I'm currently working on the findings of my Masters dissertation where I've defined hypotheses, created a questionnaire, and then testing whether a relationship exists between different ordinal variables. Many of the hypotheses are based upon the more than one question, which I've computed together (unit-weighted) to provide a holistic value. Regarding the strength of the relationships, I'm using Spearman's Rho (Correlation) to perform this task on the individual questions and computed variables.

However, I've included some demographic variables in the question such as an age grouping and education level, etc. with the intention of providing somekind of demographic insight to the ordinal responses received. An example would be a single Likert question 1-5 scale, is there an insight based upon Age or education level; alternatively with the computed variable (unit-weighted from 2 or 3 Likert questions), is there an insight based upon Age or education level.

My question is whether it is possible extract should demographic data using SPSS in this manner and what functionality should I be using?

Thanks in advance.


So ultimately the nature of my question is how can I extract demographic associations from two ordinal variables using one or more dependant values. My limited understanding has come to the following...

  • MANOVA sounds like an appealing direction but I'm using ordinal / non-parametric variables, thus not normally distributed. In using a computed variables of one or more ordinal values unit-weighted, I could use the MANOVA (assuming the data is normally distributed) to determine interesting demographic associations.

  • Ordinal regression sounds like a possible direction using the predictive results to explicitly identify associations.

  • Visually inspecting cross-tabs between specific questions either on individual ordinal variables or one dependent ordinal variable and one dependent computed variable.

  • Using filter each demographic, i.e. Age group (16-25, 26-35, etc.) and perform another Spearman correlation to measure the strength of the relationship. This feels clunky.

Unfortunately, I've hit a brick wall in determining the appropriate test / method thus I'm asking something of an open ended question towards what test / method people would recommend given my circumstances...

  • 1
    $\begingroup$ I can't follow your question. Can you try rephrasing it? $\endgroup$ – gung Jul 27 '15 at 13:39
  • $\begingroup$ Is that any better? $\endgroup$ – SeanCocteau Jul 27 '15 at 15:44
  • $\begingroup$ That does help (I voted to reopen). I think some general advice can be given now. But this is still rather vague & software specific. It's been a long tome since I've used SPSS & I don't have access to it, so I don't know what "select cases" is, eg (I suspect this is true of most on CV). You would do well to further clarify your situation & make this software neutral. What kind of thing is "an insight based upon Age", eg? $\endgroup$ – gung Jul 27 '15 at 16:20
  • $\begingroup$ Consider it removed. Insight based upon Age, an example would be the relationship between two ordinal variables split by the another ordinal variable; so if the responses are split by age, are older people selecting 'strongly agree' to the two ordinal questions. I guess in a simplier example would be "Do you like chocolate?" and "Do you like fruit?" both on a 1-5 Likert, is there a way to analyse the responses by age to retrieve insights such as "Younger people like Chocolate but not Fruit", "Older people like Fruit not Chocolate" $\endgroup$ – SeanCocteau Jul 27 '15 at 20:30
  • $\begingroup$ @SeanCocteau it may be useful to follow the criteria outlined in the analysis wiki. Read it. You should, for instance, start by defining your measures as well as the question of interest. $\endgroup$ – AdamO Jul 28 '15 at 18:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.