '16 at 20:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.