I've been trying to run some stats to check for correlation between an ordinal categories (such as body conditions classes) and a continuous variable (such as body measurements in cm). I'm really struggling to find a test that I can run that is suitable for this type of data. I have paired measurements from each individual entered in the spreadsheet.

Example hypothesis to test: Does a higher body condition index, tend to result in greater head length measurements?

I've tried: *Linear regression (but don't think this is appropriate given the x-axis variable is ordinal).

*Ordinal regression (but although the model fit is usually highly significant (<0.001), the data being analysed often seems to fail Pearsons goodness of fit test.

Multinominal regression (but Pearson's goodness of fit was highly significant (<0.001) so probably also not suitable to use.)

Principle coordinate analysis (but as I'm only using 2 variables this doesn't seem to be the correct test either).

I considered a spearman's rank order test, but that seems to require 2 ordinal values rather than just one.

I'm kind of out of ideas. Can anyone help please? (I know this should be easy, but I don't seem to be able to find a test that fits.) I've been using SPSS.

Thank you in advance.

  • $\begingroup$ Welcome to Cross Calidated! Ordinal regression with the ordinal variable as $y$ was my first thought. Could you please explain your objection about the failed test? $\endgroup$
    – Dave
    Oct 11, 2022 at 5:26
  • $\begingroup$ Hi Dave. When I ran an ordinal regression on the data, it gave a significant Pearsons goodness of fit test (P<0.05). I from what I could gather, you need a non-significant Pearson's test in order for the ordinal regression results to valid I think? $\endgroup$
    – Jaq
    Oct 11, 2022 at 5:45
  • $\begingroup$ No unfortunately. They are recommending tests that require ordinal data on both scales rather than ordinal/ continuous which would be more ideal if it is available. (It will be hard to turn the continuous data into ordinal data in a way that is justifiable.) $\endgroup$
    – Jaq
    Oct 11, 2022 at 12:23
  • 2
    $\begingroup$ You don't have to "turn" a continuous variable into an ordinal one: it already is ordinal. The issue is that tests exploiting only its ordinal character might be less powerful than other tests. $\endgroup$
    – whuber
    Oct 11, 2022 at 14:02