# SPSS twostep cluster analysis to use with ordinal variables

How do you distinguish between ordinal (ordered) and non-ordinal (nominal) categorical variables when setting up a SPSS Two-step cluster analysis? The procedure seems to have only one undifferentiated input field "categorical variables".

Example of an ordinal variable: 1=low 2=medium 3=high.

• It isn't clear to me exactly what you're asking for here - is it how to know substantively which variables are ordinal and which are not? Or is this about how to use SPSS? The latter is not outside the scope of our site - see our help center for more details. Commented Oct 9, 2016 at 15:20
• TwoStep cluster analysis does not allow ordinal variables. By "categorical" in means nominal only. Consider using your ordinal ones as if they are continuous (which wouldn't great sin for, for example likert type questions). Or refuse doing TwoStep and do Hierarchical cluster, if you number of cases isn't huge. Or you might take the trick to convert your ordinal data first to inteval via optimal scaling of Categorical PCA procedure, then input to TwoStep with clear conscience. Commented Oct 9, 2016 at 15:22
• @Silverfish, this Q is both about SPSS and the method as it exists there. Commented Oct 9, 2016 at 15:24
• @ttnphns In your opinion is there is an underlying statistical issue here, rather than just an SPSS one? I think your edit makes the situation rather clearer (and have retracted my close vote). Commented Oct 9, 2016 at 15:30
• @Silverfish, I think the Q might be kept, because it asks about a data type suitable for a statistical method, although it was worded as SPSS specific question. Commented Oct 9, 2016 at 15:32

I would put it differently. TWOSTEP certainly allows ordinal variables, but it treats them the same as nominal variables. They are simply assumed to be multinomial. As ttnphns says, though, treating ordinal variables as continuous might be satisfactory Clustering is a rather ad hoc technique. You might also want to consider using a Support Vector Machine (SVM) in unsupervised mode via the STATS SVM extension command. That procedure distinguishes between ordinal and nominal scales, although unsupervised mode is mainly useful for anomaly detection.

You can install this extension from the Utilities menu in Statistics 22-23 or the Extensions menu in V24. It requires the R Essentials.

• This is the confirmation I was looking for: " TWOSTEP certainly allows ordinal variables, but it treats them the same as nominal variables. They are simply assumed to be multinomial". Thank you! Commented Oct 10, 2016 at 17:19