Correlation test. Spearman's or Pearson's?

#1
Hello guys!
I am a poor MA student trying to survive statistics!

I have the following problem:

The participants of a questionnaire were asked to rate the quality of

a)education
b)Intervention
c)Support

having as choices: not very useful, a little bit useful, useful , very useful.

In the analysis, the authors turn these phrases in numbers from 1 to 4 and wanted to see if these 3 ratings were correlated.

They used Pearson's correlation test.

And my question is! Aren't these data ordinal? If yes, shouldn't they have used Spearman's test, which is suitable for ordinal data, while Pearson's is for interval/ratio?

Thank you in advance!!
 

CowboyBear

Super Moderator
#2
Yes. This is commonly done, but Pearsons may not be the best choice here. Another relevant option is a polychoric correlation (correlation between the two variables, assuming that a normally distributed latent variable underlies each, with this latent variable disected into ordered categories by the rating scale).
 

noetsi

Fortran must die
#3
Depending on the data, from my limited experience, it may not matter if you run pearson, spearman, or polychoric, but its worth running all three and see how they differ. Unfortunately, unless you are using R, major software such as SPSS and SAS do not have polychoric analysis built into their system. SAS has a macro that does this, SPSS has an option to use R.
 

noetsi

Fortran must die
#5
Personally I don't think its is a very positive element of a software that they have to rely on another software to work:p