1

I have ordinal data on three IVs ranging from 1 to 5 as below:

IV1: Not at all Important - Very Important

IV2: Not at all Satisfied - Very Satisfied

IV3: Performs much Worse - Performs much better

The data is not normally distributed and I want to perform some parametric tests (I'm aware of the problems on doing parametric tests on ordinal data).

Is it considered OK to normalize this kind of data?

Nick Cox
  • 48,377
  • 8
  • 110
  • 156
GentlemanEddie
  • 147
  • 3
  • 10
  • 2
    Limiting yourself to parametric tests needs an explanation. And normalizing ordinal data will not help when the variable is discrete. – Frank Harrell Aug 28 '14 at 12:07
  • Thanks Frank. I read (http://www.theanalysisfactor.com/can-likert-scale-data-ever-be-continuous/) that one could run parametric tests on ordinal data in case some conditions are filled. One of these conditions is that one has normality in the data. But perhaps I've misunderstood something. – GentlemanEddie Aug 28 '14 at 13:44
  • 1
    That's odd. When a variable is discrete it is impossible for it to be Gaussian or to be transformed to have a Gaussian distribution. Any why not use statistical methods that are dedicated to ordinal data? – Frank Harrell Aug 28 '14 at 14:11
  • I'm not familiar with Gaussian and a real rookie in statistics. I have done an ordinal logistic regression, but I find it very hard to interpret the outcomes. Especially since I've also imputed the data for some missing variables. I cannot make any sense of the Pooled Location coefficients. – GentlemanEddie Aug 28 '14 at 14:21
  • 1
    Gaussian = normal. What outcomes do you need help interpreting? Which ordinal model did you use? How did you model the IVs in that model? What is the frequency distribution of $Y$? – Frank Harrell Aug 28 '14 at 14:27
  • If your data are only in a few categories, then after a transformation they'll all still be with the same observations that were with before. What does that achieve, other than changing one set of arbitrary distances between categories to another set? – Glen_b Aug 28 '14 at 23:59
  • Clearly I need to rethink the analysis here.. Thanks! – GentlemanEddie Aug 29 '14 at 12:15

0 Answers0