Dear colleagues,

I am trying to determine the appropriate statistical test for my problem. I have read several discussions on the topic here, however was unable to extract an answer, and so I hope you can help me:

We have performed a questionnaire-based observational study in ~200 patients, using likert-scales and categorical/nominal items. All patients are in one group, the study design ist hypothesis-generating. Our analysis now wants to compare different items to one another, within this group of patients.

All questionnaire items are independent to each other (i.e. there were no repeated measurements). However the patient's likelihood of answers for some of the items may be dependent (not sure about this, please correct me), example: We asked patients for treatment goals of HbA1c and LDL-c, and their likelihood of a correct answer may be dependent on patient knowledge, participation preference etc.; from the researcher's perspective, there is a high likelihood of contingency.

My thoughts so far are:

1. Both, Likert-scales and categorical/nominal data are non-parametric, thus non-parametric tests should be used.

2. All patients are in one group, thus I cannot design a classical contingency analysis (Chi-Square, Fisher) for categorical/nominal data, where one variable is group assignment (2x2, 2x3, 2x4, depending on data). However, I could design a contingency table analyzing, whether interdependence of answers to different items (2x2 for binary items, 4x4 for 4-point-liked-scaled items)

3. I am unsure of whether I have to correct for possibly interdependent answers somehow.

Questions:

1. How do I appropriately compare the distribution of 4-point-likert-scaled ordinal data within this group? -> my best guess would be Mann-Whitney-U (non-parametric, non-dependent, ordinal data); an alternative is a 4x4 contingency table with Chi-Square/Fisher

2. How do I appropriately compare the distribution of 2-point binary/3-point nominal data within this group? -> my best guess would be 2x2 (McNemar?) resp. 3x3 contingency table with Chi-Square/Fisher

3. Do I have to correct for dependency of answers? If yes, how?

Thank you very much in advance ;-)

Best regards,

George

More Georg Wolff's questions See All
Similar questions and discussions