Hi! I'm asking what are hopefully relatively simple questions - I'm just getting in a muddle over small details in how to treat and report "prefer not to say" responses.

I recently conducted a survey comparing two groups, for all questions (3 likert questionnaires and demographic qs), I included a "prefer not to say" option.

To my understanding, "prefer not to say" responses are user-entered missing values that holds more "meaning" than a value that's missing because a user skipped the question and allow researchers to differentiate between active non-responses and errors/laziness. So, I understand should not disregard them and lump them in with missing values.

Therefore, I was planning on reporting the % of missing and the % of "prefer not to say" responses for each individual question in the appendix?

Based on informal sources, I was then going to label "prefer not to say" as a missing value/category in SPSS (e.g., -9) and perform some sort of imputation technique.

My main issue is that I cannot find a clear reference to justify this process - or is it a such a standard procedure that a justification is not needed? I'm also confused because does this not mean that I am treating these active nonresponses like nonactive nonresponses? Or, is reporting the % of prefer not to say responses beforehand enough information for the reader?

Secondly, what am I to do when particular questions have a numerically much greater % of "prefer not to say" responses (e.g.., ~20% responded with "prefer not to say" for this question when the next highest is ~3.5%)? I understand this indicates that there is something wrong with the question, and looking at it this is true. There is also a numerically big difference in % between my two groups for this question (e.g., 20% vs 1.4%), and this difference in response makes sense given differences in the groups - the question does not apply much to the 20% group, indicating that these responses are not because of participants suddenly wanted more privacy but because they weren't sure how to respond with the other options provided).

But how should I test this significantly? I've read that I should use dummy variables for the responders versus nonresponders (who answered "prefer not to say" AND missed the question?) and test for differences between the groups, but I was hoping for more clarification on this procedure? As in, are there any established protocols or numbers to use as guides? I'm assuming I explain in the "instruments" section that upon testing for differences this question was taken out (because it was deemed a "poor" question)

I'm also assuming I should test this for all questions where people have picked "prefer not to say", not just for the ones where it's clear something's fishy!

Apologies for this large wall of text - I've been going around in circles and feel I am missing something that is glaringly obvious!