I'm performing regression on subjects' response to a question on a 1-5 likert scale (1=low, 5=high, etc)- This is the DV and its continuous and normally distributed. The IV is a count of how many times they were exposed to a stimuli (a count of the number of times).

So the IV is count data.... It's not normally distributed and a full third of the scores are zero (the average is around 3, though the max is 20). Is there any specific type of regression I could run to look at the effect of this stimuli? If the count data was the DV I'd do a zero inflated negative binomial regression or something like that, but in this instance the count data is the IV... Would simple OLS regression be suitable?

  • 3
    $\begingroup$ In your title you say the DV is normal. In your text you say it is on a 1 to 5 scale. It can't be both. Which one is it? $\endgroup$ – Peter Flom Jun 22 at 9:41

Your DV is not continuous or normally distributed -- do not be constrained by the normality assumption -- I would suggest you run an ordinal regression model (see: Can I use discrete ordinal variable as an outcome in Linear Regression? If not what other choices do i have?). If you treat ordinal data as continuous you make the erroneous assumption that your data contains more information than it actually does (consider reading this tutorial by Bürkner and Vuorre https://psyarxiv.com/x8swp/).

Your predictor variable does not have to be normally distributed.

  • $\begingroup$ Upvoted for excellence of the answer. $\endgroup$ – James Phillips Jun 22 at 11:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.