Testing for an effect of a treatment on an ordinal score beyond Kruskall-Wallis...

[SBuc]

Dear Brain trust,

I want to determine the effect of a treatment on a score of pain in animals. Many published literature has used GLM models using score as a continuous dependent variable accounting then for treatment, block and animal repetition using linear mixed model.

however, this approach is per se assuming that the score is a continuous variable (which is not) and also that a score of 8 is 2 times worst than a score of 4 which is not demonstrated (so the interval between 2 unit scale can not be assumed to be the same throughout the scale (0-15 max points).

initially we used non parametric tests such as rank comparison but I want to know if a more elegant and powerful analysis can be performed.

thanks

[Reeza]

You can try an ordinal regression. 

The documentation has a fully worked example:

https://documentation.sas.com/?docsetId=statug&docsetTarget=statug_logistic_examples03.htm&docsetVer...

Under the title there's a link to the full code that you can run on your machine and then follow through. Once you understand the example, you can replicate it on your own data. 

---

@SBuc wrote:

Dear Brain trust,

I want to determine the effect of a treatment on a score of pain in animals. Many published literature has used GLM models using score as a continuous dependent variable accounting then for treatment, block and animal repetition using linear mixed model.

however, this approach is per se assuming that the score is a continuous variable (which is not) and also that a score of 8 is 2 times worst than a score of 4 which is not demonstrated (so the interval between 2 unit scale can not be assumed to be the same throughout the scale (0-15 max points).

initially we used non parametric tests such as rank comparison but I want to know if a more elegant and powerful analysis can be performed.

thanks

---