Random Sample Assumption Violation and Chi-Square/Reg: Ordinal and Nominal Data

#1
Hello Everyone,

I'm a grad student in biostatistics and am currently working at my first job with real world data.

I'm analyzing survey data accessing employee's comfort with their work environment (DV) by several characteristics (IVs) (e.g., sex, race, religion). The DV, Comfort, is measured using a likert scale, with 1= Very Comfortable and 5=Very Uncomfortable. The DVs are coded categorically (nominal).

I've been running Chi-Square analysis and Fisher's exact (when the cells are less than 5). I've also done logistic regression analyses.

My question is regarding the assumption that the sample was randomly selected. In this case, the survey went out to every employee and had 40% response. Does this fact violate the random sample violation? If so, should I be using another procedure, such as Spearman rank?


Thank you,

Brenda
 

CB

Super Moderator
#2
The DV, Comfort, is measured using a likert scale, with 1= Very Comfortable and 5=Very Uncomfortable. The DVs are coded categorically (nominal).
Hiya,

A Likert scale would usually be treated as at least ordinal, not nominal. It'd probably be more useful to select tests that treat it as such. E.g., Kruskal-Wallis test if you're looking at the relationship between your DV and a nominal predictor.

My question is regarding the assumption that the sample was randomly selected. In this case, the survey went out to every employee and had 40% response. Does this fact violate the random sample violation?
Yes, but it's very rare that one has access to a true random sample. It would be fairly conventional to treat the sample as random, but mention this limitation when discussing your findings.

If so, should I be using another procedure, such as Spearman rank?
Spearman's rank correlation could be suitable if you are correlating your ordinal DV with another ordinal variable. The choice between Spearmans coefficient and the others you mention doesn't really have anything to do with the random sampling assumption, though.
 
#3
Thank you so much for responding! It was really helpful.

(For the record, I misspoke: The DV, Comfort, is ordinal. The IVs (predictor variables) are nominal).