You have anominal scale data so you use methods for nominal scales. If your sample size is less than 30 you use non-parametric methods, if it is more than 30 you use parametric methods.

Appropriate is Chi-square.. its quite easier and would do justice to your work..

Many people are usually using T test especially to compare responses derived from the likert scale, don't bother doing such, what you arrive at is not appropriate..

In my institution, if it was a Phd or Post Doc work, I would suggest Mann and Withney's U test to get the interaction between the Variables.. Hope this helps you..

Dear Izzah,

Parametric tests are not appropriate for your study since your variables are ordinal and nominal.

Use non-parametric tests like chi-square, Spearman's test, Mann-Whitney U test, Kruskal Wallis test etc where appropriate.

Regards!

Eric

Idowu Goodnews Oluwaseyi Thank you for the response! I have a sample of 300 participants and according to Beatrice, if the sample size is less than 30 you use non-parametric methods if it is more than 30 you use parametric methods. So in this study, would Chi-square would still be appropriate?

i believe it is appropriate as long as the individual observations of samples are independent.. which means that the occurrence of one individual observation (event) has no effect on upon the occurence of any other observation in the sample under consideration.

Also, you can also make a 'GOODNESS FOR FIT TEST' too

You can use parametric statistics on the Likert items if your sample size is that large. However, first you should closely examine the responses to each item (or scales thereof) for their distributions. If the distributions are too skewed, parametric statistics might not be the best choice no matter the sample size. The categorical choice item you describe violates the assumption of independence mentioned above, since respondents can choose more than one item; I would be cautious about using anything other than just reporting frequencies of the different choices--going with just descriptive rather than inferential statistics. But it's hard to get specific unless we know what your hypotheses are and what sorts of patterns you might be interested in exploring in the data set.

As previously stated it is difficult to mention exactly the ideal way to analyze the data since for me it is not clear what is your "dependent variable or outcome of interest" vs covariates that may impact this outcome.

Ordinal regression models can be used for analysing Likert scale as dependent variable (seen Bürkner, P. C., & Vuorre, M. (2019). Ordinal regression models in psychology: A tutorial. Advances in Methods and Practices in Psychological Science, 2(1), 77-101. for an indepth discussion of the topic using R and Stan software).

On the other side traditional Chi-squared like tests can also be reported but would answer to a different question.

If trying to determine some participants specific characteristics they share based on the answers profiles, principal component analysis -like methods would also be possible (eg to find some specific "clusters" within respondents).

A good descriptive analysis is also important!