3

My dependent variable is categorical (with 3 levels) and my predictor variables are a mix of continuous (age in months, test 1 score, test 2 score, test 3 score) and categorical (gender). I believe I should run a multinomial regression, but when I do the results are really uninterpretable because each age is basically treated as a category (and the same for the test scores).

Does anyone know a different test that would be appropriate to use in my case? I would really appreciate any help.

Andre Silva
  • 3,080
user32544
  • 31
  • 1
    You have ordered categories; if you take account of the ordering, you may find the results more interpretable. – Glen_b Nov 09 '13 at 01:42
  • Thank you Glen. Does this sound like the appropriate test to use? – user32544 Nov 09 '13 at 01:46
  • I'm not sure what you refer to by 'this'. I meant things like proportional odds models, adjacent category logit, cumulative models, and so on. – Glen_b Nov 09 '13 at 01:53
  • See the answers here and here for some of the possibilities – Glen_b Nov 09 '13 at 01:58
  • 1
    Adressing the question of categorical treatment of age and scores: Did you find out how to let your software treat them as numerical? – Michael M Nov 09 '13 at 11:34
  • 1
    Sounds to me like you are treating age, a continuous independent variable (unless the range of months is small) as categorical. I don't know of any methods for ordinal IVs. – Peter Flom Nov 09 '13 at 11:50
  • You should tell us what is your categorical dependent/response variable, without knowing that we are blind. And you could model age using splines! – kjetil b halvorsen Mar 17 '18 at 13:26

1 Answers1

2

First, if your DV is ordinal (that is, has an order, e.g. "small", "medium", "large") then a better starting point is probably ordinal logistic regression.

Second, the way SPSS is treating age and the test scores as categorical has nothing to do with nominal logistic regression. You somehow told SPSS that these variables are categorical. You need to tell it that they are numeric, or continuous, or whatever. I don't know SPSS so I don't know how to do this.

Peter Flom
  • 119,535
  • 36
  • 175
  • 383