Questionnaire Analysis with Likert Scale: Using PCA to Assess Differences in Subpopulations

[Barney1998]

Good morning everyone,
I need to analyze a questionnaire, essentially revisiting a study after some years.

My issue is that all variables are on a Likert scale. The categories from which I need to draw conclusions have been predefined. By that, I mean I've been told that certain questions correspond to a person's value, others to their knowledge, etc. The purpose of the research is to examine how subpopulations I have—such as men/women, smokers/non-smokers—differ.

Initially, I thought of conducting an explanatory factor analysis to get scores for these categories, but I couldn't get the desired categories the way I wanted. So, I thought of doing the following: Is it correct to conduct PCA separately for both categories to obtain the desired scores and proceed with the analysis?

Thank you very much for your time.

[Ksharp]

Check PROC PRINQUAL and its examples in DOC .

[Ksharp]

https://blogs.sas.com/content/iml/2014/11/07/distribution-of-blood-types.html

[Barney1998]

Thanks for answering!
Definitely, this article does not solve my problem. The question was if can procced via PCA as I told above.

Burney

[Ksharp]

Answer is Yes.

Did you check PROC PRINQUAL and its examples ?

[Ksharp]

Another way is using PROC CORRESP,but it is more suited to non-Likert scale.
Check its example .

[PaigeMiller]

So, I thought of doing the following: Is it correct to conduct PCA separately for both categories to obtain the desired scores and proceed with the analysis?

First, I agree with the advice from @Ksharp 

If I was to do PCA (or PRINQUAL or CORRESP) separately for both categories to see how the categories differed, I would want to look at the loadings, not the scores, to determine differences between the groups.

Another possibility (and one I recommend here) is to fit one PCA (or PRINQUAL or CORRESP) model to the entire data, then plot the scores (not the loadings) and see if the groups overlap or are separated. To determine what variables are associated with the "separation", you can then use Contribution Plots (for example, as in PROC MVPDIAGNOSE), see Miller, P., Swanson, R. E., and Heckler, C. H. E. (1998). “Contribution Plots: A Missing Link in Multivariate Quality Control.” Applied Mathematics and Computer Science 8:775–792. By the way, that first author is me.