I think my question speaks for itself.

It may, however, consist of three parts:

1) Can we consider a likert scale as count data?

Based on my search so far, I would say we can probably not consider that because a likert scale has an underlying scale this does not make sense. I think that for this particular question, the discussion of ordinal vs interval properties for likert scale is not relevant (but please let me know if this matters for assumptions for negative binomial regression).

2a) Can a likert scale dependent variable be analyzed as a continuous variable?

I think this would be a fair argument. Not an expert on this, but as long as long as you meet the assumptions for your regression model it should be fine (see also http://www.theanalysisfactor.com/can-likert-scale-data-ever-be-continuous/ & https://link.springer.com/article/10.1007/s10459-010-9222-y)

2b) Can we use negative binomial regression for non-integer or continuous variables?

Based on experience I have with comparing the fit of models with likert scale data (comparing neg. bin. with normal regression for example), and a brief search (https://www.johndcook.com/blog/2009/09/22/negative-binomial-distribution/) I think that this should be possible.

Under the assumption that a likert scale variable can be analyzed as a continuous outcome, and that a negative binomial models is appropriate for continuous/non-integer data, I would assume that the answer to my question would be yes. However, I would like to hear other people's views on this.

More Tim I M Korevaar's questions See All
Similar questions and discussions