Edward,

this is more of a hunch than a well-thought-out argument.

When you treat the data as nominal instead of ordinal, you are making less stringent assumptions than the data would allow. In other words, you are making things more difficult for yourself, which, in my mind, is fine. [The opposite would be to assume an interval level of measurement, which would be problematic.]

I guess the question is: Are you losing anything in terms of the inferences you are able to make? Or are you merely making data analysis more cumbersome?

Hello Edward,

I don't think the issue is whether ignoring ordinal nature of a scale "restricts analyses." I think it is, do you lose information that would potentially change your interpretation based on the analysis?

The answer, I suspect, is idiosyncratic to the specific variables, distribution(s), and relationships observed in a data set. In other words, you could run simulations to determine empirically what happens with a specific data set (or data system). To me, that would be about the only definitive way to answer the question.

For you, in your situation, if treating variables as nominal is genuinely the only way to elicit an analysis, then you simply acknowledge that as a potential limitation of the study when you report and interpret results.

Good luck with your work.

Potentially great, but certainly inflated SEs.

Hope this helps,

Matt